2023-05-16 06:33:57 +00:00
|
|
|
#include "git-compat-util.h"
|
2017-05-24 05:15:34 +00:00
|
|
|
#include "refs.h"
|
2023-05-16 06:34:06 +00:00
|
|
|
#include "object-store-ll.h"
|
2017-05-24 05:15:34 +00:00
|
|
|
#include "cache-tree.h"
|
2017-05-24 05:15:35 +00:00
|
|
|
#include "mergesort.h"
|
2023-04-11 03:00:40 +00:00
|
|
|
#include "convert.h"
|
2017-05-24 05:15:35 +00:00
|
|
|
#include "diff.h"
|
|
|
|
#include "diffcore.h"
|
2023-03-21 06:25:54 +00:00
|
|
|
#include "gettext.h"
|
2023-02-24 00:09:27 +00:00
|
|
|
#include "hex.h"
|
2023-05-16 06:33:59 +00:00
|
|
|
#include "path.h"
|
2023-05-16 06:33:56 +00:00
|
|
|
#include "read-cache.h"
|
2023-03-21 06:26:05 +00:00
|
|
|
#include "setup.h"
|
2017-05-24 05:15:36 +00:00
|
|
|
#include "tag.h"
|
2023-04-11 03:00:38 +00:00
|
|
|
#include "trace2.h"
|
2017-05-24 05:15:33 +00:00
|
|
|
#include "blame.h"
|
2018-05-15 21:48:42 +00:00
|
|
|
#include "alloc.h"
|
2018-05-19 05:28:19 +00:00
|
|
|
#include "commit-slab.h"
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
#include "bloom.h"
|
|
|
|
#include "commit-graph.h"
|
2018-05-19 05:28:19 +00:00
|
|
|
|
|
|
|
define_commit_slab(blame_suspects, struct blame_origin *);
|
|
|
|
static struct blame_suspects blame_suspects;
|
|
|
|
|
|
|
|
struct blame_origin *get_blame_suspects(struct commit *commit)
|
|
|
|
{
|
|
|
|
struct blame_origin **result;
|
|
|
|
|
|
|
|
result = blame_suspects_peek(&blame_suspects, commit);
|
|
|
|
|
|
|
|
return result ? *result : NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void set_blame_suspects(struct commit *commit, struct blame_origin *origin)
|
|
|
|
{
|
|
|
|
*blame_suspects_at(&blame_suspects, commit) = origin;
|
|
|
|
}
|
2017-05-24 05:15:33 +00:00
|
|
|
|
|
|
|
void blame_origin_decref(struct blame_origin *o)
|
|
|
|
{
|
|
|
|
if (o && --o->refcnt <= 0) {
|
|
|
|
struct blame_origin *p, *l = NULL;
|
|
|
|
if (o->previous)
|
|
|
|
blame_origin_decref(o->previous);
|
|
|
|
free(o->file.ptr);
|
|
|
|
/* Should be present exactly once in commit chain */
|
2018-05-19 05:28:19 +00:00
|
|
|
for (p = get_blame_suspects(o->commit); p; l = p, p = p->next) {
|
2017-05-24 05:15:33 +00:00
|
|
|
if (p == o) {
|
|
|
|
if (l)
|
|
|
|
l->next = p->next;
|
|
|
|
else
|
2018-05-19 05:28:19 +00:00
|
|
|
set_blame_suspects(o->commit, p->next);
|
2017-05-24 05:15:33 +00:00
|
|
|
free(o);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
die("internal error in blame_origin_decref");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Given a commit and a path in it, create a new origin structure.
|
|
|
|
* The callers that add blame to the scoreboard should use
|
|
|
|
* get_origin() to obtain shared, refcounted copy instead of calling
|
|
|
|
* this function directly.
|
|
|
|
*/
|
2017-05-24 05:15:34 +00:00
|
|
|
static struct blame_origin *make_origin(struct commit *commit, const char *path)
|
2017-05-24 05:15:33 +00:00
|
|
|
{
|
|
|
|
struct blame_origin *o;
|
|
|
|
FLEX_ALLOC_STR(o, path, path);
|
|
|
|
o->commit = commit;
|
|
|
|
o->refcnt = 1;
|
2018-05-19 05:28:19 +00:00
|
|
|
o->next = get_blame_suspects(commit);
|
|
|
|
set_blame_suspects(commit, o);
|
2017-05-24 05:15:33 +00:00
|
|
|
return o;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Locate an existing origin or create a new one.
|
|
|
|
* This moves the origin to front position in the commit util list.
|
|
|
|
*/
|
2017-05-24 05:15:36 +00:00
|
|
|
static struct blame_origin *get_origin(struct commit *commit, const char *path)
|
2017-05-24 05:15:33 +00:00
|
|
|
{
|
|
|
|
struct blame_origin *o, *l;
|
|
|
|
|
2018-05-19 05:28:19 +00:00
|
|
|
for (o = get_blame_suspects(commit), l = NULL; o; l = o, o = o->next) {
|
2017-05-24 05:15:33 +00:00
|
|
|
if (!strcmp(o->path, path)) {
|
|
|
|
/* bump to front */
|
|
|
|
if (l) {
|
|
|
|
l->next = o->next;
|
2018-05-19 05:28:19 +00:00
|
|
|
o->next = get_blame_suspects(commit);
|
|
|
|
set_blame_suspects(commit, o);
|
2017-05-24 05:15:33 +00:00
|
|
|
}
|
|
|
|
return blame_origin_incref(o);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return make_origin(commit, path);
|
|
|
|
}
|
2017-05-24 05:15:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
2018-09-21 15:57:21 +00:00
|
|
|
static void verify_working_tree_path(struct repository *r,
|
2018-08-13 16:14:41 +00:00
|
|
|
struct commit *work_tree, const char *path)
|
2017-05-24 05:15:34 +00:00
|
|
|
{
|
|
|
|
struct commit_list *parents;
|
|
|
|
int pos;
|
|
|
|
|
|
|
|
for (parents = work_tree->parents; parents; parents = parents->next) {
|
|
|
|
const struct object_id *commit_oid = &parents->item->object.oid;
|
|
|
|
struct object_id blob_oid;
|
2019-04-05 15:00:12 +00:00
|
|
|
unsigned short mode;
|
2017-05-24 05:15:34 +00:00
|
|
|
|
2019-06-27 09:28:49 +00:00
|
|
|
if (!get_tree_entry(r, commit_oid, path, &blob_oid, &mode) &&
|
2018-09-21 15:57:21 +00:00
|
|
|
oid_object_info(r, &blob_oid, NULL) == OBJ_BLOB)
|
2017-05-24 05:15:34 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2018-09-21 15:57:21 +00:00
|
|
|
pos = index_name_pos(r->index, path, strlen(path));
|
2017-05-24 05:15:34 +00:00
|
|
|
if (pos >= 0)
|
|
|
|
; /* path is in the index */
|
2018-09-21 15:57:21 +00:00
|
|
|
else if (-1 - pos < r->index->cache_nr &&
|
|
|
|
!strcmp(r->index->cache[-1 - pos]->name, path))
|
2017-05-24 05:15:34 +00:00
|
|
|
; /* path is in the index, unmerged */
|
|
|
|
else
|
|
|
|
die("no such path '%s' in HEAD", path);
|
|
|
|
}
|
|
|
|
|
2018-11-10 05:48:58 +00:00
|
|
|
static struct commit_list **append_parent(struct repository *r,
|
|
|
|
struct commit_list **tail,
|
|
|
|
const struct object_id *oid)
|
2017-05-24 05:15:34 +00:00
|
|
|
{
|
|
|
|
struct commit *parent;
|
|
|
|
|
2018-11-10 05:48:58 +00:00
|
|
|
parent = lookup_commit_reference(r, oid);
|
2017-05-24 05:15:34 +00:00
|
|
|
if (!parent)
|
|
|
|
die("no such commit %s", oid_to_hex(oid));
|
|
|
|
return &commit_list_insert(parent, tail)->next;
|
|
|
|
}
|
|
|
|
|
2018-11-10 05:48:58 +00:00
|
|
|
static void append_merge_parents(struct repository *r,
|
|
|
|
struct commit_list **tail)
|
2017-05-24 05:15:34 +00:00
|
|
|
{
|
|
|
|
int merge_head;
|
|
|
|
struct strbuf line = STRBUF_INIT;
|
|
|
|
|
2018-11-10 05:48:58 +00:00
|
|
|
merge_head = open(git_path_merge_head(r), O_RDONLY);
|
2017-05-24 05:15:34 +00:00
|
|
|
if (merge_head < 0) {
|
|
|
|
if (errno == ENOENT)
|
|
|
|
return;
|
2018-05-17 22:51:51 +00:00
|
|
|
die("cannot open '%s' for reading",
|
2018-11-10 05:48:58 +00:00
|
|
|
git_path_merge_head(r));
|
2017-05-24 05:15:34 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
while (!strbuf_getwholeline_fd(&line, merge_head, '\n')) {
|
|
|
|
struct object_id oid;
|
2019-08-18 20:04:08 +00:00
|
|
|
if (get_oid_hex(line.buf, &oid))
|
2018-05-17 22:51:51 +00:00
|
|
|
die("unknown line in '%s': %s",
|
2018-11-10 05:48:58 +00:00
|
|
|
git_path_merge_head(r), line.buf);
|
|
|
|
tail = append_parent(r, tail, &oid);
|
2017-05-24 05:15:34 +00:00
|
|
|
}
|
|
|
|
close(merge_head);
|
|
|
|
strbuf_release(&line);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This isn't as simple as passing sb->buf and sb->len, because we
|
|
|
|
* want to transfer ownership of the buffer to the commit (so we
|
|
|
|
* must use detach).
|
|
|
|
*/
|
2018-11-10 05:48:58 +00:00
|
|
|
static void set_commit_buffer_from_strbuf(struct repository *r,
|
|
|
|
struct commit *c,
|
|
|
|
struct strbuf *sb)
|
2017-05-24 05:15:34 +00:00
|
|
|
{
|
|
|
|
size_t len;
|
|
|
|
void *buf = strbuf_detach(sb, &len);
|
2018-11-10 05:48:58 +00:00
|
|
|
set_commit_buffer(r, c, buf, len);
|
2017-05-24 05:15:34 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Prepare a dummy commit that represents the work tree (or staged) item.
|
|
|
|
* Note that annotating work tree item never works in the reverse.
|
|
|
|
*/
|
2018-09-21 15:57:21 +00:00
|
|
|
static struct commit *fake_working_tree_commit(struct repository *r,
|
2018-08-13 16:14:41 +00:00
|
|
|
struct diff_options *opt,
|
2017-05-24 05:15:36 +00:00
|
|
|
const char *path,
|
2023-03-24 17:08:00 +00:00
|
|
|
const char *contents_from,
|
|
|
|
struct object_id *oid)
|
2017-05-24 05:15:34 +00:00
|
|
|
{
|
|
|
|
struct commit *commit;
|
|
|
|
struct blame_origin *origin;
|
|
|
|
struct commit_list **parent_tail, *parent;
|
|
|
|
struct strbuf buf = STRBUF_INIT;
|
|
|
|
const char *ident;
|
|
|
|
time_t now;
|
2018-07-02 19:49:31 +00:00
|
|
|
int len;
|
2017-05-24 05:15:34 +00:00
|
|
|
struct cache_entry *ce;
|
|
|
|
unsigned mode;
|
|
|
|
struct strbuf msg = STRBUF_INIT;
|
|
|
|
|
2019-01-12 02:13:26 +00:00
|
|
|
repo_read_index(r);
|
2017-05-24 05:15:34 +00:00
|
|
|
time(&now);
|
2018-11-10 05:48:58 +00:00
|
|
|
commit = alloc_commit_node(r);
|
2017-05-24 05:15:34 +00:00
|
|
|
commit->object.parsed = 1;
|
|
|
|
commit->date = now;
|
|
|
|
parent_tail = &commit->parents;
|
|
|
|
|
2023-03-24 17:08:00 +00:00
|
|
|
parent_tail = append_parent(r, parent_tail, oid);
|
2018-11-10 05:48:58 +00:00
|
|
|
append_merge_parents(r, parent_tail);
|
2018-09-21 15:57:21 +00:00
|
|
|
verify_working_tree_path(r, commit, path);
|
2017-05-24 05:15:34 +00:00
|
|
|
|
|
|
|
origin = make_origin(commit, path);
|
|
|
|
|
2023-04-24 19:35:08 +00:00
|
|
|
if (contents_from)
|
|
|
|
ident = fmt_ident("External file (--contents)", "external.file",
|
|
|
|
WANT_BLANK_IDENT, NULL, 0);
|
|
|
|
else
|
|
|
|
ident = fmt_ident("Not Committed Yet", "not.committed.yet",
|
|
|
|
WANT_BLANK_IDENT, NULL, 0);
|
2017-05-24 05:15:34 +00:00
|
|
|
strbuf_addstr(&msg, "tree 0000000000000000000000000000000000000000\n");
|
|
|
|
for (parent = commit->parents; parent; parent = parent->next)
|
|
|
|
strbuf_addf(&msg, "parent %s\n",
|
|
|
|
oid_to_hex(&parent->item->object.oid));
|
|
|
|
strbuf_addf(&msg,
|
|
|
|
"author %s\n"
|
|
|
|
"committer %s\n\n"
|
|
|
|
"Version of %s from %s\n",
|
|
|
|
ident, ident, path,
|
|
|
|
(!contents_from ? path :
|
|
|
|
(!strcmp(contents_from, "-") ? "standard input" : contents_from)));
|
2018-11-10 05:48:58 +00:00
|
|
|
set_commit_buffer_from_strbuf(r, commit, &msg);
|
2017-05-24 05:15:34 +00:00
|
|
|
|
|
|
|
if (!contents_from || strcmp("-", contents_from)) {
|
|
|
|
struct stat st;
|
|
|
|
const char *read_from;
|
|
|
|
char *buf_ptr;
|
|
|
|
unsigned long buf_len;
|
|
|
|
|
|
|
|
if (contents_from) {
|
|
|
|
if (stat(contents_from, &st) < 0)
|
|
|
|
die_errno("Cannot stat '%s'", contents_from);
|
|
|
|
read_from = contents_from;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
if (lstat(path, &st) < 0)
|
|
|
|
die_errno("Cannot lstat '%s'", path);
|
|
|
|
read_from = path;
|
|
|
|
}
|
|
|
|
mode = canon_mode(st.st_mode);
|
|
|
|
|
|
|
|
switch (st.st_mode & S_IFMT) {
|
|
|
|
case S_IFREG:
|
2017-10-31 18:19:11 +00:00
|
|
|
if (opt->flags.allow_textconv &&
|
2021-04-26 01:02:56 +00:00
|
|
|
textconv_object(r, read_from, mode, null_oid(), 0, &buf_ptr, &buf_len))
|
2017-05-24 05:15:34 +00:00
|
|
|
strbuf_attach(&buf, buf_ptr, buf_len, buf_len + 1);
|
|
|
|
else if (strbuf_read_file(&buf, read_from, st.st_size) != st.st_size)
|
|
|
|
die_errno("cannot open or read '%s'", read_from);
|
|
|
|
break;
|
|
|
|
case S_IFLNK:
|
|
|
|
if (strbuf_readlink(&buf, read_from, st.st_size) < 0)
|
|
|
|
die_errno("cannot readlink '%s'", read_from);
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
die("unsupported file type %s", read_from);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
/* Reading from stdin */
|
|
|
|
mode = 0;
|
|
|
|
if (strbuf_read(&buf, 0, 0) < 0)
|
|
|
|
die_errno("failed to read from stdin");
|
|
|
|
}
|
2018-09-21 15:57:21 +00:00
|
|
|
convert_to_git(r->index, path, buf.buf, buf.len, &buf, 0);
|
2017-05-24 05:15:34 +00:00
|
|
|
origin->file.ptr = buf.buf;
|
|
|
|
origin->file.size = buf.len;
|
2018-01-28 00:13:11 +00:00
|
|
|
pretend_object_file(buf.buf, buf.len, OBJ_BLOB, &origin->blob_oid);
|
2017-05-24 05:15:34 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Read the current index, replace the path entry with
|
|
|
|
* origin->blob_sha1 without mucking with its mode or type
|
|
|
|
* bits; we are not going to write this index out -- we just
|
|
|
|
* want to run "diff-index --cached".
|
|
|
|
*/
|
2018-09-21 15:57:21 +00:00
|
|
|
discard_index(r->index);
|
2019-01-12 02:13:26 +00:00
|
|
|
repo_read_index(r);
|
2017-05-24 05:15:34 +00:00
|
|
|
|
|
|
|
len = strlen(path);
|
|
|
|
if (!mode) {
|
2018-09-21 15:57:21 +00:00
|
|
|
int pos = index_name_pos(r->index, path, len);
|
2017-05-24 05:15:34 +00:00
|
|
|
if (0 <= pos)
|
2018-09-21 15:57:21 +00:00
|
|
|
mode = r->index->cache[pos]->ce_mode;
|
2017-05-24 05:15:34 +00:00
|
|
|
else
|
|
|
|
/* Let's not bother reading from HEAD tree */
|
|
|
|
mode = S_IFREG | 0644;
|
|
|
|
}
|
2018-09-21 15:57:21 +00:00
|
|
|
ce = make_empty_cache_entry(r->index, len);
|
2017-05-24 05:15:34 +00:00
|
|
|
oidcpy(&ce->oid, &origin->blob_oid);
|
|
|
|
memcpy(ce->name, path, len);
|
|
|
|
ce->ce_flags = create_ce_flags(0);
|
|
|
|
ce->ce_namelen = len;
|
|
|
|
ce->ce_mode = create_ce_mode(mode);
|
2018-09-21 15:57:21 +00:00
|
|
|
add_index_entry(r->index, ce,
|
2018-08-13 16:14:41 +00:00
|
|
|
ADD_CACHE_OK_TO_ADD | ADD_CACHE_OK_TO_REPLACE);
|
2017-05-24 05:15:34 +00:00
|
|
|
|
2018-09-21 15:57:21 +00:00
|
|
|
cache_tree_invalidate_path(r->index, path);
|
2017-05-24 05:15:34 +00:00
|
|
|
|
|
|
|
return commit;
|
|
|
|
}
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
static int diff_hunks(mmfile_t *file_a, mmfile_t *file_b,
|
|
|
|
xdl_emit_hunk_consume_func_t hunk_func, void *cb_data, int xdl_opts)
|
|
|
|
{
|
|
|
|
xpparam_t xpp = {0};
|
|
|
|
xdemitconf_t xecfg = {0};
|
|
|
|
xdemitcb_t ecb = {NULL};
|
|
|
|
|
|
|
|
xpp.flags = xdl_opts;
|
|
|
|
xecfg.hunk_func = hunk_func;
|
|
|
|
ecb.priv = cb_data;
|
|
|
|
return xdi_diff(file_a, file_b, &xpp, &xecfg, &ecb);
|
|
|
|
}
|
|
|
|
|
2019-05-15 21:45:01 +00:00
|
|
|
static const char *get_next_line(const char *start, const char *end)
|
|
|
|
{
|
|
|
|
const char *nl = memchr(start, '\n', end - start);
|
|
|
|
|
|
|
|
return nl ? nl + 1 : end;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int find_line_starts(int **line_starts, const char *buf,
|
|
|
|
unsigned long len)
|
|
|
|
{
|
|
|
|
const char *end = buf + len;
|
|
|
|
const char *p;
|
|
|
|
int *lineno;
|
|
|
|
int num = 0;
|
|
|
|
|
|
|
|
for (p = buf; p < end; p = get_next_line(p, end))
|
|
|
|
num++;
|
|
|
|
|
|
|
|
ALLOC_ARRAY(*line_starts, num + 1);
|
|
|
|
lineno = *line_starts;
|
|
|
|
|
|
|
|
for (p = buf; p < end; p = get_next_line(p, end))
|
|
|
|
*lineno++ = p - buf;
|
|
|
|
|
|
|
|
*lineno = len;
|
|
|
|
|
|
|
|
return num;
|
|
|
|
}
|
|
|
|
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
struct fingerprint_entry;
|
|
|
|
|
|
|
|
/* A fingerprint is intended to loosely represent a string, such that two
|
|
|
|
* fingerprints can be quickly compared to give an indication of the similarity
|
|
|
|
* of the strings that they represent.
|
|
|
|
*
|
|
|
|
* A fingerprint is represented as a multiset of the lower-cased byte pairs in
|
|
|
|
* the string that it represents. Whitespace is added at each end of the
|
|
|
|
* string. Whitespace pairs are ignored. Whitespace is converted to '\0'.
