git/rerere.c

1254 lines
32 KiB
C
Raw Normal View History

#include "git-compat-util.h"
#include "abspath.h"
#include "config.h"
#include "copy.h"
#include "gettext.h"
#include "hex.h"
#include "lockfile.h"
#include "string-list.h"
#include "read-cache-ll.h"
#include "rerere.h"
#include "xdiff-interface.h"
#include "dir.h"
#include "resolve-undo.h"
merge-ll: rename from ll-merge A long term (but rather minor) pet-peeve of mine was the name ll-merge.[ch]. I thought it made it harder to realize what stuff was related to merging when I was working on the merge machinery and trying to improve it. Further, back in d1cbe1e6d8a ("hash-ll.h: split out of hash.h to remove dependency on repository.h", 2023-04-22), we have split the portions of hash.h that do not depend upon repository.h into a "hash-ll.h" (due to the recommendation to use "ll" for "low-level" in its name[1], but which I used as a suffix precisely because of my distaste for "ll-merge"). When we discussed adding additional "*-ll.h" files, a request was made that we use "ll" consistently as either a prefix or a suffix. Since it is already in use as both a prefix and a suffix, the only way to do so is to rename some files. Besides my distaste for the ll-merge.[ch] name, let me also note that the files ll-fsmonitor.h, ll-hash.h, ll-merge.h, ll-object-store.h, ll-read-cache.h would have essentially nothing to do with each other and make no sense to group. But giving them the common "ll-" prefix would group them. Using "-ll" as a suffix thus seems just much more logical to me. Rename ll-merge.[ch] to merge-ll.[ch] to achieve this consistency, and to ensure we get a more logical grouping of files. [1] https://lore.kernel.org/git/kl6lsfcu1g8w.fsf@chooglen-macbookpro.roam.corp.google.com/ Signed-off-by: Elijah Newren <newren@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-05-16 06:34:04 +00:00
#include "merge-ll.h"
#include "path.h"
#include "pathspec.h"
#include "object-file.h"
#include "object-store-ll.h"
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
#include "strmap.h"
#define RESOLVED 0
#define PUNTED 1
#define THREE_STAGED 2
void *RERERE_RESOLVED = &RERERE_RESOLVED;
/* if rerere_enabled == -1, fall back to detection of .git/rr-cache */
static int rerere_enabled = -1;
/* automatically update cleanly resolved paths to the index */
static int rerere_autoupdate;
#define RR_HAS_POSTIMAGE 1
#define RR_HAS_PREIMAGE 2
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
struct rerere_dir {
int status_alloc, status_nr;
unsigned char *status;
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
char name[FLEX_ARRAY];
};
static struct strmap rerere_dirs = STRMAP_INIT;
static void free_rerere_dirs(void)
{
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
struct hashmap_iter iter;
struct strmap_entry *ent;
strmap_for_each_entry(&rerere_dirs, &iter, ent) {
struct rerere_dir *rr_dir = ent->value;
free(rr_dir->status);
free(rr_dir);
}
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
strmap_clear(&rerere_dirs, 0);
}
static void free_rerere_id(struct string_list_item *item)
{
free(item->util);
}
static const char *rerere_id_hex(const struct rerere_id *id)
{
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
return id->collection->name;
}
static void fit_variant(struct rerere_dir *rr_dir, int variant)
{
variant++;
ALLOC_GROW(rr_dir->status, variant, rr_dir->status_alloc);
if (rr_dir->status_nr < variant) {
memset(rr_dir->status + rr_dir->status_nr,
'\0', variant - rr_dir->status_nr);
rr_dir->status_nr = variant;
}
}
static void assign_variant(struct rerere_id *id)
{
int variant;
struct rerere_dir *rr_dir = id->collection;
variant = id->variant;
if (variant < 0) {
for (variant = 0; variant < rr_dir->status_nr; variant++)
if (!rr_dir->status[variant])
break;
}
fit_variant(rr_dir, variant);
id->variant = variant;
}
const char *rerere_path(const struct rerere_id *id, const char *file)
{
if (!file)
return git_path("rr-cache/%s", rerere_id_hex(id));
if (id->variant <= 0)
return git_path("rr-cache/%s/%s", rerere_id_hex(id), file);
return git_path("rr-cache/%s/%s.%d",
rerere_id_hex(id), file, id->variant);
}
static int is_rr_file(const char *name, const char *filename, int *variant)
{
const char *suffix;
char *ep;
if (!strcmp(name, filename)) {
*variant = 0;
return 1;
}
if (!skip_prefix(name, filename, &suffix) || *suffix != '.')
return 0;
errno = 0;
*variant = strtol(suffix + 1, &ep, 10);
if (errno || *ep)
return 0;
return 1;
}
static void scan_rerere_dir(struct rerere_dir *rr_dir)
{
struct dirent *de;
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
DIR *dir = opendir(git_path("rr-cache/%s", rr_dir->name));
if (!dir)
return;
while ((de = readdir(dir)) != NULL) {
int variant;
if (is_rr_file(de->d_name, "postimage", &variant)) {
fit_variant(rr_dir, variant);
rr_dir->status[variant] |= RR_HAS_POSTIMAGE;
} else if (is_rr_file(de->d_name, "preimage", &variant)) {
fit_variant(rr_dir, variant);
rr_dir->status[variant] |= RR_HAS_PREIMAGE;
}
}
closedir(dir);
}
static struct rerere_dir *find_rerere_dir(const char *hex)
{
struct rerere_dir *rr_dir;
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
rr_dir = strmap_get(&rerere_dirs, hex);
if (!rr_dir) {
FLEX_ALLOC_STR(rr_dir, name, hex);
rr_dir->status = NULL;
rr_dir->status_nr = 0;
rr_dir->status_alloc = 0;
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
strmap_put(&rerere_dirs, hex, rr_dir);
scan_rerere_dir(rr_dir);
}
rerere: use strmap to store rerere directories We store a struct for each directory we access under .git/rr-cache. The structs are kept in an array sorted by the binary hash associated with their name (and we do lookups with a binary search). This works OK, but there are a few small downsides: - the amount of code isn't huge, but it's more than we'd need using one of our other stock data structures - the insertion into a sorted array is quadratic (though in practice it's unlikely anybody has enough conflicts for this to matter) - it's intimately tied to the representation of an object hash. This isn't a big deal, as the conflict ids we generate use the same hash, but it produces a few awkward bits (e.g., we are the only user of hash_pos() that is not using object_id). Let's instead just treat the directory names as strings, and store them in a strmap. This is less code, and removes the use of hash_pos(). Insertion is now non-quadratic, though we probably use a bit more memory. Besides the hash table overhead, and storing hex bytes instead of a binary hash, we actually store each name twice. Other code expects to access the name of a rerere_dir struct from the struct itself, so we need a copy there. But strmap keeps its own copy of the name, as well. Using a bare hashmap instead of strmap means we could use the name for both, but at the cost of extra code (e.g., our own comparison function). Likewise, strmap has a feature to use a pointer to the in-struct name at the cost of a little extra code. I didn't do either here, as simple code seemed more important than squeezing out a few bytes of efficiency. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2021-01-28 06:34:31 +00:00
return rr_dir;
}
static int has_rerere_resolution(const struct rerere_id *id)
{
const int both = RR_HAS_POSTIMAGE|RR_HAS_PREIMAGE;
int variant = id->variant;
if (variant < 0)
return 0;
return ((id->collection->status[variant] & both) == both);
}
static struct rerere_id *new_rerere_id_hex(char *hex)
{
struct rerere_id *id = xmalloc(sizeof(*id));
id->collection = find_rerere_dir(hex);
id->variant = -1; /* not known yet */
return id;
}
static struct rerere_id *new_rerere_id(unsigned char *hash)
{
return new_rerere_id_hex(hash_to_hex(hash));
}
/*
* $GIT_DIR/MERGE_RR file is a collection of records, each of which is
* "conflict ID", a HT and pathname, terminated with a NUL, and is
* used to keep track of the set of paths that "rerere" may need to
* work on (i.e. what is left by the previous invocation of "git
* rerere" during the current conflict resolution session).
