The point of disallowing ".git" in the index is that we
would never want to accidentally overwrite files in the
repository directory. But this means we need to respect the
filesystem's idea of when two paths are equal. The prior
commit added a helper to make such a comparison for NTFS
and FAT32; let's use it in verify_path().
We make this check optional for two reasons:
1. It restricts the set of allowable filenames, which is
unnecessary for people who are not on NTFS nor FAT32.
In practice this probably doesn't matter, though, as
the restricted names are rather obscure and almost
certainly would never come up in practice.
2. It has a minor performance penalty for every path we
insert into the index.
This patch ties the check to the core.protectNTFS config
option. Though this is expected to be most useful on Windows,
we allow it to be set everywhere, as NTFS may be mounted on
other platforms. The variable does default to on for Windows,
though.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We do not allow paths with a ".git" component to be added to
the index, as that would mean repository contents could
overwrite our repository files. However, asking "is this
path the same as .git" is not as simple as strcmp() on some
filesystems.
On NTFS (and FAT32), there exist so-called "short names" for
backwards-compatibility: 8.3 compliant names that refer to the same files
as their long names. As ".git" is not an 8.3 compliant name, a short name
is generated automatically, typically "git~1".
Depending on the Windows version, any combination of trailing spaces and
periods are ignored, too, so that both "git~1." and ".git." still refer
to the Git directory. The reason is that 8.3 stores file names shorter
than 8 characters with trailing spaces. So literally, it does not matter
for the short name whether it is padded with spaces or whether it is
shorter than 8 characters, it is considered to be the exact same.
The period is the separator between file name and file extension, and
again, an empty extension consists just of spaces in 8.3 format. So
technically, we would need only take care of the equivalent of this
regex:
(\.git {0,4}|git~1 {0,3})\. {0,3}
However, there are indications that at least some Windows versions might
be more lenient and accept arbitrary combinations of trailing spaces and
periods and strip them out. So we're playing it real safe here. Besides,
there can be little doubt about the intention behind using file names
matching even the more lenient pattern specified above, therefore we
should be fine with disallowing such patterns.
Extra care is taken to catch names such as '.\\.git\\booh' because the
backslash is marked as a directory separator only on Windows, and we want
to use this new helper function also in fsck on other platforms.
A big thank you goes to Ed Thomson and an unnamed Microsoft engineer for
the detailed analysis performed to come up with the corresponding fixes
for libgit2.
This commit adds a function to detect whether a given file name can refer
to the Git directory by mistake.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The point of disallowing ".git" in the index is that we
would never want to accidentally overwrite files in the
repository directory. But this means we need to respect the
filesystem's idea of when two paths are equal. The prior
commit added a helper to make such a comparison for HFS+;
let's use it in verify_path.
We make this check optional for two reasons:
1. It restricts the set of allowable filenames, which is
unnecessary for people who are not on HFS+. In practice
this probably doesn't matter, though, as the restricted
names are rather obscure and almost certainly would
never come up in practice.
2. It has a minor performance penalty for every path we
insert into the index.
This patch ties the check to the core.protectHFS config
option. Though this is expected to be most useful on OS X,
we allow it to be set everywhere, as HFS+ may be mounted on
other platforms. The variable does default to on for OS X,
though.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git config --add section.var val" used to lose existing
section.var whose value was an empty string.
* ta/config-add-to-empty-or-true-fix:
config: avoid a funny sentinel value "a^"
make config --add behave correctly for empty and NULL values
Introduce CONFIG_REGEX_NONE as a more explicit sentinel value to say
"we do not want to replace any existing entry" and use it in the
implementation of "git config --add".
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We keep a static counter to set the commit index on newly
allocated objects. However, since we also need to set the
index on any_objects which are converted to commits, let's
make the counter available as a public function.
While we're moving it, let's make sure the counter is
allocated as an unsigned integer to match the index field in
"struct commit".
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make sure all in-core commit objects are assigned a unique number
so that they can be annotated using the commit-slab API.