|
|
|
|
* For example, the string "Darth Radar" will be converted to the following
|
|
|
|
* fingerprint:
|
|
|
|
* {"\0d", "da", "da", "ar", "ar", "rt", "th", "h\0", "\0r", "ra", "ad", "r\0"}
|
|
|
|
*
|
|
|
|
* The similarity between two fingerprints is the size of the intersection of
|
|
|
|
* their multisets, including repeated elements. See fingerprint_similarity for
|
|
|
|
* examples.
|
|
|
|
*
|
|
|
|
* For ease of implementation, the fingerprint is implemented as a map
|
|
|
|
* of byte pairs to the count of that byte pair in the string, instead of
|
|
|
|
* allowing repeated elements in a set.
|
|
|
|
*/
|
|
|
|
struct fingerprint {
|
|
|
|
struct hashmap map;
|
|
|
|
/* As we know the maximum number of entries in advance, it's
|
|
|
|
* convenient to store the entries in a single array instead of having
|
|
|
|
* the hashmap manage the memory.
|
|
|
|
*/
|
|
|
|
struct fingerprint_entry *entries;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* A byte pair in a fingerprint. Stores the number of times the byte pair
|
|
|
|
* occurs in the string that the fingerprint represents.
|
|
|
|
*/
|
|
|
|
struct fingerprint_entry {
|
|
|
|
/* The hashmap entry - the hash represents the byte pair in its
|
|
|
|
* entirety so we don't need to store the byte pair separately.
|
|
|
|
*/
|
|
|
|
struct hashmap_entry entry;
|
|
|
|
/* The number of times the byte pair occurs in the string that the
|
|
|
|
* fingerprint represents.
|
|
|
|
*/
|
|
|
|
int count;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* See `struct fingerprint` for an explanation of what a fingerprint is.
|
|
|
|
* \param result the fingerprint of the string is stored here. This must be
|
|
|
|
* freed later using free_fingerprint.
|
|
|
|
* \param line_begin the start of the string
|
|
|
|
* \param line_end the end of the string
|
|
|
|
*/
|
|
|
|
static void get_fingerprint(struct fingerprint *result,
|
|
|
|
const char *line_begin,
|
|
|
|
const char *line_end)
|
|
|
|
{
|
|
|
|
unsigned int hash, c0 = 0, c1;
|
|
|
|
const char *p;
|
|
|
|
int max_map_entry_count = 1 + line_end - line_begin;
|
|
|
|
struct fingerprint_entry *entry = xcalloc(max_map_entry_count,
|
|
|
|
sizeof(struct fingerprint_entry));
|
|
|
|
struct fingerprint_entry *found_entry;
|
|
|
|
|
|
|
|
hashmap_init(&result->map, NULL, NULL, max_map_entry_count);
|
|
|
|
result->entries = entry;
|
|
|
|
for (p = line_begin; p <= line_end; ++p, c0 = c1) {
|
|
|
|
/* Always terminate the string with whitespace.
|
|
|
|
* Normalise whitespace to 0, and normalise letters to
|
|
|
|
* lower case. This won't work for multibyte characters but at
|
|
|
|
* worst will match some unrelated characters.
|
|
|
|
*/
|
|
|
|
if ((p == line_end) || isspace(*p))
|
|
|
|
c1 = 0;
|
|
|
|
else
|
|
|
|
c1 = tolower(*p);
|
|
|
|
hash = c0 | (c1 << 8);
|
|
|
|
/* Ignore whitespace pairs */
|
|
|
|
if (hash == 0)
|
|
|
|
continue;
|
2019-10-06 23:30:27 +00:00
|
|
|
hashmap_entry_init(&entry->entry, hash);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
|
2019-10-06 23:30:42 +00:00
|
|
|
found_entry = hashmap_get_entry(&result->map, entry,
|
|
|
|
/* member name */ entry, NULL);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
if (found_entry) {
|
|
|
|
found_entry->count += 1;
|
|
|
|
} else {
|
|
|
|
entry->count = 1;
|
2019-10-06 23:30:29 +00:00
|
|
|
hashmap_add(&result->map, &entry->entry);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
++entry;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void free_fingerprint(struct fingerprint *f)
|
|
|
|
{
|
2020-11-02 18:55:05 +00:00
|
|
|
hashmap_clear(&f->map);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
free(f->entries);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Calculates the similarity between two fingerprints as the size of the
|
|
|
|
* intersection of their multisets, including repeated elements. See
|
|
|
|
* `struct fingerprint` for an explanation of the fingerprint representation.
|
|
|
|
* The similarity between "cat mat" and "father rather" is 2 because "at" is
|
|
|
|
* present twice in both strings while the similarity between "tim" and "mit"
|
|
|
|
* is 0.
|
|
|
|
*/
|
|
|
|
static int fingerprint_similarity(struct fingerprint *a, struct fingerprint *b)
|
|
|
|
{
|
|
|
|
int intersection = 0;
|
|
|
|
struct hashmap_iter iter;
|
|
|
|
const struct fingerprint_entry *entry_a, *entry_b;
|
|
|
|
|
2019-10-06 23:30:38 +00:00
|
|
|
hashmap_for_each_entry(&b->map, &iter, entry_b,
|
|
|
|
entry /* member name */) {
|
2019-10-06 23:30:42 +00:00
|
|
|
entry_a = hashmap_get_entry(&a->map, entry_b, entry, NULL);
|
2019-10-06 23:30:36 +00:00
|
|
|
if (entry_a) {
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
intersection += entry_a->count < entry_b->count ?
|
|
|
|
entry_a->count : entry_b->count;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return intersection;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Subtracts byte-pair elements in B from A, modifying A in place.
|
|
|
|
*/
|
|
|
|
static void fingerprint_subtract(struct fingerprint *a, struct fingerprint *b)
|
|
|
|
{
|
|
|
|
struct hashmap_iter iter;
|
|
|
|
struct fingerprint_entry *entry_a;
|
|
|
|
const struct fingerprint_entry *entry_b;
|
|
|
|
|
|
|
|
hashmap_iter_init(&b->map, &iter);
|
|
|
|
|
2019-10-06 23:30:38 +00:00
|
|
|
hashmap_for_each_entry(&b->map, &iter, entry_b,
|
|
|
|
entry /* member name */) {
|
2019-10-06 23:30:42 +00:00
|
|
|
entry_a = hashmap_get_entry(&a->map, entry_b, entry, NULL);
|
2019-10-06 23:30:36 +00:00
|
|
|
if (entry_a) {
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
if (entry_a->count <= entry_b->count)
|
2019-10-06 23:30:31 +00:00
|
|
|
hashmap_remove(&a->map, &entry_b->entry, NULL);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
else
|
|
|
|
entry_a->count -= entry_b->count;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Calculate fingerprints for a series of lines.
|
|
|
|
* Puts the fingerprints in the fingerprints array, which must have been
|
|
|
|
* preallocated to allow storing line_count elements.
|
|
|
|
*/
|
|
|
|
static void get_line_fingerprints(struct fingerprint *fingerprints,
|
|
|
|
const char *content, const int *line_starts,
|
|
|
|
long first_line, long line_count)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
const char *linestart, *lineend;
|
|
|
|
|
|
|
|
line_starts += first_line;
|
|
|
|
for (i = 0; i < line_count; ++i) {
|
|
|
|
linestart = content + line_starts[i];
|
|
|
|
lineend = content + line_starts[i + 1];
|
|
|
|
get_fingerprint(fingerprints + i, linestart, lineend);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void free_line_fingerprints(struct fingerprint *fingerprints,
|
|
|
|
int nr_fingerprints)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < nr_fingerprints; i++)
|
|
|
|
free_fingerprint(&fingerprints[i]);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* This contains the data necessary to linearly map a line number in one half
|
|
|
|
* of a diff chunk to the line in the other half of the diff chunk that is
|
|
|
|
* closest in terms of its position as a fraction of the length of the chunk.
|
|
|
|
*/
|
|
|
|
struct line_number_mapping {
|
|
|
|
int destination_start, destination_length,
|
|
|
|
source_start, source_length;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Given a line number in one range, offset and scale it to map it onto the
|
|
|
|
* other range.
|
|
|
|
* Essentially this mapping is a simple linear equation but the calculation is
|
|
|
|
* more complicated to allow performing it with integer operations.
|
|
|
|
* Another complication is that if a line could map onto many lines in the
|
|
|
|
* destination range then we want to choose the line at the center of those
|
|
|
|
* possibilities.
|
|
|
|
* Example: if the chunk is 2 lines long in A and 10 lines long in B then the
|
|
|
|
* first 5 lines in B will map onto the first line in the A chunk, while the
|
|
|
|
* last 5 lines will all map onto the second line in the A chunk.
|
|
|
|
* Example: if the chunk is 10 lines long in A and 2 lines long in B then line
|
|
|
|
* 0 in B will map onto line 2 in A, and line 1 in B will map onto line 7 in A.
|
|
|
|
*/
|
|
|
|
static int map_line_number(int line_number,
|
|
|
|
const struct line_number_mapping *mapping)
|
|
|
|
{
|
|
|
|
return ((line_number - mapping->source_start) * 2 + 1) *
|
|
|
|
mapping->destination_length /
|
|
|
|
(mapping->source_length * 2) +
|
|
|
|
mapping->destination_start;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Get a pointer to the element storing the similarity between a line in A
|
|
|
|
* and a line in B.
|
|
|
|
*
|
|
|
|
* The similarities are stored in a 2-dimensional array. Each "row" in the
|
|
|
|
* array contains the similarities for a line in B. The similarities stored in
|
|
|
|
* a row are the similarities between the line in B and the nearby lines in A.
|
|
|
|
* To keep the length of each row the same, it is padded out with values of -1
|
|
|
|
* where the search range extends beyond the lines in A.
|
|
|
|
* For example, if max_search_distance_a is 2 and the two sides of a diff chunk
|
|
|
|
* look like this:
|
|
|
|
* a | m
|
|
|
|
* b | n
|
|
|
|
* c | o
|
|
|
|
* d | p
|
|
|
|
* e | q
|
|
|
|
* Then the similarity array will contain:
|
|
|
|
* [-1, -1, am, bm, cm,
|
|
|
|
* -1, an, bn, cn, dn,
|
|
|
|
* ao, bo, co, do, eo,
|
|
|
|
* bp, cp, dp, ep, -1,
|
|
|
|
* cq, dq, eq, -1, -1]
|
|
|
|
* Where similarities are denoted either by -1 for invalid, or the
|
|
|
|
* concatenation of the two lines in the diff being compared.
|
|
|
|
*
|
|
|
|
* \param similarities array of similarities between lines in A and B
|
|
|
|
* \param line_a the index of the line in A, in the same frame of reference as
|
|
|
|
* closest_line_a.
|
|
|
|
* \param local_line_b the index of the line in B, relative to the first line
|
|
|
|
* in B that similarities represents.
|
|
|
|
* \param closest_line_a the index of the line in A that is deemed to be
|
|
|
|
* closest to local_line_b. This must be in the same
|
|
|
|
* frame of reference as line_a. This value defines
|
|
|
|
* where similarities is centered for the line in B.
|
|
|
|
* \param max_search_distance_a maximum distance in lines from the closest line
|
|
|
|
* in A for other lines in A for which
|
|
|
|
* similarities may be calculated.
|
|
|
|
*/
|
|
|
|
static int *get_similarity(int *similarities,
|
|
|
|
int line_a, int local_line_b,
|
|
|
|
int closest_line_a, int max_search_distance_a)
|
|
|
|
{
|
|
|
|
assert(abs(line_a - closest_line_a) <=
|
|
|
|
max_search_distance_a);
|
|
|
|
return similarities + line_a - closest_line_a +
|
|
|
|
max_search_distance_a +
|
|
|
|
local_line_b * (max_search_distance_a * 2 + 1);
|
|
|
|
}
|
|
|
|
|
|
|
|
#define CERTAIN_NOTHING_MATCHES -2
|
|
|
|
#define CERTAINTY_NOT_CALCULATED -1
|
|
|
|
|
|
|
|
/* Given a line in B, first calculate its similarities with nearby lines in A
|
|
|
|
* if not already calculated, then identify the most similar and second most
|
|
|
|
* similar lines. The "certainty" is calculated based on those two
|
|
|
|
* similarities.
|
|
|
|
*
|
|
|
|
* \param start_a the index of the first line of the chunk in A
|
|
|
|
* \param length_a the length in lines of the chunk in A
|
|
|
|
* \param local_line_b the index of the line in B, relative to the first line
|
|
|
|
* in the chunk.
|
|
|
|
* \param fingerprints_a array of fingerprints for the chunk in A
|
|
|
|
* \param fingerprints_b array of fingerprints for the chunk in B
|
|
|
|
* \param similarities 2-dimensional array of similarities between lines in A
|
|
|
|
* and B. See get_similarity() for more details.
|
|
|
|
* \param certainties array of values indicating how strongly a line in B is
|
|
|
|
* matched with some line in A.
|
|
|
|
* \param second_best_result array of absolute indices in A for the second
|
|
|
|
* closest match of a line in B.
|
|
|
|
* \param result array of absolute indices in A for the closest match of a line
|
|
|
|
* in B.
|
|
|
|
* \param max_search_distance_a maximum distance in lines from the closest line
|
|
|
|
* in A for other lines in A for which
|
|
|
|
* similarities may be calculated.
|
|
|
|
* \param map_line_number_in_b_to_a parameter to map_line_number().
|
|
|
|
*/
|
|
|
|
static void find_best_line_matches(
|
|
|
|
int start_a,
|
|
|
|
int length_a,
|
|
|
|
int start_b,
|
|
|
|
int local_line_b,
|
|
|
|
struct fingerprint *fingerprints_a,
|
|
|
|
struct fingerprint *fingerprints_b,
|
|
|
|
int *similarities,
|
|
|
|
int *certainties,
|
|
|
|
int *second_best_result,
|
|
|
|
int *result,
|
|
|
|
const int max_search_distance_a,
|
|
|
|
const struct line_number_mapping *map_line_number_in_b_to_a)
|
|
|
|
{
|
|
|
|
|
|
|
|
int i, search_start, search_end, closest_local_line_a, *similarity,
|
|
|
|
best_similarity = 0, second_best_similarity = 0,
|
|
|
|
best_similarity_index = 0, second_best_similarity_index = 0;
|
|
|
|
|
|
|
|
/* certainty has already been calculated so no need to redo the work */
|
|
|
|
if (certainties[local_line_b] != CERTAINTY_NOT_CALCULATED)
|
|
|
|
return;
|
|
|
|
|
|
|
|
closest_local_line_a = map_line_number(
|
|
|
|
local_line_b + start_b, map_line_number_in_b_to_a) - start_a;
|
|
|
|
|
|
|
|
search_start = closest_local_line_a - max_search_distance_a;
|
|
|
|
if (search_start < 0)
|
|
|
|
search_start = 0;
|
|
|
|
|
|
|
|
search_end = closest_local_line_a + max_search_distance_a + 1;
|
|
|
|
if (search_end > length_a)
|
|
|
|
search_end = length_a;
|
|
|
|
|
|
|
|
for (i = search_start; i < search_end; ++i) {
|
|
|
|
similarity = get_similarity(similarities,
|
|
|
|
i, local_line_b,
|
|
|
|
closest_local_line_a,
|
|
|
|
max_search_distance_a);
|
|
|
|
if (*similarity == -1) {
|
|
|
|
/* This value will never exceed 10 but assert just in
|
|
|
|
* case
|
|
|
|
*/
|
|
|
|
assert(abs(i - closest_local_line_a) < 1000);
|
|
|
|
/* scale the similarity by (1000 - distance from
|
|
|
|
* closest line) to act as a tie break between lines
|
|
|
|
* that otherwise are equally similar.
|
|
|
|
*/
|
|
|
|
*similarity = fingerprint_similarity(
|
|
|
|
fingerprints_b + local_line_b,
|
|
|
|
fingerprints_a + i) *
|
|
|
|
(1000 - abs(i - closest_local_line_a));
|
|
|
|
}
|
|
|
|
if (*similarity > best_similarity) {
|
|
|
|
second_best_similarity = best_similarity;
|
|
|
|
second_best_similarity_index = best_similarity_index;
|
|
|
|
best_similarity = *similarity;
|
|
|
|
best_similarity_index = i;
|
|
|
|
} else if (*similarity > second_best_similarity) {
|
|
|
|
second_best_similarity = *similarity;
|
|
|
|
second_best_similarity_index = i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (best_similarity == 0) {
|
|
|
|
/* this line definitely doesn't match with anything. Mark it
|
|
|
|
* with this special value so it doesn't get invalidated and
|
|
|
|
* won't be recalculated.
|
|
|
|
*/
|
|
|
|
certainties[local_line_b] = CERTAIN_NOTHING_MATCHES;
|
|
|
|
result[local_line_b] = -1;
|
|
|
|
} else {
|
|
|
|
/* Calculate the certainty with which this line matches.
|
|
|
|
* If the line matches well with two lines then that reduces
|
|
|
|
* the certainty. However we still want to prioritise matching
|
|
|
|
* a line that matches very well with two lines over matching a
|
|
|
|
* line that matches poorly with one line, hence doubling
|
|
|
|
* best_similarity.
|
|
|
|
* This means that if we have
|
|
|
|
* line X that matches only one line with a score of 3,
|
|
|
|
* line Y that matches two lines equally with a score of 5,
|
|
|
|
* and line Z that matches only one line with a score or 2,
|
|
|
|
* then the lines in order of certainty are X, Y, Z.
|
|
|
|
*/
|
|
|
|
certainties[local_line_b] = best_similarity * 2 -
|
|
|
|
second_best_similarity;
|
|
|
|
|
|
|
|
/* We keep both the best and second best results to allow us to
|
|
|
|
* check at a later stage of the matching process whether the
|
|
|
|
* result needs to be invalidated.
|
|
|
|
*/
|
|
|
|
result[local_line_b] = start_a + best_similarity_index;
|
|
|
|
second_best_result[local_line_b] =
|
|
|
|
start_a + second_best_similarity_index;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This finds the line that we can match with the most confidence, and
|
|
|
|
* uses it as a partition. It then calls itself on the lines on either side of
|
|
|
|
* that partition. In this way we avoid lines appearing out of order, and
|
|
|
|
* retain a sensible line ordering.
|
|
|
|
* \param start_a index of the first line in A with which lines in B may be
|
|
|
|
* compared.
|
|
|
|
* \param start_b index of the first line in B for which matching should be
|
|
|
|
* done.
|
|
|
|
* \param length_a number of lines in A with which lines in B may be compared.
|
|
|
|
* \param length_b number of lines in B for which matching should be done.
|
|
|
|
* \param fingerprints_a mutable array of fingerprints in A. The first element
|
|
|
|
* corresponds to the line at start_a.
|
|
|
|
* \param fingerprints_b array of fingerprints in B. The first element
|
|
|
|
* corresponds to the line at start_b.
|
|
|
|
* \param similarities 2-dimensional array of similarities between lines in A
|
|
|
|
* and B. See get_similarity() for more details.
|
|
|
|
* \param certainties array of values indicating how strongly a line in B is
|
|
|
|
* matched with some line in A.
|
|
|
|
* \param second_best_result array of absolute indices in A for the second
|
|
|
|
* closest match of a line in B.
|
|
|
|
* \param result array of absolute indices in A for the closest match of a line
|
|
|
|
* in B.
|
|
|
|
* \param max_search_distance_a maximum distance in lines from the closest line
|
|
|
|
* in A for other lines in A for which
|
|
|
|
* similarities may be calculated.
|
|
|
|
* \param max_search_distance_b an upper bound on the greatest possible
|
|
|
|
* distance between lines in B such that they will
|
|
|
|
* both be compared with the same line in A
|
|
|
|
* according to max_search_distance_a.
|
|
|
|
* \param map_line_number_in_b_to_a parameter to map_line_number().
|
|
|
|
*/
|
|
|
|
static void fuzzy_find_matching_lines_recurse(
|
|
|
|
int start_a, int start_b,
|
|
|
|
int length_a, int length_b,
|
|
|
|
struct fingerprint *fingerprints_a,
|
|
|
|
struct fingerprint *fingerprints_b,
|
|
|
|
int *similarities,
|
|
|
|
int *certainties,
|
|
|
|
int *second_best_result,
|
|
|
|
int *result,
|
|
|
|
int max_search_distance_a,
|
|
|
|
int max_search_distance_b,
|
|
|
|
const struct line_number_mapping *map_line_number_in_b_to_a)
|
|
|
|
{
|
|
|
|
int i, invalidate_min, invalidate_max, offset_b,
|
|
|
|
second_half_start_a, second_half_start_b,
|
|
|
|
second_half_length_a, second_half_length_b,
|
|
|
|
most_certain_line_a, most_certain_local_line_b = -1,
|
|
|
|
most_certain_line_certainty = -1,
|
|
|
|
closest_local_line_a;
|
|
|
|
|
|
|
|
for (i = 0; i < length_b; ++i) {
|
|
|
|
find_best_line_matches(start_a,
|
|
|
|
length_a,
|
|
|
|
start_b,
|
|
|
|
i,
|
|
|
|
fingerprints_a,
|
|
|
|
fingerprints_b,
|
|
|
|
similarities,
|
|
|
|
certainties,
|
|
|
|
second_best_result,
|
|
|
|
result,
|
|
|
|
max_search_distance_a,
|
|
|
|
map_line_number_in_b_to_a);
|
|
|
|
|
|
|
|
if (certainties[i] > most_certain_line_certainty) {
|
|
|
|
most_certain_line_certainty = certainties[i];
|
|
|
|
most_certain_local_line_b = i;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* No matches. */
|
|
|
|
if (most_certain_local_line_b == -1)
|
|
|
|
return;
|
|
|
|
|
|
|
|
most_certain_line_a = result[most_certain_local_line_b];
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Subtract the most certain line's fingerprint in B from the matched
|
|
|
|
* fingerprint in A. This means that other lines in B can't also match
|
|
|
|
* the same parts of the line in A.