*/
static void read_rr(struct repository *r, struct string_list *rr)
{
struct strbuf buf = STRBUF_INIT;
FILE *in = fopen_or_warn(git_path_merge_rr(r), "r");
if (!in)
return;
while (!strbuf_getwholeline(&buf, in, '\0')) {
char *path;
unsigned char hash[GIT_MAX_RAWSZ];
struct rerere_id *id;
int variant;
const unsigned hexsz = the_hash_algo->hexsz;
/* There has to be the hash, tab, path and then NUL */
hex: retire get_sha1_hex() The naming convention around get_sha1_hex() and its friends is awkward these days, after "struct object_id" was introduced. There are three public functions around this area: * get_sha1_hex() - use the implied the_hash_algo, fill uchar * * get_oid_hex() - use the implied the_hash_algo, fill oid * * get_oid_hex_algop() - use the passed algop, fill oid * Between the latter two, the "_algop" suffix signals whether the the_hash_algo is used as the implied algorithm or the caller should pass an algorithm explicitly. That is very much understandable and is a good convention. Between the former two, however, the "SHA1" vs "OID" in the names differentiate in what type of variable the result is stored. We could argue that it makes sense to use "SHA1" to mean "flat byte buffer" to honor the historical practice in the days before "struct object_id" was invented, but the natural fourth friend of the above group would take an algop and fill a flat byte buffer, and it would be strange to name it get_sha1_hex_algop(). Do we use the passed in algo, or are we limited to SHA-1 ;-)? In fact, such a function exists, albeit as a private helper function used by the implementation of these functions, and is named a lot more sensibly: get_hash_hex_algop(). Correct the misnomer of get_sha1_hex() and use "hash", instead of "sha1", as "flat byte buffer that stores binary (as opposed to hexadecimal) representation of the hash". The four (2x2) friends now become: * get_hash_hex() - use the implied the_hash_algo, fill uchar * * get_oid_hex() - use the implied the_hash_algo, fill oid * * get_hash_hex_algop() - use the passed algop, fill uchar * * get_oid_hex_algop() - use the passed algop, fill oid * As there are only two remaining calls to get_sha1_hex() in the codebase right now, the blast radious of this change is fairly small. Signed-off-by: Junio C Hamano <gitster@pobox.com>
2023-07-24 23:11:03 +00:00
if (buf.len < hexsz + 2 || get_hash_hex(buf.buf, hash))
die(_("corrupt MERGE_RR"));
if (buf.buf[hexsz] != '.') {
variant = 0;
path = buf.buf + hexsz;
} else {
errno = 0;
variant = strtol(buf.buf + hexsz + 1, &path, 10);
if (errno)
die(_("corrupt MERGE_RR"));
}
if (*(path++) != '\t')
die(_("corrupt MERGE_RR"));
buf.buf[hexsz] = '\0';
id = new_rerere_id_hex(buf.buf);
id->variant = variant;
string_list_insert(rr, path)->util = id;
}
strbuf_release(&buf);
fclose(in);
}
static struct lock_file write_lock;
static int write_rr(struct string_list *rr, int out_fd)
{
int i;
for (i = 0; i < rr->nr; i++) {
struct strbuf buf = STRBUF_INIT;
struct rerere_id *id;
assert(rr->items[i].util != RERERE_RESOLVED);
id = rr->items[i].util;
if (!id)
continue;
assert(id->variant >= 0);
if (0 < id->variant)
strbuf_addf(&buf, "%s.%d\t%s%c",
rerere_id_hex(id), id->variant,
rr->items[i].string, 0);
else
strbuf_addf(&buf, "%s\t%s%c",
rerere_id_hex(id),
rr->items[i].string, 0);
avoid "write_in_full(fd, buf, len) != len" pattern The return value of write_in_full() is either "-1", or the requested number of bytes[1]. If we make a partial write before seeing an error, we still return -1, not a partial value. This goes back to f6aa66cb95 (write_in_full: really write in full or return error on disk full., 2007-01-11). So checking anything except "was the return value negative" is pointless. And there are a couple of reasons not to do so: 1. It can do a funny signed/unsigned comparison. If your "len" is signed (e.g., a size_t) then the compiler will promote the "-1" to its unsigned variant. This works out for "!= len" (unless you really were trying to write the maximum size_t bytes), but is a bug if you check "< len" (an example of which was fixed recently in config.c). We should avoid promoting the mental model that you need to check the length at all, so that new sites are not tempted to copy us. 2. Checking for a negative value is shorter to type, especially when the length is an expression. 3. Linus says so. In d34cf19b89 (Clean up write_in_full() users, 2007-01-11), right after the write_in_full() semantics were changed, he wrote: I really wish every "write_in_full()" user would just check against "<0" now, but this fixes the nasty and stupid ones. Appeals to authority aside, this makes it clear that writing it this way does not have an intentional benefit. It's a historical curiosity that we never bothered to clean up (and which was undoubtedly cargo-culted into new sites). So let's convert these obviously-correct cases (this includes write_str_in_full(), which is just a wrapper for write_in_full()). [1] A careful reader may notice there is one way that write_in_full() can return a different value. If we ask write() to write N bytes and get a return value that is _larger_ than N, we could return a larger total. But besides the fact that this would imply a totally broken version of write(), it would already invoke undefined behavior. Our internal remaining counter is an unsigned size_t, which means that subtracting too many byte will wrap it around to a very large number. So we'll instantly begin reading off the end of the buffer, trying to write gigabytes (or petabytes) of data. Signed-off-by: Jeff King <peff@peff.net> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2017-09-13 17:16:03 +00:00
if (write_in_full(out_fd, buf.buf, buf.len) < 0)
die(_("unable to write rerere record"));
strbuf_release(&buf);
}
if (commit_lock_file(&write_lock) != 0)
die(_("unable to write rerere record"));
return 0;
}
/*
* "rerere" interacts with conflicted file contents using this I/O
* abstraction. It reads a conflicted contents from one place via
* "getline()" method, and optionally can write it out after
* normalizing the conflicted hunks to the "output". Subclasses of
* rerere_io embed this structure at the beginning of their own
* rerere_io object.