|
|
|
|
*/
|
|
|
|
fingerprint_subtract(fingerprints_a + most_certain_line_a - start_a,
|
|
|
|
fingerprints_b + most_certain_local_line_b);
|
|
|
|
|
|
|
|
/* Invalidate results that may be affected by the choice of most
|
|
|
|
* certain line.
|
|
|
|
*/
|
|
|
|
invalidate_min = most_certain_local_line_b - max_search_distance_b;
|
|
|
|
invalidate_max = most_certain_local_line_b + max_search_distance_b + 1;
|
|
|
|
if (invalidate_min < 0)
|
|
|
|
invalidate_min = 0;
|
|
|
|
if (invalidate_max > length_b)
|
|
|
|
invalidate_max = length_b;
|
|
|
|
|
|
|
|
/* As the fingerprint in A has changed, discard previously calculated
|
|
|
|
* similarity values with that fingerprint.
|
|
|
|
*/
|
|
|
|
for (i = invalidate_min; i < invalidate_max; ++i) {
|
|
|
|
closest_local_line_a = map_line_number(
|
|
|
|
i + start_b, map_line_number_in_b_to_a) - start_a;
|
|
|
|
|
|
|
|
/* Check that the lines in A and B are close enough that there
|
|
|
|
* is a similarity value for them.
|
|
|
|
*/
|
|
|
|
if (abs(most_certain_line_a - start_a - closest_local_line_a) >
|
|
|
|
max_search_distance_a) {
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
*get_similarity(similarities, most_certain_line_a - start_a,
|
|
|
|
i, closest_local_line_a,
|
|
|
|
max_search_distance_a) = -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* More invalidating of results that may be affected by the choice of
|
|
|
|
* most certain line.
|
|
|
|
* Discard the matches for lines in B that are currently matched with a
|
|
|
|
* line in A such that their ordering contradicts the ordering imposed
|
|
|
|
* by the choice of most certain line.
|
|
|
|
*/
|
|
|
|
for (i = most_certain_local_line_b - 1; i >= invalidate_min; --i) {
|
|
|
|
/* In this loop we discard results for lines in B that are
|
|
|
|
* before most-certain-line-B but are matched with a line in A
|
|
|
|
* that is after most-certain-line-A.
|
|
|
|
*/
|
|
|
|
if (certainties[i] >= 0 &&
|
|
|
|
(result[i] >= most_certain_line_a ||
|
|
|
|
second_best_result[i] >= most_certain_line_a)) {
|
|
|
|
certainties[i] = CERTAINTY_NOT_CALCULATED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
for (i = most_certain_local_line_b + 1; i < invalidate_max; ++i) {
|
|
|
|
/* In this loop we discard results for lines in B that are
|
|
|
|
* after most-certain-line-B but are matched with a line in A
|
|
|
|
* that is before most-certain-line-A.
|
|
|
|
*/
|
|
|
|
if (certainties[i] >= 0 &&
|
|
|
|
(result[i] <= most_certain_line_a ||
|
|
|
|
second_best_result[i] <= most_certain_line_a)) {
|
|
|
|
certainties[i] = CERTAINTY_NOT_CALCULATED;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Repeat the matching process for lines before the most certain line.
|
|
|
|
*/
|
|
|
|
if (most_certain_local_line_b > 0) {
|
|
|
|
fuzzy_find_matching_lines_recurse(
|
|
|
|
start_a, start_b,
|
|
|
|
most_certain_line_a + 1 - start_a,
|
|
|
|
most_certain_local_line_b,
|
|
|
|
fingerprints_a, fingerprints_b, similarities,
|
|
|
|
certainties, second_best_result, result,
|
|
|
|
max_search_distance_a,
|
|
|
|
max_search_distance_b,
|
|
|
|
map_line_number_in_b_to_a);
|
|
|
|
}
|
|
|
|
/* Repeat the matching process for lines after the most certain line.
|
|
|
|
*/
|
|
|
|
if (most_certain_local_line_b + 1 < length_b) {
|
|
|
|
second_half_start_a = most_certain_line_a;
|
|
|
|
offset_b = most_certain_local_line_b + 1;
|
|
|
|
second_half_start_b = start_b + offset_b;
|
|
|
|
second_half_length_a =
|
|
|
|
length_a + start_a - second_half_start_a;
|
|
|
|
second_half_length_b =
|
|
|
|
length_b + start_b - second_half_start_b;
|
|
|
|
fuzzy_find_matching_lines_recurse(
|
|
|
|
second_half_start_a, second_half_start_b,
|
|
|
|
second_half_length_a, second_half_length_b,
|
|
|
|
fingerprints_a + second_half_start_a - start_a,
|
|
|
|
fingerprints_b + offset_b,
|
|
|
|
similarities +
|
|
|
|
offset_b * (max_search_distance_a * 2 + 1),
|
|
|
|
certainties + offset_b,
|
|
|
|
second_best_result + offset_b, result + offset_b,
|
|
|
|
max_search_distance_a,
|
|
|
|
max_search_distance_b,
|
|
|
|
map_line_number_in_b_to_a);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find the lines in the parent line range that most closely match the lines in
|
|
|
|
* the target line range. This is accomplished by matching fingerprints in each
|
|
|
|
* blame_origin, and choosing the best matches that preserve the line ordering.
|
|
|
|
* See struct fingerprint for details of fingerprint matching, and
|
|
|
|
* fuzzy_find_matching_lines_recurse for details of preserving line ordering.
|
|
|
|
*
|
|
|
|
* The performance is believed to be O(n log n) in the typical case and O(n^2)
|
|
|
|
* in a pathological case, where n is the number of lines in the target range.
|
|
|
|
*/
|
|
|
|
static int *fuzzy_find_matching_lines(struct blame_origin *parent,
|
|
|
|
struct blame_origin *target,
|
|
|
|
int tlno, int parent_slno, int same,
|
|
|
|
int parent_len)
|
|
|
|
{
|
|
|
|
/* We use the terminology "A" for the left hand side of the diff AKA
|
|
|
|
* parent, and "B" for the right hand side of the diff AKA target. */
|
|
|
|
int start_a = parent_slno;
|
|
|
|
int length_a = parent_len;
|
|
|
|
int start_b = tlno;
|
|
|
|
int length_b = same - tlno;
|
|
|
|
|
|
|
|
struct line_number_mapping map_line_number_in_b_to_a = {
|
|
|
|
start_a, length_a, start_b, length_b
|
|
|
|
};
|
|
|
|
|
|
|
|
struct fingerprint *fingerprints_a = parent->fingerprints;
|
|
|
|
struct fingerprint *fingerprints_b = target->fingerprints;
|
|
|
|
|
|
|
|
int i, *result, *second_best_result,
|
|
|
|
*certainties, *similarities, similarity_count;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* max_search_distance_a means that given a line in B, compare it to
|
|
|
|
* the line in A that is closest to its position, and the lines in A
|
|
|
|
* that are no greater than max_search_distance_a lines away from the
|
|
|
|
* closest line in A.
|
|
|
|
*
|
|
|
|
* max_search_distance_b is an upper bound on the greatest possible
|
|
|
|
* distance between lines in B such that they will both be compared
|
|
|
|
* with the same line in A according to max_search_distance_a.
|
|
|
|
*/
|
|
|
|
int max_search_distance_a = 10, max_search_distance_b;
|
|
|
|
|
|
|
|
if (length_a <= 0)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (max_search_distance_a >= length_a)
|
|
|
|
max_search_distance_a = length_a ? length_a - 1 : 0;
|
|
|
|
|
|
|
|
max_search_distance_b = ((2 * max_search_distance_a + 1) * length_b
|
|
|
|
- 1) / length_a;
|
|
|
|
|
2021-03-13 16:17:22 +00:00
|
|
|
CALLOC_ARRAY(result, length_b);
|
|
|
|
CALLOC_ARRAY(second_best_result, length_b);
|
|
|
|
CALLOC_ARRAY(certainties, length_b);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
|
|
|
|
/* See get_similarity() for details of similarities. */
|
|
|
|
similarity_count = length_b * (max_search_distance_a * 2 + 1);
|
2021-03-13 16:17:22 +00:00
|
|
|
CALLOC_ARRAY(similarities, similarity_count);
|
blame: add a fingerprint heuristic to match ignored lines
This algorithm will replace the heuristic used to identify lines from
ignored commits with one that finds likely candidate lines in the
parent's version of the file. The actual replacement occurs in an
upcoming commit.
The old heuristic simply assigned lines in the target to the same line
number (plus offset) in the parent. The new function uses a
fingerprinting algorithm to detect similarity between lines.
The new heuristic is designed to accurately match changes made
mechanically by formatting tools such as clang-format and clang-tidy.
These tools make changes such as breaking up lines to fit within a
character limit or changing identifiers to fit with a naming convention.
The heuristic is not intended to match more extensive refactoring
changes and may give misleading results in such cases.
In most cases formatting tools preserve line ordering, so the heuristic
is optimised for such cases. (Some types of changes do reorder lines
e.g. sorting keep the line content identical, the git blame -M option
can already be used to address this). The reason that it is advantageous
to rely on ordering is due to source code repeating the same character
sequences often e.g. declaring an identifier on one line and using that
identifier on several subsequent lines. This means that lines can look
very similar to each other which presents a problem when doing fuzzy
matching. Relying on ordering gives us extra clues to point towards the
true match.
The heuristic operates on a single diff chunk change at a time. It
creates a “fingerprint” for each line on each side of the change.
Fingerprints are described in detail in the comment for `struct
fingerprint`, but essentially are a multiset of the character pairs in a
line. The heuristic first identifies the line in the target entry whose
fingerprint is most clearly matched to a line fingerprint in the parent
entry. Where fingerprints match identically, the position of the lines
is used as a tie-break. The heuristic locks in the best match, and
subtracts the fingerprint of the line in the target entry from the
fingerprint of the line in the parent entry to prevent other lines being
matched on the same parts of that line. It then repeats the process
recursively on the section of the chunk before the match, and then the
section of the chunk after the match.
Here's an example of the difference the fingerprinting makes. Consider
a file with two commits:
commit-a 1) void func_1(void *x, void *y);
commit-b 2) void func_2(void *x, void *y);
After a commit 'X', we have:
commit-X 1) void func_1(void *x,
commit-X 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
When we blame-ignored with the old algorithm, we get:
commit-a 1) void func_1(void *x,
commit-b 2) void *y);
commit-X 3) void func_2(void *x,
commit-X 4) void *y);
Where commit-b is blamed for 2 instead of 3. With the fingerprint
algorithm, we get:
commit-a 1) void func_1(void *x,
commit-a 2) void *y);
commit-b 3) void func_2(void *x,
commit-b 4) void *y);
Note line 2 could be matched with either commit-a or commit-b as it is
equally similar to both lines, but is matched with commit-a because its
position as a fraction of the new line range is more similar to commit-a
as a fraction of the old line range. Line 4 is also equally similar to
both lines, but as it appears after line 3 which will be matched first
it cannot be matched with an earlier line.
For many more examples, see t/t8014-blame-ignore-fuzzy.sh which contains
example parent and target files and the line numbers in the parent that
must be matched.
Signed-off-by: Michael Platings <michael@platin.gs>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-06-20 16:38:18 +00:00
|
|
|
|
|
|
|
for (i = 0; i < length_b; ++i) {
|
|
|
|
result[i] = -1;
|
|
|
|
second_best_result[i] = -1;
|
|
|
|
certainties[i] = CERTAINTY_NOT_CALCULATED;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < similarity_count; ++i)
|
|
|
|
similarities[i] = -1;
|
|
|
|
|
|
|
|
fuzzy_find_matching_lines_recurse(start_a, start_b,
|
|
|
|
length_a, length_b,
|
|
|
|
fingerprints_a + start_a,
|
|
|
|
fingerprints_b + start_b,
|
|
|
|
similarities,
|
|
|
|
certainties,
|
|
|
|
second_best_result,
|
|
|
|
result,
|
|
|
|
max_search_distance_a,
|
|
|
|
max_search_distance_b,
|
|
|
|
&map_line_number_in_b_to_a);
|
|
|
|
|
|
|
|
free(similarities);
|
|
|
|
free(certainties);
|
|
|
|
free(second_best_result);
|
|
|
|
|
|
|
|
return result;
|
|
|
|
}
|
|
|
|
|
2019-06-28 06:24:57 +00:00
|
|
|
static void fill_origin_fingerprints(struct blame_origin *o)
|
2019-05-15 21:45:01 +00:00
|
|
|
{
|
|
|
|
int *line_starts;
|
|
|
|
|
|
|
|
if (o->fingerprints)
|
|
|
|
return;
|
|
|
|
o->num_lines = find_line_starts(&line_starts, o->file.ptr,
|
|
|
|
o->file.size);
|
2021-03-13 16:17:22 +00:00
|
|
|
CALLOC_ARRAY(o->fingerprints, o->num_lines);
|
2019-06-20 16:38:19 +00:00
|
|
|
get_line_fingerprints(o->fingerprints, o->file.ptr, line_starts,
|
|
|
|
0, o->num_lines);
|
2019-05-15 21:45:01 +00:00
|
|
|
free(line_starts);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void drop_origin_fingerprints(struct blame_origin *o)
|
|
|
|
{
|
2019-06-20 16:38:19 +00:00
|
|
|
if (o->fingerprints) {
|
|
|
|
free_line_fingerprints(o->fingerprints, o->num_lines);
|
|
|
|
o->num_lines = 0;
|
|
|
|
FREE_AND_NULL(o->fingerprints);
|
|
|
|
}
|
2019-05-15 21:45:01 +00:00
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
/*
|
|
|
|
* Given an origin, prepare mmfile_t structure to be used by the
|
|
|
|
* diff machinery
|
|
|
|
*/
|
|
|
|
static void fill_origin_blob(struct diff_options *opt,
|
2019-05-15 21:45:01 +00:00
|
|
|
struct blame_origin *o, mmfile_t *file,
|
|
|
|
int *num_read_blob, int fill_fingerprints)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
if (!o->file.ptr) {
|
|
|
|
enum object_type type;
|
|
|
|
unsigned long file_size;
|
|
|
|
|
|
|
|
(*num_read_blob)++;
|
2017-10-31 18:19:11 +00:00
|
|
|
if (opt->flags.allow_textconv &&
|
2018-09-21 15:57:22 +00:00
|
|
|
textconv_object(opt->repo, o->path, o->mode,
|
|
|
|
&o->blob_oid, 1, &file->ptr, &file_size))
|
2017-05-24 05:15:35 +00:00
|
|
|
;
|
|
|
|
else
|
2023-03-28 13:58:50 +00:00
|
|
|
file->ptr = repo_read_object_file(the_repository,
|
|
|
|
&o->blob_oid, &type,
|
|
|
|
&file_size);
|
2017-05-24 05:15:35 +00:00
|
|
|
file->size = file_size;
|
|
|
|
|
|
|
|
if (!file->ptr)
|
|
|
|
die("Cannot read blob %s for path %s",
|
|
|
|
oid_to_hex(&o->blob_oid),
|
|
|
|
o->path);
|
|
|
|
o->file = *file;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
*file = o->file;
|
2019-05-15 21:45:01 +00:00
|
|
|
if (fill_fingerprints)
|
2019-06-28 06:24:57 +00:00
|
|
|
fill_origin_fingerprints(o);
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void drop_origin_blob(struct blame_origin *o)
|
|
|
|
{
|
2018-08-17 13:02:50 +00:00
|
|
|
FREE_AND_NULL(o->file.ptr);
|
2019-05-15 21:45:01 +00:00
|
|
|
drop_origin_fingerprints(o);
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Any merge of blames happens on lists of blames that arrived via
|
|
|
|
* different parents in a single suspect. In this case, we want to
|
|
|
|
* sort according to the suspect line numbers as opposed to the final
|
|
|
|
* image line numbers. The function body is somewhat longish because
|
|
|
|
* it avoids unnecessary writes.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static struct blame_entry *blame_merge(struct blame_entry *list1,
|
|
|
|
struct blame_entry *list2)
|
|
|
|
{
|
|
|
|
struct blame_entry *p1 = list1, *p2 = list2,
|
|
|
|
**tail = &list1;
|
|
|
|
|
|
|
|
if (!p1)
|
|
|
|
return p2;
|
|
|
|
if (!p2)
|
|
|
|
return p1;
|
|
|
|
|
|
|
|
if (p1->s_lno <= p2->s_lno) {
|
|
|
|
do {
|
|
|
|
tail = &p1->next;
|
2022-05-02 16:50:37 +00:00
|
|
|
if (!(p1 = *tail)) {
|
2017-05-24 05:15:35 +00:00
|
|
|
*tail = p2;
|
|
|
|
return list1;
|
|
|
|
}
|
|
|
|
} while (p1->s_lno <= p2->s_lno);
|
|
|
|
}
|
|
|
|
for (;;) {
|
|
|
|
*tail = p2;
|
|
|
|
do {
|
|
|
|
tail = &p2->next;
|
2022-05-02 16:50:37 +00:00
|
|
|
if (!(p2 = *tail)) {
|
2017-05-24 05:15:35 +00:00
|
|
|
*tail = p1;
|
|
|
|
return list1;
|
|
|
|
}
|
|
|
|
} while (p1->s_lno > p2->s_lno);
|
|
|
|
*tail = p1;
|
|
|
|
do {
|
|
|
|
tail = &p1->next;
|
2022-05-02 16:50:37 +00:00
|
|
|
if (!(p1 = *tail)) {
|
2017-05-24 05:15:35 +00:00
|
|
|
*tail = p2;
|
|
|
|
return list1;
|
|
|
|
}
|
|
|
|
} while (p1->s_lno <= p2->s_lno);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-07-16 16:58:20 +00:00
|
|
|
DEFINE_LIST_SORT(static, sort_blame_entries, struct blame_entry, next);
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Final image line numbers are all different, so we don't need a
|
|
|
|
* three-way comparison here.
|
|
|
|
*/
|
|
|
|
|
2022-07-16 16:58:20 +00:00
|
|
|
static int compare_blame_final(const struct blame_entry *e1,
|
|
|
|
const struct blame_entry *e2)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
2022-07-16 16:58:20 +00:00
|
|
|
return e1->lno > e2->lno ? 1 : -1;
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
|
2022-07-16 16:58:20 +00:00
|
|
|
static int compare_blame_suspect(const struct blame_entry *s1,
|
|
|
|
const struct blame_entry *s2)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
/*
|
|
|
|
* to allow for collating suspects, we sort according to the
|
|
|
|
* respective pointer value as the primary sorting criterion.
|
|
|
|
* The actual relation is pretty unimportant as long as it
|
|
|
|
* establishes a total order. Comparing as integers gives us
|
|
|
|
* that.
|
|
|
|
*/
|
|
|
|
if (s1->suspect != s2->suspect)
|
|
|
|
return (intptr_t)s1->suspect > (intptr_t)s2->suspect ? 1 : -1;
|
|
|
|
if (s1->s_lno == s2->s_lno)
|
|
|
|
return 0;
|
|
|
|
return s1->s_lno > s2->s_lno ? 1 : -1;
|
|
|
|
}
|
|
|
|
|
|
|
|
void blame_sort_final(struct blame_scoreboard *sb)
|
|
|
|
{
|
2022-07-16 16:58:20 +00:00
|
|
|
sort_blame_entries(&sb->ent, compare_blame_final);
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:36 +00:00
|
|
|
static int compare_commits_by_reverse_commit_date(const void *a,
|
|
|
|
const void *b,
|
|
|
|
void *c)
|
|
|
|
{
|
|
|
|
return -compare_commits_by_commit_date(a, b, c);
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
/*
|
|
|
|
* For debugging -- origin is refcounted, and this asserts that
|
|
|
|
* we do not underflow.
|
|
|
|
*/
|
|
|
|
static void sanity_check_refcnt(struct blame_scoreboard *sb)
|
|
|
|
{
|
|
|
|
int baa = 0;
|
|
|
|
struct blame_entry *ent;
|
|
|
|
|
|
|
|
for (ent = sb->ent; ent; ent = ent->next) {
|
|
|
|
/* Nobody should have zero or negative refcnt */
|
|
|
|
if (ent->suspect->refcnt <= 0) {
|
|
|
|
fprintf(stderr, "%s in %s has negative refcnt %d\n",
|
|
|
|
ent->suspect->path,
|
|
|
|
oid_to_hex(&ent->suspect->commit->object.oid),
|
|
|
|
ent->suspect->refcnt);
|
|
|
|
baa = 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (baa)
|
|
|
|
sb->on_sanity_fail(sb, baa);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If two blame entries that are next to each other came from
|
|
|
|
* contiguous lines in the same origin (i.e. <commit, path> pair),
|
|
|
|
* merge them together.
|
|
|
|
*/
|
|
|
|
void blame_coalesce(struct blame_scoreboard *sb)
|
|
|
|
{
|
|
|
|
struct blame_entry *ent, *next;
|
|
|
|
|
|
|
|
for (ent = sb->ent; ent && (next = ent->next); ent = next) {
|
|
|
|
if (ent->suspect == next->suspect &&
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
ent->s_lno + ent->num_lines == next->s_lno &&
|
blame: only coalesce lines that are adjacent in result
After blame has finished but before we produce any output, we coalesce
groups of lines that were adjacent in the original suspect (which may
have been split apart by lines in intermediate commits which went away).