*/
struct rerere_io {
int (*getline)(struct strbuf *, struct rerere_io *);
FILE *output;
int wrerror;
/* some more stuff */
};
static void ferr_write(const void *p, size_t count, FILE *fp, int *err)
{
if (!count || *err)
return;
if (fwrite(p, count, 1, fp) != 1)
*err = errno;
}
static inline void ferr_puts(const char *s, FILE *fp, int *err)
{
ferr_write(s, strlen(s), fp, err);
}
static void rerere_io_putstr(const char *str, struct rerere_io *io)
{
if (io->output)
ferr_puts(str, io->output, &io->wrerror);
}
static void rerere_io_putmem(const char *mem, size_t sz, struct rerere_io *io)
{
if (io->output)
ferr_write(mem, sz, io->output, &io->wrerror);
}
/*
* Subclass of rerere_io that reads from an on-disk file
*/
struct rerere_io_file {
struct rerere_io io;
FILE *input;
};
/*
* ... and its getline() method implementation
*/
static int rerere_file_getline(struct strbuf *sb, struct rerere_io *io_)
{
struct rerere_io_file *io = (struct rerere_io_file *)io_;
return strbuf_getwholeline(sb, io->input, '\n');
}
/*
* Require the exact number of conflict marker letters, no more, no
* less, followed by SP or any whitespace
* (including LF).
*/
static int is_cmarker(char *buf, int marker_char, int marker_size)
{
int want_sp;
/*
* The beginning of our version and the end of their version
* always are labeled like "<<<<< ours" or ">>>>> theirs",
* hence we set want_sp for them. Note that the version from
* the common ancestor in diff3-style output is not always
* labelled (e.g. "||||| common" is often seen but "|||||"
* alone is also valid), so we do not set want_sp.
*/
want_sp = (marker_char == '<') || (marker_char == '>');
while (marker_size--)
if (*buf++ != marker_char)
return 0;
if (want_sp && *buf != ' ')
return 0;
return isspace(*buf);
}
static void rerere_strbuf_putconflict(struct strbuf *buf, int ch, size_t size)
{
strbuf_addchars(buf, ch, size);
strbuf_addch(buf, '\n');
}
static int handle_conflict(struct strbuf *out, struct rerere_io *io,
int marker_size, git_hash_ctx *ctx)
{
enum {
RR_SIDE_1 = 0, RR_SIDE_2, RR_ORIGINAL
} hunk = RR_SIDE_1;
struct strbuf one = STRBUF_INIT, two = STRBUF_INIT;
struct strbuf buf = STRBUF_INIT, conflict = STRBUF_INIT;
int has_conflicts = -1;
while (!io->getline(&buf, io)) {
if (is_cmarker(buf.buf, '<', marker_size)) {
if (handle_conflict(&conflict, io, marker_size, NULL) < 0)
break;
if (hunk == RR_SIDE_1)
strbuf_addbuf(&one, &conflict);
else
strbuf_addbuf(&two, &conflict);
strbuf_release(&conflict);
} else if (is_cmarker(buf.buf, '|', marker_size)) {
if (hunk != RR_SIDE_1)
break;
hunk = RR_ORIGINAL;
} else if (is_cmarker(buf.buf, '=', marker_size)) {
if (hunk != RR_SIDE_1 && hunk != RR_ORIGINAL)
break;
hunk = RR_SIDE_2;
} else if (is_cmarker(buf.buf, '>', marker_size)) {
if (hunk != RR_SIDE_2)
break;
if (strbuf_cmp(&one, &two) > 0)
strbuf_swap(&one, &two);
has_conflicts = 1;
rerere_strbuf_putconflict(out, '<', marker_size);
strbuf_addbuf(out, &one);
rerere_strbuf_putconflict(out, '=', marker_size);
strbuf_addbuf(out, &two);
rerere_strbuf_putconflict(out, '>', marker_size);
if (ctx) {
the_hash_algo->update_fn(ctx, one.buf ?
one.buf : "",
one.len + 1);
the_hash_algo->update_fn(ctx, two.buf ?
two.buf : "",
two.len + 1);
}
break;
} else if (hunk == RR_SIDE_1)
strbuf_addbuf(&one, &buf);
else if (hunk == RR_ORIGINAL)
; /* discard */
else if (hunk == RR_SIDE_2)
strbuf_addbuf(&two, &buf);
}
strbuf_release(&one);
strbuf_release(&two);
strbuf_release(&buf);
return has_conflicts;
}
/*
* Read contents a file with conflicts, normalize the conflicts
* by (1) discarding the common ancestor version in diff3-style,
* (2) reordering our side and their side so that whichever sorts
* alphabetically earlier comes before the other one, while
* computing the "conflict ID", which is just an SHA-1 hash of
* one side of the conflict, NUL, the other side of the conflict,
* and NUL concatenated together.
*
* Return 1 if conflict hunks are found, 0 if there are no conflict
* hunks and -1 if an error occurred.
*/
static int handle_path(unsigned char *hash, struct rerere_io *io, int marker_size)
{
git_hash_ctx ctx;
struct strbuf buf = STRBUF_INIT, out = STRBUF_INIT;
int has_conflicts = 0;
if (hash)
the_hash_algo->init_fn(&ctx);
while (!io->getline(&buf, io)) {
if (is_cmarker(buf.buf, '<', marker_size)) {
has_conflicts = handle_conflict(&out, io, marker_size,
hash ? &ctx : NULL);
if (has_conflicts < 0)
break;
rerere_io_putmem(out.buf, out.len, io);
strbuf_reset(&out);
} else
rerere_io_putstr(buf.buf, io);
}
strbuf_release(&buf);
strbuf_release(&out);
if (hash)
the_hash_algo->final_fn(hash, &ctx);
return has_conflicts;
}
/*
* Scan the path for conflicts, do the "handle_path()" thing above, and
* return the number of conflict hunks found.
*/
static int handle_file(struct index_state *istate,
const char *path, unsigned char *hash, const char *output)
{
int has_conflicts = 0;
struct rerere_io_file io;
int marker_size = ll_merge_marker_size(istate, path);
memset(&io, 0, sizeof(io));
io.io.getline = rerere_file_getline;
io.input = fopen(path, "r");
io.io.wrerror = 0;
if (!io.input)
return error_errno(_("could not open '%s'"), path);
if (output) {
io.io.output = fopen(output, "w");
if (!io.io.output) {
error_errno(_("could not write '%s'"), output);
fclose(io.input);
return -1;
}
}
has_conflicts = handle_path(hash, (struct rerere_io *)&io, marker_size);
fclose(io.input);
if (io.io.wrerror)
error(_("there were errors while writing '%s' (%s)"),
path, strerror(io.io.wrerror));
if (io.io.output && fclose(io.io.output))
io.io.wrerror = error_errno(_("failed to flush '%s'"), path);
if (has_conflicts < 0) {
if (output)
unlink_or_warn(output);
return error(_("could not parse conflict hunks in '%s'"), path);
}
if (io.io.wrerror)
return -1;
return has_conflicts;
}
/*
* Look at a cache entry at "i" and see if it is not conflicting,
* conflicting and we are willing to handle, or conflicting and
* we are unable to handle, and return the determination in *type.
* Return the cache index to be looked at next, by skipping the
* stages we have already looked at in this invocation of this
* function.
*/
static int check_one_conflict(struct index_state *istate, int i, int *type)
{
const struct cache_entry *e = istate->cache[i];
if (!ce_stage(e)) {
*type = RESOLVED;
return i + 1;
}
*type = PUNTED;
while (i < istate->cache_nr && ce_stage(istate->cache[i]) == 1)
i++;
/* Only handle regular files with both stages #2 and #3 */
if (i + 1 < istate->cache_nr) {
const struct cache_entry *e2 = istate->cache[i];
const struct cache_entry *e3 = istate->cache[i + 1];
if (ce_stage(e2) == 2 &&
ce_stage(e3) == 3 &&
ce_same_name(e, e3) &&
S_ISREG(e2->ce_mode) &&
S_ISREG(e3->ce_mode))
*type = THREE_STAGED;
}
/* Skip the entries with the same name */
while (i < istate->cache_nr && ce_same_name(e, istate->cache[i]))
i++;
return i;
}
/*
* Scan the index and find paths that have conflicts that rerere can
* handle, i.e. the ones that has both stages #2 and #3.
*
* NEEDSWORK: we do not record or replay a previous "resolve by
* deletion" for a delete-modify conflict, as that is inherently risky
* without knowing what modification is being discarded. The only
* safe case, i.e. both side doing the deletion and modification that
* are identical to the previous round, might want to be handled,
* though.
*/
static int find_conflict(struct repository *r, struct string_list *conflict)
{
int i;
if (repo_read_index(r) < 0)
return error(_("index file corrupt"));
for (i = 0; i < r->index->cache_nr;) {
int conflict_type;
const struct cache_entry *e = r->index->cache[i];
i = check_one_conflict(r->index, i, &conflict_type);
if (conflict_type == THREE_STAGED)
string_list_insert(conflict, (const char *)e->name);
}
return 0;
}
/*
* The merge_rr list is meant to hold outstanding conflicted paths
* that rerere could handle. Abuse the list by adding other types of
* entries to allow the caller to show "rerere remaining".
*
* - Conflicted paths that rerere does not handle are added
* - Conflicted paths that have been resolved are marked as such
* by storing RERERE_RESOLVED to .util field (where conflict ID
* is expected to be stored).
*
* Do *not* write MERGE_RR file out after calling this function.
*
* NEEDSWORK: we may want to fix the caller that implements "rerere
* remaining" to do this without abusing merge_rr.
*/
int rerere_remaining(struct repository *r, struct string_list *merge_rr)
{
int i;
if (setup_rerere(r, merge_rr, RERERE_READONLY))
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
return 0;
if (repo_read_index(r) < 0)
return error(_("index file corrupt"));
for (i = 0; i < r->index->cache_nr;) {
int conflict_type;
const struct cache_entry *e = r->index->cache[i];
i = check_one_conflict(r->index, i, &conflict_type);
if (conflict_type == PUNTED)
string_list_insert(merge_rr, (const char *)e->name);
else if (conflict_type == RESOLVED) {
struct string_list_item *it;
it = string_list_lookup(merge_rr, (const char *)e->name);
if (it) {
free_rerere_id(it);
it->util = RERERE_RESOLVED;
}
}
}
return 0;
}
/*
* Try using the given conflict resolution "ID" to see
* if that recorded conflict resolves cleanly what we
* got in the "cur".
*/
static int try_merge(struct index_state *istate,
const struct rerere_id *id, const char *path,
mmfile_t *cur, mmbuffer_t *result)
{
enum ll_merge_result ret;
mmfile_t base = {NULL, 0}, other = {NULL, 0};
if (read_mmfile(&base, rerere_path(id, "preimage")) ||
read_mmfile(&other, rerere_path(id, "postimage"))) {
ret = LL_MERGE_CONFLICT;
} else {
/*
* A three-way merge. Note that this honors user-customizable
* low-level merge driver settings.
*/
ret = ll_merge(result, path, &base, NULL, cur, "", &other, "",
istate, NULL);
}
free(base.ptr);
free(other.ptr);
return ret;
}
/*
* Find the conflict identified by "id"; the change between its
* "preimage" (i.e. a previous contents with conflict markers) and its
* "postimage" (i.e. the corresponding contents with conflicts
* resolved) may apply cleanly to the contents stored in "path", i.e.