However, this can cause incorrect output if the lines are not also
adjacent in the result. For instance, the case in t8003 has:
ABC
DEF
which becomes
ABC
SPLIT
DEF
Blaming only lines 1 and 3 in the result yields two blame groups (one
for each line) that were adjacent in the original. That's enough for us
to coalesce them into a single group, but that loses information: our
output routines assume they're adjacent in the result as well, and we
output:
<oid> 1) ABC
<oid> 2) SPLIT
This is nonsense for two reasons:
- we were asked about line 3, not line 2; we should not output the
SPLIT line at all
- commit <oid> did not touch the SPLIT line at all! We found the
correct blame for line 3, but the bug is actually in the output
stage, which is showing the wrong line number and content from the
final file.
We can fix this by only coalescing when both the suspect and result
lines are adjacent. That fixes this bug, but keeps coalescing in cases
where want it (e.g., the existing test in t8003 where SPLIT goes away,
and the lines really are adjacent in the result).
Reported-by: Nuthan Munaiah <nm6061@rit.edu>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-08-13 05:25:31 +00:00
|
|
|
ent->lno + ent->num_lines == next->lno &&
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
ent->ignored == next->ignored &&
|
|
|
|
ent->unblamable == next->unblamable) {
|
2017-05-24 05:15:35 +00:00
|
|
|
ent->num_lines += next->num_lines;
|
|
|
|
ent->next = next->next;
|
|
|
|
blame_origin_decref(next->suspect);
|
|
|
|
free(next);
|
|
|
|
ent->score = 0;
|
|
|
|
next = ent; /* again */
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sb->debug) /* sanity */
|
|
|
|
sanity_check_refcnt(sb);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Merge the given sorted list of blames into a preexisting origin.
|
|
|
|
* If there were no previous blames to that commit, it is entered into
|
|
|
|
* the commit priority queue of the score board.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static void queue_blames(struct blame_scoreboard *sb, struct blame_origin *porigin,
|
|
|
|
struct blame_entry *sorted)
|
|
|
|
{
|
|
|
|
if (porigin->suspects)
|
|
|
|
porigin->suspects = blame_merge(porigin->suspects, sorted);
|
|
|
|
else {
|
|
|
|
struct blame_origin *o;
|
2018-05-19 05:28:19 +00:00
|
|
|
for (o = get_blame_suspects(porigin->commit); o; o = o->next) {
|
2017-05-24 05:15:35 +00:00
|
|
|
if (o->suspects) {
|
|
|
|
porigin->suspects = sorted;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
porigin->suspects = sorted;
|
|
|
|
prio_queue_put(&sb->commits, porigin->commit);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:36 +00:00
|
|
|
/*
|
|
|
|
* Fill the blob_sha1 field of an origin if it hasn't, so that later
|
|
|
|
* call to fill_origin_blob() can use it to locate the data. blob_sha1
|
|
|
|
* for an origin is also used to pass the blame for the entire file to
|
|
|
|
* the parent to detect the case where a child's blob is identical to
|
|
|
|
* that of its parent's.
|
|
|
|
*
|
|
|
|
* This also fills origin->mode for corresponding tree path.
|
|
|
|
*/
|
2018-09-21 15:57:21 +00:00
|
|
|
static int fill_blob_sha1_and_mode(struct repository *r,
|
2018-08-13 16:14:41 +00:00
|
|
|
struct blame_origin *origin)
|
2017-05-24 05:15:36 +00:00
|
|
|
{
|
|
|
|
if (!is_null_oid(&origin->blob_oid))
|
|
|
|
return 0;
|
2019-06-27 09:28:49 +00:00
|
|
|
if (get_tree_entry(r, &origin->commit->object.oid, origin->path, &origin->blob_oid, &origin->mode))
|
2017-05-24 05:15:36 +00:00
|
|
|
goto error_out;
|
2018-09-21 15:57:21 +00:00
|
|
|
if (oid_object_info(r, &origin->blob_oid, NULL) != OBJ_BLOB)
|
2017-05-24 05:15:36 +00:00
|
|
|
goto error_out;
|
|
|
|
return 0;
|
|
|
|
error_out:
|
|
|
|
oidclr(&origin->blob_oid);
|
|
|
|
origin->mode = S_IFINVALID;
|
|
|
|
return -1;
|
|
|
|
}
|
|
|
|
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
struct blame_bloom_data {
|
|
|
|
/*
|
|
|
|
* Changed-path Bloom filter keys. These can help prevent
|
|
|
|
* computing diffs against first parents, but we need to
|
|
|
|
* expand the list as code is moved or files are renamed.
|
|
|
|
*/
|
|
|
|
struct bloom_filter_settings *settings;
|
|
|
|
struct bloom_key **keys;
|
|
|
|
int nr;
|
|
|
|
int alloc;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int bloom_count_queries = 0;
|
|
|
|
static int bloom_count_no = 0;
|
|
|
|
static int maybe_changed_path(struct repository *r,
|
|
|
|
struct blame_origin *origin,
|
|
|
|
struct blame_bloom_data *bd)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
struct bloom_filter *filter;
|
|
|
|
|
|
|
|
if (!bd)
|
|
|
|
return 1;
|
|
|
|
|
2020-06-17 09:14:10 +00:00
|
|
|
if (commit_graph_generation(origin->commit) == GENERATION_NUMBER_INFINITY)
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
return 1;
|
|
|
|
|
bloom: split 'get_bloom_filter()' in two
'get_bloom_filter' takes a flag to control whether it will compute a
Bloom filter if the requested one is missing. In the next patch, we'll
add yet another parameter to this method, which would force all but one
caller to specify an extra 'NULL' parameter at the end.
Instead of doing this, split 'get_bloom_filter' into two functions:
'get_bloom_filter' and 'get_or_compute_bloom_filter'. The former only
looks up a Bloom filter (and does not compute one if it's missing,
thus dropping the 'compute_if_not_present' flag). The latter does
compute missing Bloom filters, with an additional parameter to store
whether or not it needed to do so.
This simplifies many call-sites, since the majority of existing callers
to 'get_bloom_filter' do not want missing Bloom filters to be computed
(so they can drop the parameter entirely and use the simpler version of
the function).
While we're at it, instrument the new 'get_or_compute_bloom_filter()'
with counters in the 'write_commit_graph_context' struct which store
the number of filters that we did and didn't compute, as well as filters
that were truncated.
It would be nice to drop the 'compute_if_not_present' flag entirely,
since all remaining callers of 'get_or_compute_bloom_filter' pass it as
'1', but this will change in a future patch and hence cannot be removed.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-16 18:07:32 +00:00
|
|
|
filter = get_bloom_filter(r, origin->commit);
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
|
|
|
|
if (!filter)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
bloom_count_queries++;
|
|
|
|
for (i = 0; i < bd->nr; i++) {
|
|
|
|
if (bloom_filter_contains(filter,
|
|
|
|
bd->keys[i],
|
|
|
|
bd->settings))
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
bloom_count_no++;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static void add_bloom_key(struct blame_bloom_data *bd,
|
|
|
|
const char *path)
|
|
|
|
{
|
|
|
|
if (!bd)
|
|
|
|
return;
|
|
|
|
|
|
|
|
if (bd->nr >= bd->alloc) {
|
|
|
|
bd->alloc *= 2;
|
|
|
|
REALLOC_ARRAY(bd->keys, bd->alloc);
|
|
|
|
}
|
|
|
|
|
|
|
|
bd->keys[bd->nr] = xmalloc(sizeof(struct bloom_key));
|
|
|
|
fill_bloom_key(path, strlen(path), bd->keys[bd->nr], bd->settings);
|
|
|
|
bd->nr++;
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
/*
|
|
|
|
* We have an origin -- check if the same path exists in the
|
|
|
|
* parent and return an origin structure to represent it.
|
|
|
|
*/
|
2018-09-21 15:57:24 +00:00
|
|
|
static struct blame_origin *find_origin(struct repository *r,
|
|
|
|
struct commit *parent,
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
struct blame_origin *origin,
|
|
|
|
struct blame_bloom_data *bd)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
struct blame_origin *porigin;
|
|
|
|
struct diff_options diff_opts;
|
|
|
|
const char *paths[2];
|
|
|
|
|
|
|
|
/* First check any existing origins */
|
2018-05-19 05:28:19 +00:00
|
|
|
for (porigin = get_blame_suspects(parent); porigin; porigin = porigin->next)
|
2017-05-24 05:15:35 +00:00
|
|
|
if (!strcmp(porigin->path, origin->path)) {
|
|
|
|
/*
|
|
|
|
* The same path between origin and its parent
|
|
|
|
* without renaming -- the most common case.
|
|
|
|
*/
|
|
|
|
return blame_origin_incref (porigin);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* See if the origin->path is different between parent
|
|
|
|
* and origin first. Most of the time they are the
|
|
|
|
* same and diff-tree is fairly efficient about this.
|
|
|
|
*/
|
2018-09-21 15:57:24 +00:00
|
|
|
repo_diff_setup(r, &diff_opts);
|
2017-10-31 18:19:11 +00:00
|
|
|
diff_opts.flags.recursive = 1;
|
2017-05-24 05:15:35 +00:00
|
|
|
diff_opts.detect_rename = 0;
|
|
|
|
diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
|
|
|
|
paths[0] = origin->path;
|
|
|
|
paths[1] = NULL;
|
|
|
|
|
|
|
|
parse_pathspec(&diff_opts.pathspec,
|
|
|
|
PATHSPEC_ALL_MAGIC & ~PATHSPEC_LITERAL,
|
|
|
|
PATHSPEC_LITERAL_PATH, "", paths);
|
|
|
|
diff_setup_done(&diff_opts);
|
|
|
|
|
|
|
|
if (is_null_oid(&origin->commit->object.oid))
|
2018-04-06 19:09:38 +00:00
|
|
|
do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
else {
|
|
|
|
int compute_diff = 1;
|
|
|
|
if (origin->commit->parents &&
|
2020-09-08 21:10:53 +00:00
|
|
|
oideq(&parent->object.oid,
|
|
|
|
&origin->commit->parents->item->object.oid))
|
2020-04-23 21:03:03 +00:00
|
|
|
compute_diff = maybe_changed_path(r, origin, bd);
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
|
|
|
|
if (compute_diff)
|
|
|
|
diff_tree_oid(get_commit_tree_oid(parent),
|
|
|
|
get_commit_tree_oid(origin->commit),
|
|
|
|
"", &diff_opts);
|
|
|
|
}
|
2017-05-24 05:15:35 +00:00
|
|
|
diffcore_std(&diff_opts);
|
|
|
|
|
|
|
|
if (!diff_queued_diff.nr) {
|
|
|
|
/* The path is the same as parent */
|
|
|
|
porigin = get_origin(parent, origin->path);
|
|
|
|
oidcpy(&porigin->blob_oid, &origin->blob_oid);
|
|
|
|
porigin->mode = origin->mode;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Since origin->path is a pathspec, if the parent
|
|
|
|
* commit had it as a directory, we will see a whole
|
|
|
|
* bunch of deletion of files in the directory that we
|
|
|
|
* do not care about.
|
|
|
|
*/
|
|
|
|
int i;
|
|
|
|
struct diff_filepair *p = NULL;
|
|
|
|
for (i = 0; i < diff_queued_diff.nr; i++) {
|
|
|
|
const char *name;
|
|
|
|
p = diff_queued_diff.queue[i];
|
|
|
|
name = p->one->path ? p->one->path : p->two->path;
|
|
|
|
if (!strcmp(name, origin->path))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (!p)
|
|
|
|
die("internal error in blame::find_origin");
|
|
|
|
switch (p->status) {
|
|
|
|
default:
|
|
|
|
die("internal error in blame::find_origin (%c)",
|
|
|
|
p->status);
|
|
|
|
case 'M':
|
|
|
|
porigin = get_origin(parent, origin->path);
|
|
|
|
oidcpy(&porigin->blob_oid, &p->one->oid);
|
|
|
|
porigin->mode = p->one->mode;
|
|
|
|
break;
|
|
|
|
case 'A':
|
|
|
|
case 'T':
|
|
|
|
/* Did not exist in parent, or type changed */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
diff_flush(&diff_opts);
|
|
|
|
return porigin;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We have an origin -- find the path that corresponds to it in its
|
|
|
|
* parent and return an origin structure to represent it.
|
|
|
|
*/
|
2018-09-21 15:57:24 +00:00
|
|
|
static struct blame_origin *find_rename(struct repository *r,
|
|
|
|
struct commit *parent,
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
struct blame_origin *origin,
|
|
|
|
struct blame_bloom_data *bd)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
struct blame_origin *porigin = NULL;
|
|
|
|
struct diff_options diff_opts;
|
|
|
|
int i;
|
|
|
|
|
2018-09-21 15:57:24 +00:00
|
|
|
repo_diff_setup(r, &diff_opts);
|
2017-10-31 18:19:11 +00:00
|
|
|
diff_opts.flags.recursive = 1;
|
2017-05-24 05:15:35 +00:00
|
|
|
diff_opts.detect_rename = DIFF_DETECT_RENAME;
|
|
|
|
diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
|
|
|
|
diff_opts.single_follow = origin->path;
|
|
|
|
diff_setup_done(&diff_opts);
|
|
|
|
|
|
|
|
if (is_null_oid(&origin->commit->object.oid))
|
2018-04-06 19:09:38 +00:00
|
|
|
do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
|
2017-05-24 05:15:35 +00:00
|
|
|
else
|
2018-04-06 19:09:38 +00:00
|
|
|
diff_tree_oid(get_commit_tree_oid(parent),
|
|
|
|
get_commit_tree_oid(origin->commit),
|
2017-06-19 19:38:44 +00:00
|
|
|
"", &diff_opts);
|
2017-05-24 05:15:35 +00:00
|
|
|
diffcore_std(&diff_opts);
|
|
|
|
|
|
|
|
for (i = 0; i < diff_queued_diff.nr; i++) {
|
|
|
|
struct diff_filepair *p = diff_queued_diff.queue[i];
|
|
|
|
if ((p->status == 'R' || p->status == 'C') &&
|
|
|
|
!strcmp(p->two->path, origin->path)) {
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
add_bloom_key(bd, p->one->path);
|
2017-05-24 05:15:35 +00:00
|
|
|
porigin = get_origin(parent, p->one->path);
|
|
|
|
oidcpy(&porigin->blob_oid, &p->one->oid);
|
|
|
|
porigin->mode = p->one->mode;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
diff_flush(&diff_opts);
|
|
|
|
return porigin;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Append a new blame entry to a given output queue.
|
|
|
|
*/
|
|
|
|
static void add_blame_entry(struct blame_entry ***queue,
|
|
|
|
const struct blame_entry *src)
|
|
|
|
{
|
|
|
|
struct blame_entry *e = xmalloc(sizeof(*e));
|
|
|
|
memcpy(e, src, sizeof(*e));
|
|
|
|
blame_origin_incref(e->suspect);
|
|
|
|
|
|
|
|
e->next = **queue;
|
|
|
|
**queue = e;
|
|
|
|
*queue = &e->next;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* src typically is on-stack; we want to copy the information in it to
|
|
|
|
* a malloced blame_entry that gets added to the given queue. The
|
|
|
|
* origin of dst loses a refcnt.
|
|
|
|
*/
|
|
|
|
static void dup_entry(struct blame_entry ***queue,
|
|
|
|
struct blame_entry *dst, struct blame_entry *src)
|
|
|
|
{
|
|
|
|
blame_origin_incref(src->suspect);
|
|
|
|
blame_origin_decref(dst->suspect);
|
|
|
|
memcpy(dst, src, sizeof(*src));
|
|
|
|
dst->next = **queue;
|
|
|
|
**queue = dst;
|
|
|
|
*queue = &dst->next;
|
|
|
|
}
|
|
|
|
|
|
|
|
const char *blame_nth_line(struct blame_scoreboard *sb, long lno)
|
|
|
|
{
|
|
|
|
return sb->final_buf + sb->lineno[lno];
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* It is known that lines between tlno to same came from parent, and e
|
|
|
|
* has an overlap with that range. it also is known that parent's
|
|
|
|
* line plno corresponds to e's line tlno.
|
|
|
|
*
|
|
|
|
* <---- e ----->
|
|
|
|
* <------>
|
|
|
|
* <------------>
|
|
|
|
* <------------>
|
|
|
|
* <------------------>
|
|
|
|
*
|
|
|
|
* Split e into potentially three parts; before this chunk, the chunk
|
|
|
|
* to be blamed for the parent, and after that portion.
|
|
|
|
*/
|
|
|
|
static void split_overlap(struct blame_entry *split,
|
|
|
|
struct blame_entry *e,
|
|
|
|
int tlno, int plno, int same,
|
|
|
|
struct blame_origin *parent)
|
|
|
|
{
|
|
|
|
int chunk_end_lno;
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
int i;
|
2017-05-24 05:15:35 +00:00
|
|
|
memset(split, 0, sizeof(struct blame_entry [3]));
|
|
|
|
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
for (i = 0; i < 3; i++) {
|
|
|
|
split[i].ignored = e->ignored;
|
|
|
|
split[i].unblamable = e->unblamable;
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
if (e->s_lno < tlno) {
|
|
|
|
/* there is a pre-chunk part not blamed on parent */
|
|
|
|
split[0].suspect = blame_origin_incref(e->suspect);
|
|
|
|
split[0].lno = e->lno;
|
|
|
|
split[0].s_lno = e->s_lno;
|
|
|
|
split[0].num_lines = tlno - e->s_lno;
|
|
|
|
split[1].lno = e->lno + tlno - e->s_lno;
|
|
|
|
split[1].s_lno = plno;
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
split[1].lno = e->lno;
|
|
|
|
split[1].s_lno = plno + (e->s_lno - tlno);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (same < e->s_lno + e->num_lines) {
|
|
|
|
/* there is a post-chunk part not blamed on parent */
|
|
|
|
split[2].suspect = blame_origin_incref(e->suspect);
|
|
|
|
split[2].lno = e->lno + (same - e->s_lno);
|
|
|
|
split[2].s_lno = e->s_lno + (same - e->s_lno);
|
|
|
|
split[2].num_lines = e->s_lno + e->num_lines - same;
|
|
|
|
chunk_end_lno = split[2].lno;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
chunk_end_lno = e->lno + e->num_lines;
|
|
|
|
split[1].num_lines = chunk_end_lno - split[1].lno;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* if it turns out there is nothing to blame the parent for,
|
|
|
|
* forget about the splitting. !split[1].suspect signals this.
|
|
|
|
*/
|
|
|
|
if (split[1].num_lines < 1)
|
|
|
|
return;
|
|
|
|
split[1].suspect = blame_origin_incref(parent);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* split_overlap() divided an existing blame e into up to three parts
|
|
|
|
* in split. Any assigned blame is moved to queue to
|
|
|
|
* reflect the split.