* the conflict this time around.
*
* Returns 0 for successful replay of recorded resolution, or non-zero
* for failure.
*/
static int merge(struct index_state *istate, const struct rerere_id *id, const char *path)
{
FILE *f;
int ret;
mmfile_t cur = {NULL, 0};
mmbuffer_t result = {NULL, 0};
/*
* Normalize the conflicts in path and write it out to
* "thisimage" temporary file.
*/
if ((handle_file(istate, path, NULL, rerere_path(id, "thisimage")) < 0) ||
read_mmfile(&cur, rerere_path(id, "thisimage"))) {
ret = 1;
goto out;
}
ret = try_merge(istate, id, path, &cur, &result);
if (ret)
goto out;
/*
* A successful replay of recorded resolution.
* Mark that "postimage" was used to help gc.
*/
if (utime(rerere_path(id, "postimage"), NULL) < 0)
warning_errno(_("failed utime() on '%s'"),
rerere_path(id, "postimage"));
/* Update "path" with the resolution */
f = fopen(path, "w");
if (!f)
return error_errno(_("could not open '%s'"), path);
if (fwrite(result.ptr, result.size, 1, f) != 1)
error_errno(_("could not write '%s'"), path);
if (fclose(f))
return error_errno(_("writing '%s' failed"), path);
out:
free(cur.ptr);
free(result.ptr);
return ret;
}
static void update_paths(struct repository *r, struct string_list *update)
{
struct lock_file index_lock = LOCK_INIT;
int i;
repo_hold_locked_index(r, &index_lock, LOCK_DIE_ON_ERROR);
for (i = 0; i < update->nr; i++) {
struct string_list_item *item = &update->items[i];
if (add_file_to_index(r->index, item->string, 0))
exit(128);
fprintf_ln(stderr, _("Staged '%s' using previous resolution."),
item->string);
}
if (write_locked_index(r->index, &index_lock,
COMMIT_LOCK | SKIP_IF_UNCHANGED))
die(_("unable to write new index file"));
}
static void remove_variant(struct rerere_id *id)
{
unlink_or_warn(rerere_path(id, "postimage"));
unlink_or_warn(rerere_path(id, "preimage"));
id->collection->status[id->variant] = 0;
}
/*
* The path indicated by rr_item may still have conflict for which we
* have a recorded resolution, in which case replay it and optionally
* update it. Or it may have been resolved by the user and we may
* only have the preimage for that conflict, in which case the result
* needs to be recorded as a resolution in a postimage file.
*/
static void do_rerere_one_path(struct index_state *istate,
struct string_list_item *rr_item,
struct string_list *update)
{
const char *path = rr_item->string;
struct rerere_id *id = rr_item->util;
struct rerere_dir *rr_dir = id->collection;
int variant;
variant = id->variant;
/* Has the user resolved it already? */
if (variant >= 0) {
if (!handle_file(istate, path, NULL, NULL)) {
copy_file(rerere_path(id, "postimage"), path, 0666);
id->collection->status[variant] |= RR_HAS_POSTIMAGE;
fprintf_ln(stderr, _("Recorded resolution for '%s'."), path);
free_rerere_id(rr_item);
rr_item->util = NULL;
return;
}
/*
* There may be other variants that can cleanly
* replay. Try them and update the variant number for
* this one.
*/
}
/* Does any existing resolution apply cleanly? */
for (variant = 0; variant < rr_dir->status_nr; variant++) {
const int both = RR_HAS_PREIMAGE | RR_HAS_POSTIMAGE;
struct rerere_id vid = *id;
if ((rr_dir->status[variant] & both) != both)
continue;
vid.variant = variant;
if (merge(istate, &vid, path))
continue; /* failed to replay */
/*
* If there already is a different variant that applies
* cleanly, there is no point maintaining our own variant.
*/
if (0 <= id->variant && id->variant != variant)
remove_variant(id);
if (rerere_autoupdate)
string_list_insert(update, path);
else
fprintf_ln(stderr,
_("Resolved '%s' using previous resolution."),
path);
free_rerere_id(rr_item);
rr_item->util = NULL;
return;
}
/* None of the existing one applies; we need a new variant */
assign_variant(id);
variant = id->variant;
handle_file(istate, path, NULL, rerere_path(id, "preimage"));
if (id->collection->status[variant] & RR_HAS_POSTIMAGE) {
const char *path = rerere_path(id, "postimage");
if (unlink(path))
die_errno(_("cannot unlink stray '%s'"), path);
id->collection->status[variant] &= ~RR_HAS_POSTIMAGE;
}
id->collection->status[variant] |= RR_HAS_PREIMAGE;
fprintf_ln(stderr, _("Recorded preimage for '%s'"), path);
}
static int do_plain_rerere(struct repository *r,
struct string_list *rr, int fd)
{
struct string_list conflict = STRING_LIST_INIT_DUP;
struct string_list update = STRING_LIST_INIT_DUP;
int i;
find_conflict(r, &conflict);
/*
* MERGE_RR records paths with conflicts immediately after
* merge failed. Some of the conflicted paths might have been
* hand resolved in the working tree since then, but the
* initial run would catch all and register their preimages.
*/
for (i = 0; i < conflict.nr; i++) {
struct rerere_id *id;
unsigned char hash[GIT_MAX_RAWSZ];
const char *path = conflict.items[i].string;
int ret;
/*
* Ask handle_file() to scan and assign a
* conflict ID. No need to write anything out
* yet.
*/
ret = handle_file(r->index, path, hash, NULL);
if (ret != 0 && string_list_has_string(rr, path)) {
rerere: fix crash with files rerere can't handle Currently when a user does a conflict resolution and ends it (in any way that calls 'git rerere' again) with a file 'rerere' can't handle, subsequent rerere operations that are interested in that path, such as 'rerere clear' or 'rerere forget <path>' will fail, or even worse in the case of 'rerere clear' segfault. Such states include nested conflicts, or a conflict marker that doesn't have any match. This is because 'git rerere' calculates a conflict file and writes it to the MERGE_RR file. When the user then changes the file in any way rerere can't handle, and then calls 'git rerere' on it again to record the conflict resolution, the handle_file function fails, and removes the 'preimage' file in the rr-cache in the process, while leaving the ID in the MERGE_RR file. Now when 'rerere clear' is run, it reads the ID from the MERGE_RR file, however the 'fit_variant' function for the ID is never called as the 'preimage' file does not exist anymore. This means 'collection->status' in 'has_rerere_resolution' is NULL, and the command will crash. To fix this, remove the rerere ID from the MERGE_RR file in the case when we can't handle it, just after the 'preimage' file was removed and remove the corresponding variant from .git/rr-cache/. Removing it unconditionally is fine here, because if the user would have resolved the conflict and ran rerere, the entry would no longer be in the MERGE_RR file, so we wouldn't have this problem in the first place, while if the conflict was not resolved. Currently there is nothing left in this folder, as the 'preimage' was already deleted by the 'handle_file' function, so 'remove_variant' is a no-op. Still call the function, to make sure we clean everything up, in case we add some other files corresponding to a variant in the future. Note that other variants that have the same conflict ID will not be touched. Signed-off-by: Thomas Gummerer <t.gummerer@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2018-08-05 17:20:32 +00:00
remove_variant(string_list_lookup(rr, path)->util);
string_list_remove(rr, path, 1);
}
if (ret < 1)
continue;
id = new_rerere_id(hash);
string_list_insert(rr, path)->util = id;
/* Ensure that the directory exists. */
mkdir_in_gitdir(rerere_path(id, NULL));
}
for (i = 0; i < rr->nr; i++)
do_rerere_one_path(r->index, &rr->items[i], &update);
if (update.nr)
update_paths(r, &update);
return write_rr(rr, fd);
}
static void git_rerere_config(void)
{
git_config_get_bool("rerere.enabled", &rerere_enabled);
git_config_get_bool("rerere.autoupdate", &rerere_autoupdate);
git_config(git_default_config, NULL);
}
memoize common git-path "constant" files One of the most common uses of git_path() is to pass a constant, like git_path("MERGE_MSG"). This has two drawbacks: 1. The return value is a static buffer, and the lifetime is dependent on other calls to git_path, etc. 2. There's no compile-time checking of the pathname. This is OK for a one-off (after all, we have to spell it correctly at least once), but many of these constant strings appear throughout the code. This patch introduces a series of functions to "memoize" these strings, which are essentially globals for the lifetime of the program. We compute the value once, take ownership of the buffer, and return the cached value for subsequent calls. cache.h provides a helper macro for defining these functions as one-liners, and defines a few common ones for global use. Using a macro is a little bit gross, but it does nicely document the purpose of the functions. If we need to touch them all later (e.g., because we learned how to change the git_dir variable at runtime, and need to invalidate all of the stored values), it will be much easier to have the complete list. Note that the shared-global functions have separate, manual declarations. We could do something clever with the macros (e.g., expand it to a declaration in some places, and a declaration _and_ a definition in path.c). But there aren't that many, and it's probably better to stay away from too-magical macros. Likewise, if we abandon the C preprocessor in favor of generating these with a script, we could get much fancier. E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz". But the small amount of saved typing is probably not worth the resulting confusion to readers who want to grep for the function's definition. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-08-10 09:38:57 +00:00
static GIT_PATH_FUNC(git_path_rr_cache, "rr-cache")
static int is_rerere_enabled(void)
{
int rr_cache_exists;
if (!rerere_enabled)
return 0;
memoize common git-path "constant" files One of the most common uses of git_path() is to pass a constant, like git_path("MERGE_MSG"). This has two drawbacks: 1. The return value is a static buffer, and the lifetime is dependent on other calls to git_path, etc. 2. There's no compile-time checking of the pathname. This is OK for a one-off (after all, we have to spell it correctly at least once), but many of these constant strings appear throughout the code. This patch introduces a series of functions to "memoize" these strings, which are essentially globals for the lifetime of the program. We compute the value once, take ownership of the buffer, and return the cached value for subsequent calls. cache.h provides a helper macro for defining these functions as one-liners, and defines a few common ones for global use. Using a macro is a little bit gross, but it does nicely document the purpose of the functions. If we need to touch them all later (e.g., because we learned how to change the git_dir variable at runtime, and need to invalidate all of the stored values), it will be much easier to have the complete list. Note that the shared-global functions have separate, manual declarations. We could do something clever with the macros (e.g., expand it to a declaration in some places, and a declaration _and_ a definition in path.c). But there aren't that many, and it's probably better to stay away from too-magical macros. Likewise, if we abandon the C preprocessor in favor of generating these with a script, we could get much fancier. E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz". But the small amount of saved typing is probably not worth the resulting confusion to readers who want to grep for the function's definition. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-08-10 09:38:57 +00:00
rr_cache_exists = is_directory(git_path_rr_cache());
if (rerere_enabled < 0)
return rr_cache_exists;
memoize common git-path "constant" files One of the most common uses of git_path() is to pass a constant, like git_path("MERGE_MSG"). This has two drawbacks: 1. The return value is a static buffer, and the lifetime is dependent on other calls to git_path, etc. 2. There's no compile-time checking of the pathname. This is OK for a one-off (after all, we have to spell it correctly at least once), but many of these constant strings appear throughout the code. This patch introduces a series of functions to "memoize" these strings, which are essentially globals for the lifetime of the program. We compute the value once, take ownership of the buffer, and return the cached value for subsequent calls. cache.h provides a helper macro for defining these functions as one-liners, and defines a few common ones for global use. Using a macro is a little bit gross, but it does nicely document the purpose of the functions. If we need to touch them all later (e.g., because we learned how to change the git_dir variable at runtime, and need to invalidate all of the stored values), it will be much easier to have the complete list. Note that the shared-global functions have separate, manual declarations. We could do something clever with the macros (e.g., expand it to a declaration in some places, and a declaration _and_ a definition in path.c). But there aren't that many, and it's probably better to stay away from too-magical macros. Likewise, if we abandon the C preprocessor in favor of generating these with a script, we could get much fancier. E.g., normalizing "FOO/BAR-BAZ" into "git_path_foo_bar_baz". But the small amount of saved typing is probably not worth the resulting confusion to readers who want to grep for the function's definition. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-08-10 09:38:57 +00:00
if (!rr_cache_exists && mkdir_in_gitdir(git_path_rr_cache()))
die(_("could not create directory '%s'"), git_path_rr_cache());
return 1;
}
int setup_rerere(struct repository *r, struct string_list *merge_rr, int flags)
{
int fd;
git_rerere_config();
if (!is_rerere_enabled())
return -1;
if (flags & (RERERE_AUTOUPDATE|RERERE_NOAUTOUPDATE))
rerere_autoupdate = !!(flags & RERERE_AUTOUPDATE);
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
if (flags & RERERE_READONLY)
fd = 0;
else
fd = hold_lock_file_for_update(&write_lock,
git_path_merge_rr(r),
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
LOCK_DIE_ON_ERROR);
read_rr(r, merge_rr);
return fd;
}
/*
* The main entry point that is called internally from codepaths that
* perform mergy operations, possibly leaving conflicted index entries
* and working tree files.