|
|
|
|
*/
|
|
|
|
static void split_blame(struct blame_entry ***blamed,
|
|
|
|
struct blame_entry ***unblamed,
|
|
|
|
struct blame_entry *split,
|
|
|
|
struct blame_entry *e)
|
|
|
|
{
|
|
|
|
if (split[0].suspect && split[2].suspect) {
|
|
|
|
/* The first part (reuse storage for the existing entry e) */
|
|
|
|
dup_entry(unblamed, e, &split[0]);
|
|
|
|
|
|
|
|
/* The last part -- me */
|
|
|
|
add_blame_entry(unblamed, &split[2]);
|
|
|
|
|
|
|
|
/* ... and the middle part -- parent */
|
|
|
|
add_blame_entry(blamed, &split[1]);
|
|
|
|
}
|
|
|
|
else if (!split[0].suspect && !split[2].suspect)
|
|
|
|
/*
|
|
|
|
* The parent covers the entire area; reuse storage for
|
|
|
|
* e and replace it with the parent.
|
|
|
|
*/
|
|
|
|
dup_entry(blamed, e, &split[1]);
|
|
|
|
else if (split[0].suspect) {
|
|
|
|
/* me and then parent */
|
|
|
|
dup_entry(unblamed, e, &split[0]);
|
|
|
|
add_blame_entry(blamed, &split[1]);
|
|
|
|
}
|
|
|
|
else {
|
|
|
|
/* parent and then me */
|
|
|
|
dup_entry(blamed, e, &split[1]);
|
|
|
|
add_blame_entry(unblamed, &split[2]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* After splitting the blame, the origins used by the
|
|
|
|
* on-stack blame_entry should lose one refcnt each.
|
|
|
|
*/
|
|
|
|
static void decref_split(struct blame_entry *split)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
for (i = 0; i < 3; i++)
|
|
|
|
blame_origin_decref(split[i].suspect);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* reverse_blame reverses the list given in head, appending tail.
|
|
|
|
* That allows us to build lists in reverse order, then reverse them
|
|
|
|
* afterwards. This can be faster than building the list in proper
|
|
|
|
* order right away. The reason is that building in proper order
|
|
|
|
* requires writing a link in the _previous_ element, while building
|
|
|
|
* in reverse order just requires placing the list head into the
|
|
|
|
* _current_ element.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static struct blame_entry *reverse_blame(struct blame_entry *head,
|
|
|
|
struct blame_entry *tail)
|
|
|
|
{
|
|
|
|
while (head) {
|
|
|
|
struct blame_entry *next = head->next;
|
|
|
|
head->next = tail;
|
|
|
|
tail = head;
|
|
|
|
head = next;
|
|
|
|
}
|
|
|
|
return tail;
|
|
|
|
}
|
|
|
|
|
2019-05-15 21:44:58 +00:00
|
|
|
/*
|
|
|
|
* Splits a blame entry into two entries at 'len' lines. The original 'e'
|
|
|
|
* consists of len lines, i.e. [e->lno, e->lno + len), and the second part,
|
|
|
|
* which is returned, consists of the remainder: [e->lno + len, e->lno +
|
|
|
|
* e->num_lines). The caller needs to sort out the reference counting for the
|
|
|
|
* new entry's suspect.
|
|
|
|
*/
|
|
|
|
static struct blame_entry *split_blame_at(struct blame_entry *e, int len,
|
|
|
|
struct blame_origin *new_suspect)
|
|
|
|
{
|
|
|
|
struct blame_entry *n = xcalloc(1, sizeof(struct blame_entry));
|
|
|
|
|
|
|
|
n->suspect = new_suspect;
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
n->ignored = e->ignored;
|
|
|
|
n->unblamable = e->unblamable;
|
2019-05-15 21:44:58 +00:00
|
|
|
n->lno = e->lno + len;
|
|
|
|
n->s_lno = e->s_lno + len;
|
|
|
|
n->num_lines = e->num_lines - len;
|
|
|
|
e->num_lines = len;
|
|
|
|
e->score = 0;
|
|
|
|
return n;
|
|
|
|
}
|
|
|
|
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
struct blame_line_tracker {
|
|
|
|
int is_parent;
|
|
|
|
int s_lno;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int are_lines_adjacent(struct blame_line_tracker *first,
|
|
|
|
struct blame_line_tracker *second)
|
|
|
|
{
|
|
|
|
return first->is_parent == second->is_parent &&
|
|
|
|
first->s_lno + 1 == second->s_lno;
|
|
|
|
}
|
|
|
|
|
2019-06-20 16:38:19 +00:00
|
|
|
static int scan_parent_range(struct fingerprint *p_fps,
|
|
|
|
struct fingerprint *t_fps, int t_idx,
|
|
|
|
int from, int nr_lines)
|
|
|
|
{
|
|
|
|
int sim, p_idx;
|
|
|
|
#define FINGERPRINT_FILE_THRESHOLD 10
|
|
|
|
int best_sim_val = FINGERPRINT_FILE_THRESHOLD;
|
|
|
|
int best_sim_idx = -1;
|
|
|
|
|
|
|
|
for (p_idx = from; p_idx < from + nr_lines; p_idx++) {
|
|
|
|
sim = fingerprint_similarity(&t_fps[t_idx], &p_fps[p_idx]);
|
|
|
|
if (sim < best_sim_val)
|
|
|
|
continue;
|
|
|
|
/* Break ties with the closest-to-target line number */
|
|
|
|
if (sim == best_sim_val && best_sim_idx != -1 &&
|
|
|
|
abs(best_sim_idx - t_idx) < abs(p_idx - t_idx))
|
|
|
|
continue;
|
|
|
|
best_sim_val = sim;
|
|
|
|
best_sim_idx = p_idx;
|
|
|
|
}
|
|
|
|
return best_sim_idx;
|
|
|
|
}
|
|
|
|
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
/*
|
2019-06-20 16:38:19 +00:00
|
|
|
* The first pass checks the blame entry (from the target) against the parent's
|
|
|
|
* diff chunk. If that fails for a line, the second pass tries to match that
|
|
|
|
* line to any part of parent file. That catches cases where a change was
|
|
|
|
* broken into two chunks by 'context.'
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
*/
|
|
|
|
static void guess_line_blames(struct blame_origin *parent,
|
|
|
|
struct blame_origin *target,
|
|
|
|
int tlno, int offset, int same, int parent_len,
|
|
|
|
struct blame_line_tracker *line_blames)
|
|
|
|
{
|
|
|
|
int i, best_idx, target_idx;
|
|
|
|
int parent_slno = tlno + offset;
|
2019-06-20 16:38:19 +00:00
|
|
|
int *fuzzy_matches;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
|
2019-06-20 16:38:19 +00:00
|
|
|
fuzzy_matches = fuzzy_find_matching_lines(parent, target,
|
|
|
|
tlno, parent_slno, same,
|
|
|
|
parent_len);
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
for (i = 0; i < same - tlno; i++) {
|
|
|
|
target_idx = tlno + i;
|
2019-06-20 16:38:19 +00:00
|
|
|
if (fuzzy_matches && fuzzy_matches[i] >= 0) {
|
|
|
|
best_idx = fuzzy_matches[i];
|
|
|
|
} else {
|
|
|
|
best_idx = scan_parent_range(parent->fingerprints,
|
|
|
|
target->fingerprints,
|
|
|
|
target_idx, 0,
|
|
|
|
parent->num_lines);
|
|
|
|
}
|
|
|
|
if (best_idx >= 0) {
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
line_blames[i].is_parent = 1;
|
|
|
|
line_blames[i].s_lno = best_idx;
|
|
|
|
} else {
|
|
|
|
line_blames[i].is_parent = 0;
|
|
|
|
line_blames[i].s_lno = target_idx;
|
|
|
|
}
|
|
|
|
}
|
2019-06-20 16:38:19 +00:00
|
|
|
free(fuzzy_matches);
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This decides which parts of a blame entry go to the parent (added to the
|
|
|
|
* ignoredp list) and which stay with the target (added to the diffp list). The
|
|
|
|
* actual decision was made in a separate heuristic function, and those answers
|
|
|
|
* for the lines in 'e' are in line_blames. This consumes e, essentially
|
|
|
|
* putting it on a list.
|
|
|
|
*
|
|
|
|
* Note that the blame entries on the ignoredp list are not necessarily sorted
|
|
|
|
* with respect to the parent's line numbers yet.
|
|
|
|
*/
|
|
|
|
static void ignore_blame_entry(struct blame_entry *e,
|
|
|
|
struct blame_origin *parent,
|
|
|
|
struct blame_entry **diffp,
|
|
|
|
struct blame_entry **ignoredp,
|
|
|
|
struct blame_line_tracker *line_blames)
|
|
|
|
{
|
|
|
|
int entry_len, nr_lines, i;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We carve new entries off the front of e. Each entry comes from a
|
|
|
|
* contiguous chunk of lines: adjacent lines from the same origin
|
|
|
|
* (either the parent or the target).
|
|
|
|
*/
|
|
|
|
entry_len = 1;
|
|
|
|
nr_lines = e->num_lines; /* e changes in the loop */
|
|
|
|
for (i = 0; i < nr_lines; i++) {
|
|
|
|
struct blame_entry *next = NULL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We are often adjacent to the next line - only split the blame
|
|
|
|
* entry when we have to.
|
|
|
|
*/
|
|
|
|
if (i + 1 < nr_lines) {
|
|
|
|
if (are_lines_adjacent(&line_blames[i],
|
|
|
|
&line_blames[i + 1])) {
|
|
|
|
entry_len++;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
next = split_blame_at(e, entry_len,
|
|
|
|
blame_origin_incref(e->suspect));
|
|
|
|
}
|
|
|
|
if (line_blames[i].is_parent) {
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
e->ignored = 1;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
blame_origin_decref(e->suspect);
|
|
|
|
e->suspect = blame_origin_incref(parent);
|
|
|
|
e->s_lno = line_blames[i - entry_len + 1].s_lno;
|
|
|
|
e->next = *ignoredp;
|
|
|
|
*ignoredp = e;
|
|
|
|
} else {
|
blame: add config options for the output of ignored or unblamable lines
When ignoring commits, the commit that is blamed might not be
responsible for the change, due to the inaccuracy of our heuristic.
Users might want to know when a particular line has a potentially
inaccurate blame.
Furthermore, guess_line_blames() may fail to find any parent commit for
a given line touched by an ignored commit. Those 'unblamable' lines
remain blamed on an ignored commit. Users might want to know if a line
is unblamable so that they do not spend time investigating a commit they
know is uninteresting.
This patch adds two config options to mark these two types of lines in
the output of blame.
The first option can identify ignored lines by specifying
blame.markIgnoredLines. When this option is set, each blame line that
was blamed on a commit other than the ignored commit is marked with a
'?'.
For example:
278b6158d6fdb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
?278b6158d6fd (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
where the '?' is placed before the commit, and the hash has one fewer
characters.
Sometimes we are unable to even guess at what ancestor commit touched a
line. These lines are 'unblamable.' The second option,
blame.markUnblamableLines, will mark the line with '*'.
For example, say we ignore e5e8d36d04cbe, yet we are unable to blame
this line on another commit:
e5e8d36d04cbe (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
appears as:
*e5e8d36d04cb (Barret Rhoden 2016-04-11 13:57:54 -0400 26)
When these config options are used together, every line touched by an
ignored commit will be marked with either a '?' or a '*'.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:45:00 +00:00
|
|
|
e->unblamable = 1;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
/* e->s_lno is already in the target's address space. */
|
|
|
|
e->next = *diffp;
|
|
|
|
*diffp = e;
|
|
|
|
}
|
|
|
|
assert(e->num_lines == entry_len);
|
|
|
|
e = next;
|
|
|
|
entry_len = 1;
|
|
|
|
}
|
|
|
|
assert(!e);
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
/*
|
|
|
|
* Process one hunk from the patch between the current suspect for
|
|
|
|
* blame_entry e and its parent. This first blames any unfinished
|
|
|
|
* entries before the chunk (which is where target and parent start
|
|
|
|
* differing) on the parent, and then splits blame entries at the
|
|
|
|
* start and at the end of the difference region. Since use of -M and
|
|
|
|
* -C options may lead to overlapping/duplicate source line number
|
|
|
|
* ranges, all we can rely on from sorting/merging is the order of the
|
|
|
|
* first suspect line number.
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
*
|
|
|
|
* tlno: line number in the target where this chunk begins
|
|
|
|
* same: line number in the target where this chunk ends
|
|
|
|
* offset: add to tlno to get the chunk starting point in the parent
|
|
|
|
* parent_len: number of lines in the parent chunk
|
2017-05-24 05:15:35 +00:00
|
|
|
*/
|
|
|
|
static void blame_chunk(struct blame_entry ***dstq, struct blame_entry ***srcq,
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
int tlno, int offset, int same, int parent_len,
|
|
|
|
struct blame_origin *parent,
|
|
|
|
struct blame_origin *target, int ignore_diffs)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
struct blame_entry *e = **srcq;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
struct blame_entry *samep = NULL, *diffp = NULL, *ignoredp = NULL;
|
|
|
|
struct blame_line_tracker *line_blames = NULL;
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
while (e && e->s_lno < tlno) {
|
|
|
|
struct blame_entry *next = e->next;
|
|
|
|
/*
|
|
|
|
* current record starts before differing portion. If
|
|
|
|
* it reaches into it, we need to split it up and
|
|
|
|
* examine the second part separately.
|
|
|
|
*/
|
|
|
|
if (e->s_lno + e->num_lines > tlno) {
|
|
|
|
/* Move second half to a new record */
|
2019-05-15 21:44:58 +00:00
|
|
|
struct blame_entry *n;
|
|
|
|
|
|
|
|
n = split_blame_at(e, tlno - e->s_lno, e->suspect);
|
2017-05-24 05:15:35 +00:00
|
|
|
/* Push new record to diffp */
|
|
|
|
n->next = diffp;
|
|
|
|
diffp = n;
|
|
|
|
} else
|
|
|
|
blame_origin_decref(e->suspect);
|
|
|
|
/* Pass blame for everything before the differing
|
|
|
|
* chunk to the parent */
|
|
|
|
e->suspect = blame_origin_incref(parent);
|
|
|
|
e->s_lno += offset;
|
|
|
|
e->next = samep;
|
|
|
|
samep = e;
|
|
|
|
e = next;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* As we don't know how much of a common stretch after this
|
|
|
|
* diff will occur, the currently blamed parts are all that we
|
|
|
|
* can assign to the parent for now.
|
|
|
|
*/
|
|
|
|
|
|
|
|
if (samep) {
|
|
|
|
**dstq = reverse_blame(samep, **dstq);
|
|
|
|
*dstq = &samep->next;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* Prepend the split off portions: everything after e starts
|
|
|
|
* after the blameable portion.
|
|
|
|
*/
|
|
|
|
e = reverse_blame(diffp, e);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now retain records on the target while parts are different
|
|
|
|
* from the parent.
|
|
|
|
*/
|
|
|
|
samep = NULL;
|
|
|
|
diffp = NULL;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
|
|
|
|
if (ignore_diffs && same - tlno > 0) {
|
2021-03-13 16:17:22 +00:00
|
|
|
CALLOC_ARRAY(line_blames, same - tlno);
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
guess_line_blames(parent, target, tlno, offset, same,
|
|
|
|
parent_len, line_blames);
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
while (e && e->s_lno < same) {
|
|
|
|
struct blame_entry *next = e->next;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If current record extends into sameness, need to split.
|
|
|
|
*/
|
|
|
|
if (e->s_lno + e->num_lines > same) {
|
|
|
|
/*
|
|
|
|
* Move second half to a new record to be
|
|
|
|
* processed by later chunks
|
|
|
|
*/
|
2019-05-15 21:44:58 +00:00
|
|
|
struct blame_entry *n;
|
|
|
|
|
|
|
|
n = split_blame_at(e, same - e->s_lno,
|
|
|
|
blame_origin_incref(e->suspect));
|
2017-05-24 05:15:35 +00:00
|
|
|
/* Push new record to samep */
|
|
|
|
n->next = samep;
|
|
|
|
samep = n;
|
|
|
|
}
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
if (ignore_diffs) {
|
2019-06-28 06:24:57 +00:00
|
|
|
ignore_blame_entry(e, parent, &diffp, &ignoredp,
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
line_blames + e->s_lno - tlno);
|
|
|
|
} else {
|
|
|
|
e->next = diffp;
|
|
|
|
diffp = e;
|
|
|
|
}
|
2017-05-24 05:15:35 +00:00
|
|
|
e = next;
|
|
|
|
}
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
free(line_blames);
|
|
|
|
if (ignoredp) {
|
|
|
|
/*
|
|
|
|
* Note ignoredp is not sorted yet, and thus neither is dstq.
|
|
|
|
* That list must be sorted before we queue_blames(). We defer
|
|
|
|
* sorting until after all diff hunks are processed, so that
|
|
|
|
* guess_line_blames() can pick *any* line in the parent. The
|
|
|
|
* slight drawback is that we end up sorting all blame entries
|
|
|
|
* passed to the parent, including those that are unrelated to
|
|
|
|
* changes made by the ignored commit.
|
|
|
|
*/
|
|
|
|
**dstq = reverse_blame(ignoredp, **dstq);
|
|
|
|
*dstq = &ignoredp->next;
|
|
|
|
}
|
2017-05-24 05:15:35 +00:00
|
|
|
**srcq = reverse_blame(diffp, reverse_blame(samep, e));
|
|
|
|
/* Move across elements that are in the unblamable portion */
|
|
|
|
if (diffp)
|
|
|
|
*srcq = &diffp->next;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct blame_chunk_cb_data {
|
|
|
|
struct blame_origin *parent;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
struct blame_origin *target;
|
2017-05-24 05:15:35 +00:00
|
|
|
long offset;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
int ignore_diffs;
|
2017-05-24 05:15:35 +00:00
|
|
|
struct blame_entry **dstq;
|
|
|
|
struct blame_entry **srcq;
|
|
|
|
};
|
|
|
|
|
|
|
|
/* diff chunks are from parent to target */
|
|
|
|
static int blame_chunk_cb(long start_a, long count_a,
|
|
|
|
long start_b, long count_b, void *data)
|
|
|
|
{
|
|
|
|
struct blame_chunk_cb_data *d = data;
|
|
|
|
if (start_a - start_b != d->offset)
|
|
|
|
die("internal error in blame::blame_chunk_cb");
|
|
|
|
blame_chunk(&d->dstq, &d->srcq, start_b, start_a - start_b,
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
start_b + count_b, count_a, d->parent, d->target,
|
|
|
|
d->ignore_diffs);
|
2017-05-24 05:15:35 +00:00
|
|
|
d->offset = start_a + count_a - (start_b + count_b);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We are looking at the origin 'target' and aiming to pass blame
|
|
|
|
* for the lines it is suspected to its parent. Run diff to find
|
|
|
|
* which lines came from parent and pass blame for them.
|
|
|
|
*/
|
|
|
|
static void pass_blame_to_parent(struct blame_scoreboard *sb,
|
|
|
|
struct blame_origin *target,
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
struct blame_origin *parent, int ignore_diffs)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
mmfile_t file_p, file_o;
|
|
|
|
struct blame_chunk_cb_data d;
|
|
|
|
struct blame_entry *newdest = NULL;
|
|
|
|
|
|
|
|
if (!target->suspects)
|
|
|
|
return; /* nothing remains for this target */
|
|
|
|
|
|
|
|
d.parent = parent;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
d.target = target;
|
2017-05-24 05:15:35 +00:00
|
|
|
d.offset = 0;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
d.ignore_diffs = ignore_diffs;
|
2017-05-24 05:15:35 +00:00
|
|
|
d.dstq = &newdest; d.srcq = &target->suspects;
|
|
|
|
|
2019-05-15 21:45:01 +00:00
|
|
|
fill_origin_blob(&sb->revs->diffopt, parent, &file_p,
|
|
|
|
&sb->num_read_blob, ignore_diffs);
|
|
|
|
fill_origin_blob(&sb->revs->diffopt, target, &file_o,
|
|
|
|
&sb->num_read_blob, ignore_diffs);
|
2017-05-24 05:15:35 +00:00
|
|
|
sb->num_get_patch++;
|
|
|
|
|
|
|
|
if (diff_hunks(&file_p, &file_o, blame_chunk_cb, &d, sb->xdl_opts))
|
|
|
|
die("unable to generate diff (%s -> %s)",
|
|
|
|
oid_to_hex(&parent->commit->object.oid),
|
|
|
|
oid_to_hex(&target->commit->object.oid));
|
|
|
|
/* The rest are the same as the parent */
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
blame_chunk(&d.dstq, &d.srcq, INT_MAX, d.offset, INT_MAX, 0,
|
|
|
|
parent, target, 0);
|
2017-05-24 05:15:35 +00:00
|
|
|
*d.dstq = NULL;
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
if (ignore_diffs)
|
2022-07-16 16:58:20 +00:00
|
|
|
sort_blame_entries(&newdest, compare_blame_suspect);
|
2017-05-24 05:15:35 +00:00
|
|
|
queue_blames(sb, parent, newdest);
|
|
|
|
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The lines in blame_entry after splitting blames many times can become
|
|
|
|
* very small and trivial, and at some point it becomes pointless to
|
|
|
|
* blame the parents. E.g. "\t\t}\n\t}\n\n" appears everywhere in any
|
|
|
|
* ordinary C program, and it is not worth to say it was copied from
|
|
|
|
* totally unrelated file in the parent.