*/
int repo_rerere(struct repository *r, int flags)
{
struct string_list merge_rr = STRING_LIST_INIT_DUP;
int fd, status;
fd = setup_rerere(r, &merge_rr, flags);
if (fd < 0)
return 0;
status = do_plain_rerere(r, &merge_rr, fd);
free_rerere_dirs();
return status;
}
/*
* Subclass of rerere_io that reads from an in-core buffer that is a
* strbuf
*/
struct rerere_io_mem {
struct rerere_io io;
struct strbuf input;
};
/*
* ... and its getline() method implementation
*/
static int rerere_mem_getline(struct strbuf *sb, struct rerere_io *io_)
{
struct rerere_io_mem *io = (struct rerere_io_mem *)io_;
char *ep;
size_t len;
strbuf_release(sb);
if (!io->input.len)
return -1;
ep = memchr(io->input.buf, '\n', io->input.len);
if (!ep)
ep = io->input.buf + io->input.len;
else if (*ep == '\n')
ep++;
len = ep - io->input.buf;
strbuf_add(sb, io->input.buf, len);
strbuf_remove(&io->input, 0, len);
return 0;
}
static int handle_cache(struct index_state *istate,
const char *path, unsigned char *hash, const char *output)
{
mmfile_t mmfile[3] = {{NULL}};
mmbuffer_t result = {NULL, 0};
const struct cache_entry *ce;
int pos, len, i, has_conflicts;
struct rerere_io_mem io;
int marker_size = ll_merge_marker_size(istate, path);
/*
* Reproduce the conflicted merge in-core
*/
len = strlen(path);
pos = index_name_pos(istate, path, len);
if (0 <= pos)
return -1;
pos = -pos - 1;
while (pos < istate->cache_nr) {
enum object_type type;
unsigned long size;
ce = istate->cache[pos++];
if (ce_namelen(ce) != len || memcmp(ce->name, path, len))
break;
i = ce_stage(ce) - 1;
if (!mmfile[i].ptr) {
mmfile[i].ptr = repo_read_object_file(the_repository,
&ce->oid, &type,
&size);
mmfile[i].size = size;
}
}
for (i = 0; i < 3; i++)
if (!mmfile[i].ptr && !mmfile[i].size)
mmfile[i].ptr = xstrdup("");
/*
* NEEDSWORK: handle conflicts from merges with
* merge.renormalize set, too?
*/
ll_merge(&result, path, &mmfile[0], NULL,
&mmfile[1], "ours",
&mmfile[2], "theirs",
istate, NULL);
for (i = 0; i < 3; i++)
free(mmfile[i].ptr);
memset(&io, 0, sizeof(io));
io.io.getline = rerere_mem_getline;
if (output)
io.io.output = fopen(output, "w");
else
io.io.output = NULL;
strbuf_init(&io.input, 0);
strbuf_attach(&io.input, result.ptr, result.size, result.size);
/*
* Grab the conflict ID and optionally write the original
* contents with conflict markers out.
*/
has_conflicts = handle_path(hash, (struct rerere_io *)&io, marker_size);
strbuf_release(&io.input);
if (io.io.output)
fclose(io.io.output);
return has_conflicts;
}
static int rerere_forget_one_path(struct index_state *istate,
const char *path,
struct string_list *rr)
{
const char *filename;
struct rerere_id *id;
unsigned char hash[GIT_MAX_RAWSZ];
int ret;
struct string_list_item *item;
/*
* Recreate the original conflict from the stages in the
* index and compute the conflict ID
*/
ret = handle_cache(istate, path, hash, NULL);
if (ret < 1)
return error(_("could not parse conflict hunks in '%s'"), path);
/* Nuke the recorded resolution for the conflict */
id = new_rerere_id(hash);
for (id->variant = 0;
id->variant < id->collection->status_nr;
id->variant++) {
mmfile_t cur = { NULL, 0 };
mmbuffer_t result = {NULL, 0};
int cleanly_resolved;
if (!has_rerere_resolution(id))
continue;
handle_cache(istate, path, hash, rerere_path(id, "thisimage"));
if (read_mmfile(&cur, rerere_path(id, "thisimage"))) {
free(cur.ptr);
error(_("failed to update conflicted state in '%s'"), path);
goto fail_exit;
}
cleanly_resolved = !try_merge(istate, id, path, &cur, &result);
free(result.ptr);
free(cur.ptr);
if (cleanly_resolved)
break;
}
if (id->collection->status_nr <= id->variant) {
error(_("no remembered resolution for '%s'"), path);
goto fail_exit;
}
filename = rerere_path(id, "postimage");
if (unlink(filename)) {
if (errno == ENOENT)
error(_("no remembered resolution for '%s'"), path);
else
error_errno(_("cannot unlink '%s'"), filename);
goto fail_exit;
}
/*
* Update the preimage so that the user can resolve the
* conflict in the working tree, run us again to record
* the postimage.
*/
handle_cache(istate, path, hash, rerere_path(id, "preimage"));
fprintf_ln(stderr, _("Updated preimage for '%s'"), path);
/*
* And remember that we can record resolution for this
* conflict when the user is done.
*/
item = string_list_insert(rr, path);
free_rerere_id(item);
item->util = id;
fprintf(stderr, _("Forgot resolution for '%s'\n"), path);
return 0;
fail_exit:
free(id);
return -1;
}
int rerere_forget(struct repository *r, struct pathspec *pathspec)
{
int i, fd;
struct string_list conflict = STRING_LIST_INIT_DUP;
struct string_list merge_rr = STRING_LIST_INIT_DUP;
if (repo_read_index(r) < 0)
return error(_("index file corrupt"));
fd = setup_rerere(r, &merge_rr, RERERE_NOAUTOUPDATE);
if (fd < 0)
return 0;
/*
* The paths may have been resolved (incorrectly);
* recover the original conflicted state and then
* find the conflicted paths.