|
|
|
|
*
|
|
|
|
* Compute how trivial the lines in the blame_entry are.
|
|
|
|
*/
|
|
|
|
unsigned blame_entry_score(struct blame_scoreboard *sb, struct blame_entry *e)
|
|
|
|
{
|
|
|
|
unsigned score;
|
|
|
|
const char *cp, *ep;
|
|
|
|
|
|
|
|
if (e->score)
|
|
|
|
return e->score;
|
|
|
|
|
|
|
|
score = 1;
|
|
|
|
cp = blame_nth_line(sb, e->lno);
|
|
|
|
ep = blame_nth_line(sb, e->lno + e->num_lines);
|
|
|
|
while (cp < ep) {
|
|
|
|
unsigned ch = *((unsigned char *)cp);
|
|
|
|
if (isalnum(ch))
|
|
|
|
score++;
|
|
|
|
cp++;
|
|
|
|
}
|
|
|
|
e->score = score;
|
|
|
|
return score;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
2018-02-14 18:59:25 +00:00
|
|
|
* best_so_far[] and potential[] are both a split of an existing blame_entry
|
|
|
|
* that passes blame to the parent. Maintain best_so_far the best split so
|
|
|
|
* far, by comparing potential and best_so_far and copying potential into
|
2017-05-24 05:15:35 +00:00
|
|
|
* bst_so_far as needed.
|
|
|
|
*/
|
|
|
|
static void copy_split_if_better(struct blame_scoreboard *sb,
|
|
|
|
struct blame_entry *best_so_far,
|
2018-02-14 18:59:25 +00:00
|
|
|
struct blame_entry *potential)
|
2017-05-24 05:15:35 +00:00
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
2018-02-14 18:59:25 +00:00
|
|
|
if (!potential[1].suspect)
|
2017-05-24 05:15:35 +00:00
|
|
|
return;
|
|
|
|
if (best_so_far[1].suspect) {
|
2018-02-14 18:59:25 +00:00
|
|
|
if (blame_entry_score(sb, &potential[1]) <
|
|
|
|
blame_entry_score(sb, &best_so_far[1]))
|
2017-05-24 05:15:35 +00:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < 3; i++)
|
2018-02-14 18:59:25 +00:00
|
|
|
blame_origin_incref(potential[i].suspect);
|
2017-05-24 05:15:35 +00:00
|
|
|
decref_split(best_so_far);
|
2018-02-14 18:59:25 +00:00
|
|
|
memcpy(best_so_far, potential, sizeof(struct blame_entry[3]));
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We are looking at a part of the final image represented by
|
|
|
|
* ent (tlno and same are offset by ent->s_lno).
|
|
|
|
* tlno is where we are looking at in the final image.
|
|
|
|
* up to (but not including) same match preimage.
|
|
|
|
* plno is where we are looking at in the preimage.
|
|
|
|
*
|
|
|
|
* <-------------- final image ---------------------->
|
|
|
|
* <------ent------>
|
|
|
|
* ^tlno ^same
|
|
|
|
* <---------preimage----->
|
|
|
|
* ^plno
|
|
|
|
*
|
|
|
|
* All line numbers are 0-based.
|
|
|
|
*/
|
|
|
|
static void handle_split(struct blame_scoreboard *sb,
|
|
|
|
struct blame_entry *ent,
|
|
|
|
int tlno, int plno, int same,
|
|
|
|
struct blame_origin *parent,
|
|
|
|
struct blame_entry *split)
|
|
|
|
{
|
|
|
|
if (ent->num_lines <= tlno)
|
|
|
|
return;
|
|
|
|
if (tlno < same) {
|
2018-02-14 18:59:25 +00:00
|
|
|
struct blame_entry potential[3];
|
2017-05-24 05:15:35 +00:00
|
|
|
tlno += ent->s_lno;
|
|
|
|
same += ent->s_lno;
|
2018-02-14 18:59:25 +00:00
|
|
|
split_overlap(potential, ent, tlno, plno, same, parent);
|
|
|
|
copy_split_if_better(sb, split, potential);
|
|
|
|
decref_split(potential);
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
struct handle_split_cb_data {
|
|
|
|
struct blame_scoreboard *sb;
|
|
|
|
struct blame_entry *ent;
|
|
|
|
struct blame_origin *parent;
|
|
|
|
struct blame_entry *split;
|
|
|
|
long plno;
|
|
|
|
long tlno;
|
|
|
|
};
|
|
|
|
|
|
|
|
static int handle_split_cb(long start_a, long count_a,
|
|
|
|
long start_b, long count_b, void *data)
|
|
|
|
{
|
|
|
|
struct handle_split_cb_data *d = data;
|
|
|
|
handle_split(d->sb, d->ent, d->tlno, d->plno, start_b, d->parent,
|
|
|
|
d->split);
|
|
|
|
d->plno = start_a + count_a;
|
|
|
|
d->tlno = start_b + count_b;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Find the lines from parent that are the same as ent so that
|
|
|
|
* we can pass blames to it. file_p has the blob contents for
|
|
|
|
* the parent.
|
|
|
|
*/
|
|
|
|
static void find_copy_in_blob(struct blame_scoreboard *sb,
|
|
|
|
struct blame_entry *ent,
|
|
|
|
struct blame_origin *parent,
|
|
|
|
struct blame_entry *split,
|
|
|
|
mmfile_t *file_p)
|
|
|
|
{
|
|
|
|
const char *cp;
|
|
|
|
mmfile_t file_o;
|
|
|
|
struct handle_split_cb_data d;
|
|
|
|
|
|
|
|
memset(&d, 0, sizeof(d));
|
|
|
|
d.sb = sb; d.ent = ent; d.parent = parent; d.split = split;
|
|
|
|
/*
|
|
|
|
* Prepare mmfile that contains only the lines in ent.
|
|
|
|
*/
|
|
|
|
cp = blame_nth_line(sb, ent->lno);
|
|
|
|
file_o.ptr = (char *) cp;
|
|
|
|
file_o.size = blame_nth_line(sb, ent->lno + ent->num_lines) - cp;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* file_o is a part of final image we are annotating.
|
|
|
|
* file_p partially may match that image.
|
|
|
|
*/
|
|
|
|
memset(split, 0, sizeof(struct blame_entry [3]));
|
|
|
|
if (diff_hunks(file_p, &file_o, handle_split_cb, &d, sb->xdl_opts))
|
|
|
|
die("unable to generate diff (%s)",
|
|
|
|
oid_to_hex(&parent->commit->object.oid));
|
|
|
|
/* remainder, if any, all match the preimage */
|
|
|
|
handle_split(sb, ent, d.tlno, d.plno, ent->num_lines, parent, split);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Move all blame entries from list *source that have a score smaller
|
|
|
|
* than score_min to the front of list *small.
|
|
|
|
* Returns a pointer to the link pointing to the old head of the small list.
|
|
|
|
*/
|
|
|
|
|
|
|
|
static struct blame_entry **filter_small(struct blame_scoreboard *sb,
|
|
|
|
struct blame_entry **small,
|
|
|
|
struct blame_entry **source,
|
|
|
|
unsigned score_min)
|
|
|
|
{
|
|
|
|
struct blame_entry *p = *source;
|
|
|
|
struct blame_entry *oldsmall = *small;
|
|
|
|
while (p) {
|
|
|
|
if (blame_entry_score(sb, p) <= score_min) {
|
|
|
|
*small = p;
|
|
|
|
small = &p->next;
|
|
|
|
p = *small;
|
|
|
|
} else {
|
|
|
|
*source = p;
|
|
|
|
source = &p->next;
|
|
|
|
p = *source;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
*small = oldsmall;
|
|
|
|
*source = NULL;
|
|
|
|
return small;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* See if lines currently target is suspected for can be attributed to
|
|
|
|
* parent.
|
|
|
|
*/
|
|
|
|
static void find_move_in_parent(struct blame_scoreboard *sb,
|
|
|
|
struct blame_entry ***blamed,
|
|
|
|
struct blame_entry **toosmall,
|
|
|
|
struct blame_origin *target,
|
|
|
|
struct blame_origin *parent)
|
|
|
|
{
|
|
|
|
struct blame_entry *e, split[3];
|
|
|
|
struct blame_entry *unblamed = target->suspects;
|
|
|
|
struct blame_entry *leftover = NULL;
|
|
|
|
mmfile_t file_p;
|
|
|
|
|
|
|
|
if (!unblamed)
|
|
|
|
return; /* nothing remains for this target */
|
|
|
|
|
2019-05-15 21:45:01 +00:00
|
|
|
fill_origin_blob(&sb->revs->diffopt, parent, &file_p,
|
|
|
|
&sb->num_read_blob, 0);
|
2017-05-24 05:15:35 +00:00
|
|
|
if (!file_p.ptr)
|
|
|
|
return;
|
|
|
|
|
|
|
|
/* At each iteration, unblamed has a NULL-terminated list of
|
|
|
|
* entries that have not yet been tested for blame. leftover
|
|
|
|
* contains the reversed list of entries that have been tested
|
|
|
|
* without being assignable to the parent.
|
|
|
|
*/
|
|
|
|
do {
|
|
|
|
struct blame_entry **unblamedtail = &unblamed;
|
|
|
|
struct blame_entry *next;
|
|
|
|
for (e = unblamed; e; e = next) {
|
|
|
|
next = e->next;
|
|
|
|
find_copy_in_blob(sb, e, parent, split, &file_p);
|
|
|
|
if (split[1].suspect &&
|
|
|
|
sb->move_score < blame_entry_score(sb, &split[1])) {
|
|
|
|
split_blame(blamed, &unblamedtail, split, e);
|
|
|
|
} else {
|
|
|
|
e->next = leftover;
|
|
|
|
leftover = e;
|
|
|
|
}
|
|
|
|
decref_split(split);
|
|
|
|
}
|
|
|
|
*unblamedtail = NULL;
|
|
|
|
toosmall = filter_small(sb, toosmall, &unblamed, sb->move_score);
|
|
|
|
} while (unblamed);
|
|
|
|
target->suspects = reverse_blame(leftover, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
struct blame_list {
|
|
|
|
struct blame_entry *ent;
|
|
|
|
struct blame_entry split[3];
|
|
|
|
};
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Count the number of entries the target is suspected for,
|
|
|
|
* and prepare a list of entry and the best split.
|
|
|
|
*/
|
|
|
|
static struct blame_list *setup_blame_list(struct blame_entry *unblamed,
|
|
|
|
int *num_ents_p)
|
|
|
|
{
|
|
|
|
struct blame_entry *e;
|
|
|
|
int num_ents, i;
|
|
|
|
struct blame_list *blame_list = NULL;
|
|
|
|
|
|
|
|
for (e = unblamed, num_ents = 0; e; e = e->next)
|
|
|
|
num_ents++;
|
|
|
|
if (num_ents) {
|
2021-03-13 16:17:22 +00:00
|
|
|
CALLOC_ARRAY(blame_list, num_ents);
|
2017-05-24 05:15:35 +00:00
|
|
|
for (e = unblamed, i = 0; e; e = e->next)
|
|
|
|
blame_list[i++].ent = e;
|
|
|
|
}
|
|
|
|
*num_ents_p = num_ents;
|
|
|
|
return blame_list;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For lines target is suspected for, see if we can find code movement
|
|
|
|
* across file boundary from the parent commit. porigin is the path
|
|
|
|
* in the parent we already tried.
|
|
|
|
*/
|
|
|
|
static void find_copy_in_parent(struct blame_scoreboard *sb,
|
|
|
|
struct blame_entry ***blamed,
|
|
|
|
struct blame_entry **toosmall,
|
|
|
|
struct blame_origin *target,
|
|
|
|
struct commit *parent,
|
|
|
|
struct blame_origin *porigin,
|
|
|
|
int opt)
|
|
|
|
{
|
|
|
|
struct diff_options diff_opts;
|
|
|
|
int i, j;
|
|
|
|
struct blame_list *blame_list;
|
|
|
|
int num_ents;
|
|
|
|
struct blame_entry *unblamed = target->suspects;
|
|
|
|
struct blame_entry *leftover = NULL;
|
|
|
|
|
|
|
|
if (!unblamed)
|
|
|
|
return; /* nothing remains for this target */
|
|
|
|
|
2018-09-21 15:57:24 +00:00
|
|
|
repo_diff_setup(sb->repo, &diff_opts);
|
2017-10-31 18:19:11 +00:00
|
|
|
diff_opts.flags.recursive = 1;
|
2017-05-24 05:15:35 +00:00
|
|
|
diff_opts.output_format = DIFF_FORMAT_NO_OUTPUT;
|
|
|
|
|
|
|
|
diff_setup_done(&diff_opts);
|
|
|
|
|
|
|
|
/* Try "find copies harder" on new path if requested;
|
|
|
|
* we do not want to use diffcore_rename() actually to
|
|
|
|
* match things up; find_copies_harder is set only to
|
2017-06-19 19:38:44 +00:00
|
|
|
* force diff_tree_oid() to feed all filepairs to diff_queue,
|
2017-05-24 05:15:35 +00:00
|
|
|
* and this code needs to be after diff_setup_done(), which
|
|
|
|
* usually makes find-copies-harder imply copy detection.
|
|
|
|
*/
|
|
|
|
if ((opt & PICKAXE_BLAME_COPY_HARDEST)
|
|
|
|
|| ((opt & PICKAXE_BLAME_COPY_HARDER)
|
|
|
|
&& (!porigin || strcmp(target->path, porigin->path))))
|
2017-10-31 18:19:11 +00:00
|
|
|
diff_opts.flags.find_copies_harder = 1;
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
if (is_null_oid(&target->commit->object.oid))
|
2018-04-06 19:09:38 +00:00
|
|
|
do_diff_cache(get_commit_tree_oid(parent), &diff_opts);
|
2017-05-24 05:15:35 +00:00
|
|
|
else
|
2018-04-06 19:09:38 +00:00
|
|
|
diff_tree_oid(get_commit_tree_oid(parent),
|
|
|
|
get_commit_tree_oid(target->commit),
|
2017-06-19 19:38:44 +00:00
|
|
|
"", &diff_opts);
|
2017-05-24 05:15:35 +00:00
|
|
|
|
2017-10-31 18:19:11 +00:00
|
|
|
if (!diff_opts.flags.find_copies_harder)
|
2017-05-24 05:15:35 +00:00
|
|
|
diffcore_std(&diff_opts);
|
|
|
|
|
|
|
|
do {
|
|
|
|
struct blame_entry **unblamedtail = &unblamed;
|
|
|
|
blame_list = setup_blame_list(unblamed, &num_ents);
|
|
|
|
|
|
|
|
for (i = 0; i < diff_queued_diff.nr; i++) {
|
|
|
|
struct diff_filepair *p = diff_queued_diff.queue[i];
|
|
|
|
struct blame_origin *norigin;
|
|
|
|
mmfile_t file_p;
|
2018-02-14 18:59:25 +00:00
|
|
|
struct blame_entry potential[3];
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
if (!DIFF_FILE_VALID(p->one))
|
|
|
|
continue; /* does not exist in parent */
|
|
|
|
if (S_ISGITLINK(p->one->mode))
|
|
|
|
continue; /* ignore git links */
|
|
|
|
if (porigin && !strcmp(p->one->path, porigin->path))
|
|
|
|
/* find_move already dealt with this path */
|
|
|
|
continue;
|
|
|
|
|
|
|
|
norigin = get_origin(parent, p->one->path);
|
|
|
|
oidcpy(&norigin->blob_oid, &p->one->oid);
|
|
|
|
norigin->mode = p->one->mode;
|
2019-05-15 21:45:01 +00:00
|
|
|
fill_origin_blob(&sb->revs->diffopt, norigin, &file_p,
|
|
|
|
&sb->num_read_blob, 0);
|
2017-05-24 05:15:35 +00:00
|
|
|
if (!file_p.ptr)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
for (j = 0; j < num_ents; j++) {
|
|
|
|
find_copy_in_blob(sb, blame_list[j].ent,
|
2018-02-14 18:59:25 +00:00
|
|
|
norigin, potential, &file_p);
|
2017-05-24 05:15:35 +00:00
|
|
|
copy_split_if_better(sb, blame_list[j].split,
|
2018-02-14 18:59:25 +00:00
|
|
|
potential);
|
|
|
|
decref_split(potential);
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
blame_origin_decref(norigin);
|
|
|
|
}
|
|
|
|
|
|
|
|
for (j = 0; j < num_ents; j++) {
|
|
|
|
struct blame_entry *split = blame_list[j].split;
|
|
|
|
if (split[1].suspect &&
|
|
|
|
sb->copy_score < blame_entry_score(sb, &split[1])) {
|
|
|
|
split_blame(blamed, &unblamedtail, split,
|
|
|
|
blame_list[j].ent);
|
|
|
|
} else {
|
|
|
|
blame_list[j].ent->next = leftover;
|
|
|
|
leftover = blame_list[j].ent;
|
|
|
|
}
|
|
|
|
decref_split(split);
|
|
|
|
}
|
|
|
|
free(blame_list);
|
|
|
|
*unblamedtail = NULL;
|
|
|
|
toosmall = filter_small(sb, toosmall, &unblamed, sb->copy_score);
|
|
|
|
} while (unblamed);
|
|
|
|
target->suspects = reverse_blame(leftover, NULL);
|
|
|
|
diff_flush(&diff_opts);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The blobs of origin and porigin exactly match, so everything
|
|
|
|
* origin is suspected for can be blamed on the parent.
|
|
|
|
*/
|
|
|
|
static void pass_whole_blame(struct blame_scoreboard *sb,
|
|
|
|
struct blame_origin *origin, struct blame_origin *porigin)
|
|
|
|
{
|
|
|
|
struct blame_entry *e, *suspects;
|
|
|
|
|
|
|
|
if (!porigin->file.ptr && origin->file.ptr) {
|
|
|
|
/* Steal its file */
|
|
|
|
porigin->file = origin->file;
|
|
|
|
origin->file.ptr = NULL;
|
|
|
|
}
|
|
|
|
suspects = origin->suspects;
|
|
|
|
origin->suspects = NULL;
|
|
|
|
for (e = suspects; e; e = e->next) {
|
|
|
|
blame_origin_incref(porigin);
|
|
|
|
blame_origin_decref(e->suspect);
|
|
|
|
e->suspect = porigin;
|
|
|
|
}
|
|
|
|
queue_blames(sb, porigin, suspects);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We pass blame from the current commit to its parents. We keep saying
|
|
|
|
* "parent" (and "porigin"), but what we mean is to find scapegoat to
|
|
|
|
* exonerate ourselves.
|
|
|
|
*/
|
|
|
|
static struct commit_list *first_scapegoat(struct rev_info *revs, struct commit *commit,
|
|
|
|
int reverse)
|
|
|
|
{
|
|
|
|
if (!reverse) {
|
|
|
|
if (revs->first_parent_only &&
|
|
|
|
commit->parents &&
|
|
|
|
commit->parents->next) {
|
|
|
|
free_commit_list(commit->parents->next);
|
|
|
|
commit->parents->next = NULL;
|
|
|
|
}
|
|
|
|
return commit->parents;
|
|
|
|
}
|
|
|
|
return lookup_decoration(&revs->children, &commit->object);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int num_scapegoats(struct rev_info *revs, struct commit *commit, int reverse)
|
|
|
|
{
|
|
|
|
struct commit_list *l = first_scapegoat(revs, commit, reverse);
|
|
|
|
return commit_list_count(l);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Distribute collected unsorted blames to the respected sorted lists
|
|
|
|
* in the various origins.
|
|
|
|
*/
|
|
|
|
static void distribute_blame(struct blame_scoreboard *sb, struct blame_entry *blamed)
|
|
|
|
{
|
2022-07-16 16:58:20 +00:00
|
|
|
sort_blame_entries(&blamed, compare_blame_suspect);
|
2017-05-24 05:15:35 +00:00
|
|
|
while (blamed)
|
|
|
|
{
|
|
|
|
struct blame_origin *porigin = blamed->suspect;
|
|
|
|
struct blame_entry *suspects = NULL;
|
|
|
|
do {
|
|
|
|
struct blame_entry *next = blamed->next;
|
|
|
|
blamed->next = suspects;
|
|
|
|
suspects = blamed;
|
|
|
|
blamed = next;
|
|
|
|
} while (blamed && blamed->suspect == porigin);
|
|
|
|
suspects = reverse_blame(suspects, NULL);
|
|
|
|
queue_blames(sb, porigin, suspects);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
#define MAXSG 16
|
|
|
|
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
typedef struct blame_origin *(*blame_find_alg)(struct repository *,
|
|
|
|
struct commit *,
|
|
|
|
struct blame_origin *,
|
|
|
|
struct blame_bloom_data *);
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
static void pass_blame(struct blame_scoreboard *sb, struct blame_origin *origin, int opt)
|
|
|
|
{
|
|
|
|
struct rev_info *revs = sb->revs;
|
|
|
|
int i, pass, num_sg;
|
|
|
|
struct commit *commit = origin->commit;
|
|
|
|
struct commit_list *sg;
|
|
|
|
struct blame_origin *sg_buf[MAXSG];
|
|
|
|
struct blame_origin *porigin, **sg_origin = sg_buf;
|
|
|
|
struct blame_entry *toosmall = NULL;
|
|
|
|
struct blame_entry *blames, **blametail = &blames;
|
|
|
|
|
|
|
|
num_sg = num_scapegoats(revs, commit, sb->reverse);
|
|
|
|
if (!num_sg)
|
|
|
|
goto finish;
|
|
|
|
else if (num_sg < ARRAY_SIZE(sg_buf))
|
|
|
|
memset(sg_buf, 0, sizeof(sg_buf));
|
|
|
|
else
|
2021-03-13 16:17:22 +00:00
|
|
|
CALLOC_ARRAY(sg_origin, num_sg);
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* The first pass looks for unrenamed path to optimize for
|
|
|
|
* common cases, then we look for renames in the second pass.
|
|
|
|
*/
|
|
|
|
for (pass = 0; pass < 2 - sb->no_whole_file_rename; pass++) {
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
blame_find_alg find = pass ? find_rename : find_origin;
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
|
|
|
|
i < num_sg && sg;
|
|
|
|
sg = sg->next, i++) {
|
|
|
|
struct commit *p = sg->item;
|
|
|
|
int j, same;
|
|
|
|
|
|
|
|
if (sg_origin[i])
|
|
|
|
continue;
|
2023-03-28 13:58:48 +00:00
|
|
|
if (repo_parse_commit(the_repository, p))
|
2017-05-24 05:15:35 +00:00
|
|
|
continue;
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
porigin = find(sb->repo, p, origin, sb->bloom_data);
|
2017-05-24 05:15:35 +00:00
|
|
|
if (!porigin)
|
|
|
|
continue;
|
convert "oidcmp() == 0" to oideq()
Using the more restrictive oideq() should, in the long run,
give the compiler more opportunities to optimize these
callsites. For now, this conversion should be a complete
noop with respect to the generated code.
The result is also perhaps a little more readable, as it
avoids the "zero is equal" idiom. Since it's so prevalent in
C, I think seasoned programmers tend not to even notice it
anymore, but it can sometimes make for awkward double
negations (e.g., we can drop a few !!oidcmp() instances
here).
This patch was generated almost entirely by the included
coccinelle patch. This mechanical conversion should be
completely safe, because we check explicitly for cases where
oidcmp() is compared to 0, which is what oideq() is doing
under the hood. Note that we don't have to catch "!oidcmp()"
separately; coccinelle's standard isomorphisms make sure the
two are treated equivalently.
I say "almost" because I did hand-edit the coccinelle output
to fix up a few style violations (it mostly keeps the
original formatting, but sometimes unwraps long lines).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-08-28 21:22:40 +00:00
|
|
|
if (oideq(&porigin->blob_oid, &origin->blob_oid)) {
|
2017-05-24 05:15:35 +00:00
|
|
|
pass_whole_blame(sb, origin, porigin);
|
|
|
|
blame_origin_decref(porigin);
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
for (j = same = 0; j < i; j++)
|
|
|
|
if (sg_origin[j] &&
|
convert "oidcmp() == 0" to oideq()
Using the more restrictive oideq() should, in the long run,
give the compiler more opportunities to optimize these
callsites. For now, this conversion should be a complete
noop with respect to the generated code.
The result is also perhaps a little more readable, as it
avoids the "zero is equal" idiom. Since it's so prevalent in
C, I think seasoned programmers tend not to even notice it
anymore, but it can sometimes make for awkward double
negations (e.g., we can drop a few !!oidcmp() instances
here).
This patch was generated almost entirely by the included
coccinelle patch. This mechanical conversion should be
completely safe, because we check explicitly for cases where
oidcmp() is compared to 0, which is what oideq() is doing
under the hood. Note that we don't have to catch "!oidcmp()"
separately; coccinelle's standard isomorphisms make sure the
two are treated equivalently.
I say "almost" because I did hand-edit the coccinelle output
to fix up a few style violations (it mostly keeps the
original formatting, but sometimes unwraps long lines).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-08-28 21:22:40 +00:00
|
|
|
oideq(&sg_origin[j]->blob_oid, &porigin->blob_oid)) {
|
2017-05-24 05:15:35 +00:00
|
|
|
same = 1;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (!same)
|
|
|
|
sg_origin[i] = porigin;
|
|
|
|
else
|
|
|
|
blame_origin_decref(porigin);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
sb->num_commits++;
|
|
|
|
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
|
|
|
|
i < num_sg && sg;
|
|
|
|
sg = sg->next, i++) {
|
|
|
|
struct blame_origin *porigin = sg_origin[i];
|
|
|
|
if (!porigin)
|
|
|
|
continue;
|
|
|
|
if (!origin->previous) {
|
|
|
|
blame_origin_incref(porigin);
|
|
|
|
origin->previous = porigin;
|
|
|
|
}
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
pass_blame_to_parent(sb, origin, porigin, 0);
|
2017-05-24 05:15:35 +00:00
|
|
|
if (!origin->suspects)
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
/*
|
|
|
|
* Pass remaining suspects for ignored commits to their parents.
|
|
|
|
*/
|
|
|
|
if (oidset_contains(&sb->ignore_list, &commit->object.oid)) {
|
|
|
|
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
|
|
|
|
i < num_sg && sg;
|
|
|
|
sg = sg->next, i++) {
|
|
|
|
struct blame_origin *porigin = sg_origin[i];
|
|
|
|
|
|
|
|
if (!porigin)
|
|
|
|
continue;
|
|
|
|
pass_blame_to_parent(sb, origin, porigin, 1);
|
2019-06-20 16:38:19 +00:00
|
|
|
/*
|
|
|
|
* Preemptively drop porigin so we can refresh the
|
|
|
|
* fingerprints if we use the parent again, which can
|
|
|
|
* occur if you ignore back-to-back commits.
|
|
|
|
*/
|
|
|
|
drop_origin_blob(porigin);
|
blame: add the ability to ignore commits and their changes
Commits that make formatting changes or function renames are often not
interesting when blaming a file. A user may deem such a commit as 'not
interesting' and want to ignore and its changes it when assigning blame.
For example, say a file has the following git history / rev-list:
---O---A---X---B---C---D---Y---E---F
Commits X and Y both touch a particular line, and the other commits do
not:
X: "Take a third parameter"
-MyFunc(1, 2);
+MyFunc(1, 2, 3);
Y: "Remove camelcase"
-MyFunc(1, 2, 3);
+my_func(1, 2, 3);
git-blame will blame Y for the change. I'd like to be able to ignore Y:
both the existence of the commit as well as any changes it made. This
differs from -S rev-list, which specifies the list of commits to
process for the blame. We would still process Y, but just don't let the
blame 'stick.'
This patch adds the ability for users to ignore a revision with
--ignore-rev=rev, which may be repeated. They can specify a set of
files of full object names of revs, e.g. SHA-1 hashes, one per line. A
single file may be specified with the blame.ignoreRevFile config option
or with --ignore-rev-file=file. Both the config option and the command
line option may be repeated multiple times. An empty file name "" will
clear the list of revs from previously processed files. Config options
are processed before command line options.
For a typical use case, projects will maintain the file containing
revisions for commits that perform mass reformatting, and their users
have the option to ignore all of the commits in that file.
Additionally, a user can use the --ignore-rev option for one-off
investigation. To go back to the example above, X was a substantive
change to the function, but not the change the user is interested in.
The user inspected X, but wanted to find the previous change to that
line - perhaps a commit that introduced that function call.
To make this work, we can't simply remove all ignored commits from the
rev-list. We need to diff the changes introduced by Y so that we can
ignore them. We let the blames get passed to Y, just like when
processing normally. When Y is the target, we make sure that Y does not
*keep* any blames. Any changes that Y is responsible for get passed to
its parent. Note we make one pass through all of the scapegoats
(parents) to attempt to pass blame normally; we don't know if we *need*
to ignore the commit until we've checked all of the parents.
The blame_entry will get passed up the tree until we find a commit that
has a diff chunk that affects those lines.
One issue is that the ignored commit *did* make some change, and there is
no general solution to finding the line in the parent commit that
corresponds to a given line in the ignored commit. That makes it hard
to attribute a particular line within an ignored commit's diff
correctly.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) #include "a.h"
commit-b 12) #include "b.h"
Commit X, which we will ignore, swaps these lines:
commit-X 11) #include "b.h"
commit-X 12) #include "a.h"
We can pass that blame entry to the parent, but line 11 will be
attributed to commit A, even though "include b.h" came from commit B.
The blame mechanism will be looking at the parent's view of the file at
line number 11.
ignore_blame_entry() is set up to allow alternative algorithms for
guessing per-line blames. Any line that is not attributed to the parent
will continue to be blamed on the ignored commit as if that commit was
not ignored. Upcoming patches have the ability to detect these lines
and mark them in the blame output.
The existing algorithm is simple: blame each line on the corresponding
line in the parent's diff chunk. Any lines beyond that stay with the
target.
For example, the parent of an ignored commit has this, say at line 11:
commit-a 11) void new_func_1(void *x, void *y);
commit-b 12) void new_func_2(void *x, void *y);
commit-c 13) some_line_c
commit-d 14) some_line_d
After a commit 'X', we have:
commit-X 11) void new_func_1(void *x,
commit-X 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Commit X nets two additionally lines: 13 and 14. The current
guess_line_blames() algorithm will not attribute these to the parent,
whose diff chunk is only two lines - not four.
When we ignore with the current algorithm, we get:
commit-a 11) void new_func_1(void *x,
commit-b 12) void *y);
commit-X 13) void new_func_2(void *x,
commit-X 14) void *y);
commit-c 15) some_line_c
commit-d 16) some_line_d
Note that line 12 was blamed on B, though B was the commit for
new_func_2(), not new_func_1(). Even when guess_line_blames() finds a
line in the parent, it may still be incorrect.
Signed-off-by: Barret Rhoden <brho@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2019-05-15 21:44:59 +00:00
|
|
|
if (!origin->suspects)
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:35 +00:00
|
|
|
/*
|
|
|
|
* Optionally find moves in parents' files.
|
|
|
|
*/
|
|
|
|
if (opt & PICKAXE_BLAME_MOVE) {
|
|
|
|
filter_small(sb, &toosmall, &origin->suspects, sb->move_score);
|
|
|
|
if (origin->suspects) {
|
|
|
|
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
|
|
|
|
i < num_sg && sg;
|
|
|
|
sg = sg->next, i++) {
|
|
|
|
struct blame_origin *porigin = sg_origin[i];
|
|
|
|
if (!porigin)
|
|
|
|
continue;
|
|
|
|
find_move_in_parent(sb, &blametail, &toosmall, origin, porigin);
|
|
|
|
if (!origin->suspects)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Optionally find copies from parents' files.
|
|
|
|
*/
|
|
|
|
if (opt & PICKAXE_BLAME_COPY) {
|
|
|
|
if (sb->copy_score > sb->move_score)
|
|
|
|
filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
|
|
|
|
else if (sb->copy_score < sb->move_score) {
|
|
|
|
origin->suspects = blame_merge(origin->suspects, toosmall);
|
|
|
|
toosmall = NULL;
|
|
|
|
filter_small(sb, &toosmall, &origin->suspects, sb->copy_score);
|
|
|
|
}
|
|
|
|
if (!origin->suspects)
|
|
|
|
goto finish;
|
|
|
|
|
|
|
|
for (i = 0, sg = first_scapegoat(revs, commit, sb->reverse);
|
|
|
|
i < num_sg && sg;
|
|
|
|
sg = sg->next, i++) {
|
|
|
|
struct blame_origin *porigin = sg_origin[i];
|
|
|
|
find_copy_in_parent(sb, &blametail, &toosmall,
|
|
|
|
origin, sg->item, porigin, opt);
|
|
|
|
if (!origin->suspects)
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
finish:
|
|
|
|
*blametail = NULL;
|
|
|
|
distribute_blame(sb, blames);
|
|
|
|
/*
|
|
|
|
* prepend toosmall to origin->suspects
|
|
|
|
*
|
|
|
|
* There is no point in sorting: this ends up on a big
|
|
|
|
* unsorted list in the caller anyway.
|
|
|
|
*/
|
|
|
|
if (toosmall) {
|
|
|
|
struct blame_entry **tail = &toosmall;
|
|
|
|
while (*tail)
|
|
|
|
tail = &(*tail)->next;
|
|
|
|
*tail = origin->suspects;
|
|
|
|
origin->suspects = toosmall;
|
|
|
|
}
|
|
|
|
for (i = 0; i < num_sg; i++) {
|
|
|
|
if (sg_origin[i]) {
|
2019-04-02 11:56:25 +00:00
|
|
|
if (!sg_origin[i]->suspects)
|
|
|
|
drop_origin_blob(sg_origin[i]);
|
2017-05-24 05:15:35 +00:00
|
|
|
blame_origin_decref(sg_origin[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
drop_origin_blob(origin);
|
|
|
|
if (sg_buf != sg_origin)
|
|
|
|
free(sg_origin);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* The main loop -- while we have blobs with lines whose true origin
|
|
|
|
* is still unknown, pick one blob, and allow its lines to pass blames
|
|
|
|
* to its parents. */
|
|
|
|
void assign_blame(struct blame_scoreboard *sb, int opt)
|
|
|
|
{
|
|
|
|
struct rev_info *revs = sb->revs;
|
|
|
|
struct commit *commit = prio_queue_get(&sb->commits);
|
|
|
|
|
|
|
|
while (commit) {
|
|
|
|
struct blame_entry *ent;
|
2018-05-19 05:28:19 +00:00
|
|
|
struct blame_origin *suspect = get_blame_suspects(commit);
|
2017-05-24 05:15:35 +00:00
|
|
|
|
|
|
|
/* find one suspect to break down */
|
|
|
|
while (suspect && !suspect->suspects)
|
|
|
|
suspect = suspect->next;
|
|
|
|
|
|
|
|
if (!suspect) {
|
|
|
|
commit = prio_queue_get(&sb->commits);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(commit == suspect->commit);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We will use this suspect later in the loop,
|
|
|
|
* so hold onto it in the meantime.
|
|
|
|
*/
|
|
|
|
blame_origin_incref(suspect);
|
2023-03-28 13:58:48 +00:00
|
|
|
repo_parse_commit(the_repository, commit);
|
2017-05-24 05:15:35 +00:00
|
|
|
if (sb->reverse ||
|
|
|
|
(!(commit->object.flags & UNINTERESTING) &&
|
|
|
|
!(revs->max_age != -1 && commit->date < revs->max_age)))
|
|
|
|
pass_blame(sb, suspect, opt);
|
|
|
|
else {
|
|
|
|
commit->object.flags |= UNINTERESTING;
|
|
|
|
if (commit->object.parsed)
|
git-rev-list: add --exclude-first-parent-only flag
It is useful to know when a branch first diverged in history
from some integration branch in order to be able to enumerate
the user's local changes. However, these local changes can
include arbitrary merges, so it is necessary to ignore this
merge structure when finding the divergence point.
In order to do this, teach the "rev-list" family to accept
"--exclude-first-parent-only", which restricts the traversal
of excluded commits to only follow first parent links.
-A-----E-F-G--main
\ / /
B-C-D--topic
In this example, the goal is to return the set {B, C, D} which
represents a topic branch that has been merged into main branch.
`git rev-list topic ^main` will end up returning no commits
since excluding main will end up traversing the commits on topic
as well. `git rev-list --exclude-first-parent-only topic ^main`
however will return {B, C, D} as desired.
Add docs for the new flag, and clarify the doc for --first-parent
to indicate that it applies to traversing the set of included
commits only.
Signed-off-by: Jerry Zhang <jerry@skydio.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2022-01-11 21:39:41 +00:00
|
|
|
mark_parents_uninteresting(sb->revs, commit);
|
2017-05-24 05:15:35 +00:00
|
|
|
}
|
|
|
|
/* treat root commit as boundary */
|
|
|
|
if (!commit->parents && !sb->show_root)
|
|
|
|
commit->object.flags |= UNINTERESTING;
|
|
|
|
|
|
|
|
/* Take responsibility for the remaining entries */
|
|
|
|
ent = suspect->suspects;
|
|
|
|
if (ent) {
|
|
|
|
suspect->guilty = 1;
|
|
|
|
for (;;) {
|
|
|
|
struct blame_entry *next = ent->next;
|
|
|
|
if (sb->found_guilty_entry)
|
|
|
|
sb->found_guilty_entry(ent, sb->found_guilty_entry_data);
|
|
|
|
if (next) {
|
|
|
|
ent = next;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
ent->next = sb->ent;
|
|
|
|
sb->ent = suspect->suspects;
|
|
|
|
suspect->suspects = NULL;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
blame_origin_decref(suspect);
|
|
|
|
|
|
|
|
if (sb->debug) /* sanity */
|
|
|
|
sanity_check_refcnt(sb);
|
|
|
|
}
|
|
|
|
}
|
2017-05-24 05:15:36 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* To allow quick access to the contents of nth line in the
|
|
|
|
* final image, prepare an index in the scoreboard.
|
|
|
|
*/
|
|
|
|
static int prepare_lines(struct blame_scoreboard *sb)
|
|
|
|
{
|
2019-05-15 21:45:01 +00:00
|
|
|
sb->num_lines = find_line_starts(&sb->lineno, sb->final_buf,
|
|
|
|
sb->final_buf_size);
|
2017-05-24 05:15:36 +00:00
|
|
|
return sb->num_lines;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit *find_single_final(struct rev_info *revs,
|
|
|
|
const char **name_p)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
struct commit *found = NULL;
|
|
|
|
const char *name = NULL;
|
|
|
|
|
|
|
|
for (i = 0; i < revs->pending.nr; i++) {
|
|
|
|
struct object *obj = revs->pending.objects[i].item;
|
|
|
|
if (obj->flags & UNINTERESTING)
|
|
|
|
continue;
|
2018-11-10 05:48:58 +00:00
|
|
|
obj = deref_tag(revs->repo, obj, NULL, 0);
|
2020-10-11 16:03:37 +00:00
|
|
|
if (!obj || obj->type != OBJ_COMMIT)
|
2017-05-24 05:15:36 +00:00
|
|
|
die("Non commit %s?", revs->pending.objects[i].name);
|
|
|
|
if (found)
|
|
|
|
die("More than one commit to dig from %s and %s?",
|
|
|
|
revs->pending.objects[i].name, name);
|
|
|
|
found = (struct commit *)obj;
|
|
|
|
name = revs->pending.objects[i].name;
|
|
|
|
}
|
|
|
|
if (name_p)
|
2017-07-24 21:15:50 +00:00
|
|
|
*name_p = xstrdup_or_null(name);
|
2017-05-24 05:15:36 +00:00
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit *dwim_reverse_initial(struct rev_info *revs,
|
|
|
|
const char **name_p)
|
|
|
|
{
|
|
|
|
/*
|
|
|
|
* DWIM "git blame --reverse ONE -- PATH" as
|
|
|
|
* "git blame --reverse ONE..HEAD -- PATH" but only do so
|
|
|
|
* when it makes sense.