*/
unmerge_index(r->index, pathspec, 0);
find_conflict(r, &conflict);
for (i = 0; i < conflict.nr; i++) {
struct string_list_item *it = &conflict.items[i];
if (!match_pathspec(r->index, pathspec, it->string,
strlen(it->string), 0, NULL, 0))
continue;
rerere_forget_one_path(r->index, it->string, &merge_rr);
}
return write_rr(&merge_rr, fd);
}
/*
* Garbage collection support
*/
static timestamp_t rerere_created_at(struct rerere_id *id)
{
struct stat st;
return stat(rerere_path(id, "preimage"), &st) ? (time_t) 0 : st.st_mtime;
}
static timestamp_t rerere_last_used_at(struct rerere_id *id)
{
struct stat st;
return stat(rerere_path(id, "postimage"), &st) ? (time_t) 0 : st.st_mtime;
}
/*
* Remove the recorded resolution for a given conflict ID
*/
static void unlink_rr_item(struct rerere_id *id)
{
unlink_or_warn(rerere_path(id, "thisimage"));
remove_variant(id);
id->collection->status[id->variant] = 0;
}
static void prune_one(struct rerere_id *id,
timestamp_t cutoff_resolve, timestamp_t cutoff_noresolve)
{
timestamp_t then;
timestamp_t cutoff;
then = rerere_last_used_at(id);
if (then)
cutoff = cutoff_resolve;
else {
then = rerere_created_at(id);
if (!then)
return;
cutoff = cutoff_noresolve;
}
if (then < cutoff)
unlink_rr_item(id);
}
/* Does the basename in "path" look plausibly like an rr-cache entry? */
static int is_rr_cache_dirname(const char *path)
{
struct object_id oid;
const char *end;
return !parse_oid_hex(path, &oid, &end) && !*end;
}
void rerere_gc(struct repository *r, struct string_list *rr)
{
struct string_list to_remove = STRING_LIST_INIT_DUP;
DIR *dir;
struct dirent *e;
int i;
timestamp_t now = time(NULL);
timestamp_t cutoff_noresolve = now - 15 * 86400;
timestamp_t cutoff_resolve = now - 60 * 86400;
if (setup_rerere(r, rr, 0) < 0)
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
return;
git_config_get_expiry_in_days("gc.rerereresolved", &cutoff_resolve, now);
git_config_get_expiry_in_days("gc.rerereunresolved", &cutoff_noresolve, now);
git_config(git_default_config, NULL);
dir = opendir(git_path("rr-cache"));
if (!dir)
die_errno(_("unable to open rr-cache directory"));
/* Collect stale conflict IDs ... */
while ((e = readdir_skip_dot_and_dotdot(dir))) {
struct rerere_dir *rr_dir;
struct rerere_id id;
int now_empty;
if (!is_rr_cache_dirname(e->d_name))
continue; /* or should we remove e->d_name? */
rr_dir = find_rerere_dir(e->d_name);
now_empty = 1;
for (id.variant = 0, id.collection = rr_dir;
id.variant < id.collection->status_nr;
id.variant++) {
prune_one(&id, cutoff_resolve, cutoff_noresolve);
if (id.collection->status[id.variant])
now_empty = 0;
}
if (now_empty)
string_list_append(&to_remove, e->d_name);
}
closedir(dir);
/* ... and then remove the empty directories */
for (i = 0; i < to_remove.nr; i++)
rmdir(git_path("rr-cache/%s", to_remove.items[i].string));
string_list_clear(&to_remove, 0);
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
rollback_lock_file(&write_lock);
}
/*
* During a conflict resolution, after "rerere" recorded the
* preimages, abandon them if the user did not resolve them or
* record their resolutions. And drop $GIT_DIR/MERGE_RR.
*
* NEEDSWORK: shouldn't we be calling this from "reset --hard"?
*/
void rerere_clear(struct repository *r, struct string_list *merge_rr)
{
int i;
if (setup_rerere(r, merge_rr, 0) < 0)
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
return;
for (i = 0; i < merge_rr->nr; i++) {
struct rerere_id *id = merge_rr->items[i].util;
if (!has_rerere_resolution(id)) {
unlink_rr_item(id);
rmdir(rerere_path(id, NULL));
}
}
unlink_or_warn(git_path_merge_rr(r));
rerere: release lockfile in non-writing functions There's a bug in builtin/am.c in which we take a lock on MERGE_RR recursively. But rather than fix am.c, this patch fixes the confusing interface from rerere.c that caused the bug. Read on for the gory details. The setup_rerere() function both reads the existing MERGE_RR file, and takes MERGE_RR.lock. In the rerere() and rerere_forget() functions, we end up in write_rr(), which will then commit the lock file. But for functions like rerere_clear() that do not write to MERGE_RR, we expect the caller to have handled setup_rerere(). That caller would then need to release the lockfile, but it can't; the lock struct is local to rerere.c. For builtin/rerere.c, this is OK. We run a single rerere operation and then exit immediately, which has the side effect of rolling back the lockfile. But in builtin/am.c, this is actively wrong. If we run "git am -3 --skip", we call setup-rerere twice without releasing the lock: 1. The "--skip" causes us to call am_rerere_clear(), which calls setup_rerere(), but never drops the lock. 2. We then proceed to the next patch. 3. The "--3way" may cause us to call rerere() to handle conflicts in that patch, but we are already holding the lock. The lockfile code dies with: BUG: prepare_tempfile_object called for active object We could fix this by having rerere_clear() call rollback_lock_file(). But it feels a bit odd for it to roll back a lockfile that it did not itself take. So let's simplify the interface further, and handle setup_rerere in the function itself, taking away the question from the caller over whether they need to do so. We can give rerere_gc() the same treatment, as well (even though it doesn't have any callers besides builtin/rerere.c at this point). Note that these functions don't take flags from their callers to pass along to setup_rerere; that's OK, because the flags would not be meaningful for what they are doing. Both of those functions need to hold the lock because even though they do not write to MERGE_RR, they are still writing and should be protected from a simultaneous "rerere" run. But rerere_remaining(), "rerere diff", and "rerere status" are all read-only operations. They want to setup_rerere(), but do not care about taking the lock in the first place. Since our update of MERGE_RR is the usual atomic rename done by commit_lock_file, they can just do a lockless read. For that, we teach setup_rerere a READONLY flag to avoid the lock. As a bonus, this pushes builtin/rerere.c's setup_rerere call closer to the functions that use it. Which means that "git rerere totally-bogus-command" will no longer silently exit(0) in a repository without rerere enabled. Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
2015-09-01 22:14:09 +00:00
rollback_lock_file(&write_lock);
}