|
|
|
|
*/
|
|
|
|
struct object *obj;
|
|
|
|
struct commit *head_commit;
|
2017-06-05 00:18:11 +00:00
|
|
|
struct object_id head_oid;
|
2017-05-24 05:15:36 +00:00
|
|
|
|
|
|
|
if (revs->pending.nr != 1)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Is that sole rev a committish? */
|
|
|
|
obj = revs->pending.objects[0].item;
|
2018-11-10 05:48:58 +00:00
|
|
|
obj = deref_tag(revs->repo, obj, NULL, 0);
|
2020-10-11 16:03:37 +00:00
|
|
|
if (!obj || obj->type != OBJ_COMMIT)
|
2017-05-24 05:15:36 +00:00
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Do we have HEAD? */
|
refs: convert resolve_ref_unsafe to struct object_id
Convert resolve_ref_unsafe to take a pointer to struct object_id by
converting one remaining caller to use struct object_id, removing the
temporary NULL pointer check in expand_ref, converting the declaration
and definition, and applying the following semantic patch:
@@
expression E1, E2, E3, E4;
@@
- resolve_ref_unsafe(E1, E2, E3.hash, E4)
+ resolve_ref_unsafe(E1, E2, &E3, E4)
@@
expression E1, E2, E3, E4;
@@
- resolve_ref_unsafe(E1, E2, E3->hash, E4)
+ resolve_ref_unsafe(E1, E2, E3, E4)
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-10-15 22:07:09 +00:00
|
|
|
if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
|
2017-05-24 05:15:36 +00:00
|
|
|
return NULL;
|
2018-11-10 05:48:58 +00:00
|
|
|
head_commit = lookup_commit_reference_gently(revs->repo,
|
2018-06-29 01:21:57 +00:00
|
|
|
&head_oid, 1);
|
2017-05-24 05:15:36 +00:00
|
|
|
if (!head_commit)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Turn "ONE" into "ONE..HEAD" then */
|
|
|
|
obj->flags |= UNINTERESTING;
|
|
|
|
add_pending_object(revs, &head_commit->object, "HEAD");
|
|
|
|
|
|
|
|
if (name_p)
|
|
|
|
*name_p = revs->pending.objects[0].name;
|
|
|
|
return (struct commit *)obj;
|
|
|
|
}
|
|
|
|
|
|
|
|
static struct commit *find_single_initial(struct rev_info *revs,
|
|
|
|
const char **name_p)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
struct commit *found = NULL;
|
|
|
|
const char *name = NULL;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* There must be one and only one negative commit, and it must be
|
|
|
|
* the boundary.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < revs->pending.nr; i++) {
|
|
|
|
struct object *obj = revs->pending.objects[i].item;
|
|
|
|
if (!(obj->flags & UNINTERESTING))
|
|
|
|
continue;
|
2018-11-10 05:48:58 +00:00
|
|
|
obj = deref_tag(revs->repo, obj, NULL, 0);
|
2020-10-11 16:03:37 +00:00
|
|
|
if (!obj || obj->type != OBJ_COMMIT)
|
2017-05-24 05:15:36 +00:00
|
|
|
die("Non commit %s?", revs->pending.objects[i].name);
|
|
|
|
if (found)
|
|
|
|
die("More than one commit to dig up from, %s and %s?",
|
|
|
|
revs->pending.objects[i].name, name);
|
|
|
|
found = (struct commit *) obj;
|
|
|
|
name = revs->pending.objects[i].name;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!name)
|
|
|
|
found = dwim_reverse_initial(revs, &name);
|
|
|
|
if (!name)
|
|
|
|
die("No commit to dig up from?");
|
|
|
|
|
|
|
|
if (name_p)
|
2017-07-24 21:15:50 +00:00
|
|
|
*name_p = xstrdup(name);
|
2017-05-24 05:15:36 +00:00
|
|
|
return found;
|
|
|
|
}
|
|
|
|
|
|
|
|
void init_scoreboard(struct blame_scoreboard *sb)
|
|
|
|
{
|
|
|
|
memset(sb, 0, sizeof(struct blame_scoreboard));
|
|
|
|
sb->move_score = BLAME_DEFAULT_MOVE_SCORE;
|
|
|
|
sb->copy_score = BLAME_DEFAULT_COPY_SCORE;
|
|
|
|
}
|
|
|
|
|
2018-08-13 16:14:41 +00:00
|
|
|
void setup_scoreboard(struct blame_scoreboard *sb,
|
|
|
|
struct blame_origin **orig)
|
2017-05-24 05:15:36 +00:00
|
|
|
{
|
|
|
|
const char *final_commit_name = NULL;
|
|
|
|
struct blame_origin *o;
|
|
|
|
struct commit *final_commit = NULL;
|
|
|
|
enum object_type type;
|
|
|
|
|
2018-05-19 05:28:19 +00:00
|
|
|
init_blame_suspects(&blame_suspects);
|
|
|
|
|
2017-05-24 05:15:36 +00:00
|
|
|
if (sb->reverse && sb->contents_from)
|
|
|
|
die(_("--contents and --reverse do not blend well."));
|
|
|
|
|
2018-08-13 16:14:41 +00:00
|
|
|
if (!sb->repo)
|
|
|
|
BUG("repo is NULL");
|
|
|
|
|
2017-05-24 05:15:36 +00:00
|
|
|
if (!sb->reverse) {
|
|
|
|
sb->final = find_single_final(sb->revs, &final_commit_name);
|
|
|
|
sb->commits.compare = compare_commits_by_commit_date;
|
|
|
|
} else {
|
|
|
|
sb->final = find_single_initial(sb->revs, &final_commit_name);
|
|
|
|
sb->commits.compare = compare_commits_by_reverse_commit_date;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sb->reverse && sb->revs->first_parent_only)
|
|
|
|
sb->revs->children.name = NULL;
|
|
|
|
|
2023-03-24 17:08:00 +00:00
|
|
|
if (sb->contents_from || !sb->final) {
|
|
|
|
struct object_id head_oid, *parent_oid;
|
|
|
|
|
2017-05-24 05:15:36 +00:00
|
|
|
/*
|
2023-03-24 17:08:00 +00:00
|
|
|
* Build a fake commit at the top of the history, when
|
|
|
|
* (1) "git blame [^A] --path", i.e. with no positive end
|
|
|
|
* of the history range, in which case we build such
|
|
|
|
* a fake commit on top of the HEAD to blame in-tree
|
|
|
|
* modifications.
|
|
|
|
* (2) "git blame --contents=file [A] -- path", with or
|
|
|
|
* without positive end of the history range but with
|
|
|
|
* --contents, in which case we pretend that there is
|
|
|
|
* a fake commit on top of the positive end (defaulting to
|
|
|
|
* HEAD) that has the given contents in the path.
|
2017-05-24 05:15:36 +00:00
|
|
|
*/
|
2023-03-24 17:08:00 +00:00
|
|
|
if (sb->final) {
|
|
|
|
parent_oid = &sb->final->object.oid;
|
|
|
|
} else {
|
|
|
|
if (!resolve_ref_unsafe("HEAD", RESOLVE_REF_READING, &head_oid, NULL))
|
|
|
|
die("no such ref: HEAD");
|
|
|
|
parent_oid = &head_oid;
|
|
|
|
}
|
|
|
|
|
2017-05-24 05:15:36 +00:00
|
|
|
setup_work_tree();
|
2018-08-13 16:14:41 +00:00
|
|
|
sb->final = fake_working_tree_commit(sb->repo,
|
|
|
|
&sb->revs->diffopt,
|
2023-03-24 17:08:00 +00:00
|
|
|
sb->path, sb->contents_from,
|
|
|
|
parent_oid);
|
2017-05-24 05:15:36 +00:00
|
|
|
add_pending_object(sb->revs, &(sb->final->object), ":");
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sb->reverse && sb->revs->first_parent_only) {
|
|
|
|
final_commit = find_single_final(sb->revs, NULL);
|
|
|
|
if (!final_commit)
|
|
|
|
die(_("--reverse and --first-parent together require specified latest commit"));
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If we have bottom, this will mark the ancestors of the
|
|
|
|
* bottom commits we would reach while traversing as
|
|
|
|
* uninteresting.
|
|
|
|
*/
|
|
|
|
if (prepare_revision_walk(sb->revs))
|
|
|
|
die(_("revision walk setup failed"));
|
|
|
|
|
|
|
|
if (sb->reverse && sb->revs->first_parent_only) {
|
|
|
|
struct commit *c = final_commit;
|
|
|
|
|
|
|
|
sb->revs->children.name = "children";
|
|
|
|
while (c->parents &&
|
2018-08-28 21:22:48 +00:00
|
|
|
!oideq(&c->object.oid, &sb->final->object.oid)) {
|
2017-05-24 05:15:36 +00:00
|
|
|
struct commit_list *l = xcalloc(1, sizeof(*l));
|
|
|
|
|
|
|
|
l->item = c;
|
|
|
|
if (add_decoration(&sb->revs->children,
|
|
|
|
&c->parents->item->object, l))
|
2018-05-02 09:38:39 +00:00
|
|
|
BUG("not unique item in first-parent chain");
|
2017-05-24 05:15:36 +00:00
|
|
|
c = c->parents->item;
|
|
|
|
}
|
|
|
|
|
2018-08-28 21:22:48 +00:00
|
|
|
if (!oideq(&c->object.oid, &sb->final->object.oid))
|
2017-05-24 05:15:36 +00:00
|
|
|
die(_("--reverse --first-parent together require range along first-parent chain"));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (is_null_oid(&sb->final->object.oid)) {
|
2018-05-19 05:28:19 +00:00
|
|
|
o = get_blame_suspects(sb->final);
|
2017-05-24 05:15:36 +00:00
|
|
|
sb->final_buf = xmemdupz(o->file.ptr, o->file.size);
|
|
|
|
sb->final_buf_size = o->file.size;
|
|
|
|
}
|
|
|
|
else {
|
2020-11-01 17:28:46 +00:00
|
|
|
o = get_origin(sb->final, sb->path);
|
2018-08-13 16:14:41 +00:00
|
|
|
if (fill_blob_sha1_and_mode(sb->repo, o))
|
2020-11-01 17:28:46 +00:00
|
|
|
die(_("no such path %s in %s"), sb->path, final_commit_name);
|
2017-05-24 05:15:36 +00:00
|
|
|
|
2017-10-31 18:19:11 +00:00
|
|
|
if (sb->revs->diffopt.flags.allow_textconv &&
|
2020-11-01 17:28:46 +00:00
|
|
|
textconv_object(sb->repo, sb->path, o->mode, &o->blob_oid, 1, (char **) &sb->final_buf,
|
2017-05-24 05:15:36 +00:00
|
|
|
&sb->final_buf_size))
|
|
|
|
;
|
|
|
|
else
|
2023-03-28 13:58:50 +00:00
|
|
|
sb->final_buf = repo_read_object_file(the_repository,
|
|
|
|
&o->blob_oid,
|
|
|
|
&type,
|
|
|
|
&sb->final_buf_size);
|
2017-05-24 05:15:36 +00:00
|
|
|
|
|
|
|
if (!sb->final_buf)
|
|
|
|
die(_("cannot read blob %s for path %s"),
|
|
|
|
oid_to_hex(&o->blob_oid),
|
2020-11-01 17:28:46 +00:00
|
|
|
sb->path);
|
2017-05-24 05:15:36 +00:00
|
|
|
}
|
|
|
|
sb->num_read_blob++;
|
|
|
|
prepare_lines(sb);
|
|
|
|
|
|
|
|
if (orig)
|
|
|
|
*orig = o;
|
2017-07-24 21:15:50 +00:00
|
|
|
|
|
|
|
free((char *)final_commit_name);
|
2017-05-24 05:15:36 +00:00
|
|
|
}
|
2017-05-24 05:15:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
struct blame_entry *blame_entry_prepend(struct blame_entry *head,
|
|
|
|
long start, long end,
|
|
|
|
struct blame_origin *o)
|
|
|
|
{
|
|
|
|
struct blame_entry *new_head = xcalloc(1, sizeof(struct blame_entry));
|
|
|
|
new_head->lno = start;
|
|
|
|
new_head->num_lines = end - start;
|
|
|
|
new_head->suspect = o;
|
|
|
|
new_head->s_lno = start;
|
|
|
|
new_head->next = head;
|
|
|
|
blame_origin_incref(o);
|
|
|
|
return new_head;
|
|
|
|
}
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
|
2020-11-01 17:28:47 +00:00
|
|
|
void setup_blame_bloom_data(struct blame_scoreboard *sb)
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
{
|
|
|
|
struct blame_bloom_data *bd;
|
commit-graph: introduce 'get_bloom_filter_settings()'
Many places in the code often need a pointer to the commit-graph's
'struct bloom_filter_settings', in which case they often take the value
from the top-most commit-graph.
In the non-split case, this works as expected. In the split case,
however, things get a little tricky. Not all layers in a chain of
incremental commit-graphs are required to themselves have Bloom data,
and so whether or not some part of the code uses Bloom filters depends
entirely on whether or not the top-most level of the commit-graph chain
has Bloom filters.
This has been the behavior since Bloom filters were introduced, and has
been codified into the tests since a759bfa9ee (t4216: add end to end
tests for git log with Bloom filters, 2020-04-06). In fact, t4216.130
requires that Bloom filters are not used in exactly the case described
earlier.
There is no reason that this needs to be the case, since it is perfectly
valid for commits in an earlier layer to have Bloom filters when commits
in a newer layer do not.
Since Bloom settings are guaranteed in practice to be the same for any
layer in a chain that has Bloom data, it is sufficient to traverse the
'->base_graph' pointer until either (1) a non-null 'struct
bloom_filter_settings *' is found, or (2) until we are at the root of
the commit-graph chain.
Introduce a 'get_bloom_filter_settings()' function that does just this,
and use it instead of purely dereferencing the top-most graph's
'->bloom_filter_settings' pointer.
While we're at it, add an additional test in t5324 to guard against code
in the commit-graph writing machinery that doesn't correctly handle a
NULL 'struct bloom_filter *'.
Co-authored-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-09 15:22:44 +00:00
|
|
|
struct bloom_filter_settings *bs;
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
|
|
|
|
if (!sb->repo->objects->commit_graph)
|
|
|
|
return;
|
|
|
|
|
commit-graph: introduce 'get_bloom_filter_settings()'
Many places in the code often need a pointer to the commit-graph's
'struct bloom_filter_settings', in which case they often take the value
from the top-most commit-graph.
In the non-split case, this works as expected. In the split case,
however, things get a little tricky. Not all layers in a chain of
incremental commit-graphs are required to themselves have Bloom data,
and so whether or not some part of the code uses Bloom filters depends
entirely on whether or not the top-most level of the commit-graph chain
has Bloom filters.
This has been the behavior since Bloom filters were introduced, and has
been codified into the tests since a759bfa9ee (t4216: add end to end
tests for git log with Bloom filters, 2020-04-06). In fact, t4216.130
requires that Bloom filters are not used in exactly the case described
earlier.
There is no reason that this needs to be the case, since it is perfectly
valid for commits in an earlier layer to have Bloom filters when commits
in a newer layer do not.
Since Bloom settings are guaranteed in practice to be the same for any
layer in a chain that has Bloom data, it is sufficient to traverse the
'->base_graph' pointer until either (1) a non-null 'struct
bloom_filter_settings *' is found, or (2) until we are at the root of
the commit-graph chain.
Introduce a 'get_bloom_filter_settings()' function that does just this,
and use it instead of purely dereferencing the top-most graph's
'->bloom_filter_settings' pointer.
While we're at it, add an additional test in t5324 to guard against code
in the commit-graph writing machinery that doesn't correctly handle a
NULL 'struct bloom_filter *'.
Co-authored-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-09 15:22:44 +00:00
|
|
|
bs = get_bloom_filter_settings(sb->repo);
|
|
|
|
if (!bs)
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
return;
|
|
|
|
|
|
|
|
bd = xmalloc(sizeof(struct blame_bloom_data));
|
|
|
|
|
commit-graph: introduce 'get_bloom_filter_settings()'
Many places in the code often need a pointer to the commit-graph's
'struct bloom_filter_settings', in which case they often take the value
from the top-most commit-graph.
In the non-split case, this works as expected. In the split case,
however, things get a little tricky. Not all layers in a chain of
incremental commit-graphs are required to themselves have Bloom data,
and so whether or not some part of the code uses Bloom filters depends
entirely on whether or not the top-most level of the commit-graph chain
has Bloom filters.
This has been the behavior since Bloom filters were introduced, and has
been codified into the tests since a759bfa9ee (t4216: add end to end
tests for git log with Bloom filters, 2020-04-06). In fact, t4216.130
requires that Bloom filters are not used in exactly the case described
earlier.
There is no reason that this needs to be the case, since it is perfectly
valid for commits in an earlier layer to have Bloom filters when commits
in a newer layer do not.
Since Bloom settings are guaranteed in practice to be the same for any
layer in a chain that has Bloom data, it is sufficient to traverse the
'->base_graph' pointer until either (1) a non-null 'struct
bloom_filter_settings *' is found, or (2) until we are at the root of
the commit-graph chain.
Introduce a 'get_bloom_filter_settings()' function that does just this,
and use it instead of purely dereferencing the top-most graph's
'->bloom_filter_settings' pointer.
While we're at it, add an additional test in t5324 to guard against code
in the commit-graph writing machinery that doesn't correctly handle a
NULL 'struct bloom_filter *'.
Co-authored-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-09-09 15:22:44 +00:00
|
|
|
bd->settings = bs;
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
|
|
|
|
bd->alloc = 4;
|
|
|
|
bd->nr = 0;
|
|
|
|
ALLOC_ARRAY(bd->keys, bd->alloc);
|
|
|
|
|
2020-11-01 17:28:47 +00:00
|
|
|
add_bloom_key(bd, sb->path);
|
blame: use changed-path Bloom filters
The changed-path Bloom filters help reduce the amount of tree
parsing required during history queries. Before calculating a
diff, we can ask the filter if a path changed between a commit
and its first parent. If the filter says "no" then we can move
on without parsing trees. If the filter says "maybe" then we
parse trees to discover if the answer is actually "yes" or "no".
When computing a blame, there is a section in find_origin() that
computes a diff between a commit and one of its parents. When this
is the first parent, we can check the Bloom filters before calling
diff_tree_oid().
In order to make this work with the blame machinery, we need to
initialize a struct bloom_key with the initial path. But also, we
need to add more keys to a list if a rename is detected. We then
check to see if _any_ of these keys answer "maybe" in the diff.
During development, I purposefully left out this "add a new key
when a rename is detected" to see if the test suite would catch
my error. That is how I discovered the issues with
GIT_TEST_COMMIT_GRAPH_CHANGED_PATHS from the previous change.
With that change, we can feel some confidence in the coverage of
this change.
If a user requests copy detection using "git blame -C", then there
are more places where the set of "important" files can expand. I
do not know enough about how this happens in the blame machinery.
Thus, the Bloom filter integration is explicitly disabled in this
mode. A later change could expand the bloom_key data with an
appropriate call (or calls) to add_bloom_key().
If we did not disable this mode, then the following tests would
fail:
t8003-blame-corner-cases.sh
t8011-blame-split-file.sh
Generally, this is a performance enhancement and should not
change the behavior of 'git blame' in any way. If a repo has a
commit-graph file with computed changed-path Bloom filters, then
they should notice improved performance for their 'git blame'
commands.
Here are some example timings that I found by blaming some paths
in the Linux kernel repository:
git blame arch/x86/kernel/topology.c >/dev/null
Before: 0.83s
After: 0.24s
git blame kernel/time/time.c >/dev/null
Before: 0.72s
After: 0.24s
git blame tools/perf/ui/stdio/hist.c >/dev/null
Before: 0.27s
After: 0.11s
I specifically looked for "deep" paths that were also edited many
times. As a counterpoint, the MAINTAINERS file was edited many
times but is located in the root tree. This means that the cost of
computing a diff relative to the pathspec is very small. Here are
the timings for that command:
git blame MAINTAINERS >/dev/null
Before: 20.1s
After: 18.0s
These timings are the best of five. The worst-case runs were on the
order of 2.5 minutes for both cases. Note that the MAINTAINERS file
has 18,740 lines across 17,000+ commits. This happens to be one of
the cases where this change provides the least improvement.
The lack of improvement for the MAINTAINERS file and the relatively
modest improvement for the other examples can be easily explained.
The blame machinery needs to compute line-level diffs to determine
which lines were changed by each commit. That makes up a large
proportion of the computation time, and this change does not
attempt to improve on that section of the algorithm. The
MAINTAINERS file is large and changed often, so it takes time to
determine which lines were updated by which commit. In contrast,
the code files are much smaller, and it takes longer to comute
the line-by-line diff for a single patch on the Linux mailing
lists.
Outside of the "-C" integration, I believe there is little more to
gain from the changed-path Bloom filters for 'git blame' after this
patch.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2020-04-16 20:14:04 +00:00
|
|
|
|
|
|
|
sb->bloom_data = bd;
|
|
|
|
}
|
|
|
|
|
|
|
|
void cleanup_scoreboard(struct blame_scoreboard *sb)
|
|
|
|
{
|
|
|
|
if (sb->bloom_data) {
|
|
|
|
int i;
|
|
|
|
for (i = 0; i < sb->bloom_data->nr; i++) {
|
|
|
|
free(sb->bloom_data->keys[i]->hashes);
|
|
|
|
free(sb->bloom_data->keys[i]);
|
|
|
|
}
|
|
|
|
free(sb->bloom_data->keys);
|
|
|
|
FREE_AND_NULL(sb->bloom_data);
|
|
|
|
|
|
|
|
trace2_data_intmax("blame", sb->repo,
|
|
|
|
"bloom/queries", bloom_count_queries);
|
|
|
|
trace2_data_intmax("blame", sb->repo,
|
|
|
|
"bloom/response-no", bloom_count_no);
|
|
|
|
}
|
|
|
|
}
|