* jk/alloc-commit-id:
diff-tree: avoid lookup_unknown_object
object_as_type: set commit index
alloc: factor out commit index
add object_as_type helper for casting objects
parse_object_buffer: do not set object type
move setting of object->type to alloc_* functions
alloc: write out allocator definitions
alloc.c: remove the alloc_raw_commit_node() function
* kb/perf-trace:
api-trace.txt: add trace API documentation
progress: simplify performance measurement by using getnanotime()
wt-status: simplify performance measurement by using getnanotime()
git: add performance tracing for git's main() function to debug scripts
trace: add trace_performance facility to debug performance issues
trace: add high resolution timer function to debug performance issues
trace: add 'file:line' to all trace output
trace: move code around, in preparation to file:line output
trace: add current timestamp to all trace output
trace: disable additional trace output for unit tests
trace: add infrastructure to augment trace output with additional info
sha1_file: change GIT_TRACE_PACK_ACCESS logging to use trace API
Documentation/git.txt: improve documentation of 'GIT_TRACE*' variables
trace: improve trace performance
trace: remove redundant printf format attribute
trace: consistently name the format parameter
trace: move trace declarations from cache.h to new trace.h
Early part of the "ref transaction" topic.
* rs/ref-transaction-0:
refs.c: change ref_transaction_update() to do error checking and return status
refs.c: remove the onerr argument to ref_transaction_commit
update-ref: use err argument to get error from ref_transaction_commit
refs.c: make update_ref_write update a strbuf on failure
refs.c: make ref_update_reject_duplicates take a strbuf argument for errors
refs.c: log_ref_write should try to return meaningful errno
refs.c: make resolve_ref_unsafe set errno to something meaningful on error
refs.c: commit_packed_refs to return a meaningful errno on failure
refs.c: make remove_empty_directories always set errno to something sane
refs.c: verify_lock should set errno to something meaningful
refs.c: make sure log_ref_setup returns a meaningful errno
refs.c: add an err argument to repack_without_refs
lockfile.c: make lock_file return a meaningful errno on failurei
lockfile.c: add a new public function unable_to_lock_message
refs.c: add a strbuf argument to ref_transaction_commit for error logging
refs.c: allow passing NULL to ref_transaction_free
refs.c: constify the sha arguments for ref_transaction_create|delete|update
refs.c: ref_transaction_commit should not free the transaction
refs.c: remove ref_transaction_rollback
An experiment to use two files (the base file and incremental
changes relative to it) to represent the index to reduce I/O cost
of rewriting a large index when only small part of the working tree
changes.
* nd/split-index: (32 commits)
t1700: new tests for split-index mode
t2104: make sure split index mode is off for the version test
read-cache: force split index mode with GIT_TEST_SPLIT_INDEX
read-tree: note about dropping split-index mode or index version
read-tree: force split-index mode off on --index-output
rev-parse: add --shared-index-path to get shared index path
update-index --split-index: do not split if $GIT_DIR is read only
update-index: new options to enable/disable split index mode
split-index: strip pathname of on-disk replaced entries
split-index: do not invalidate cache-tree at read time
split-index: the reading part
split-index: the writing part
read-cache: mark updated entries for split index
read-cache: save deleted entries in split index
read-cache: mark new entries for split index
read-cache: split-index mode
read-cache: save index SHA-1 after reading
entry.c: update cache_changed if refresh_cache is set in checkout_entry()
cache-tree: mark istate->cache_changed on prime_cache_tree()
cache-tree: mark istate->cache_changed on cache tree update
...
Making errno when returning from resolve_ref_unsafe() meaningful,
which should fix
* a bug in lock_ref_sha1_basic, where it assumes EISDIR
means it failed due to a directory being in the way
Signed-off-by: Ronnie Sahlberg <sahlberg@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Michael Haggerty <mhagger@alum.mit.edu>
Introducing a new unable_to_lock_message helper, which has nicer
semantics than unable_to_lock_error and cleans up lockfile.c a little.
Signed-off-by: Ronnie Sahlberg <sahlberg@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Michael Haggerty <mhagger@alum.mit.edu>
We keep a static counter to set the commit index on newly
allocated objects. However, since we also need to set the
index on any_objects which are converted to commits, let's
make the counter available as a public function.
While we're moving it, let's make sure the counter is
allocated as an unsigned integer to match the index field in
"struct commit".
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Rename cache_def_free to cache_def_clear as it doesn't free the struct
cache_def, but just clears its content.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
'git checkout' fails if a directory is longer than PATH_MAX, because the
lstat_cache in symlinks.c checks if the leading directory exists using
PATH_MAX-bounded string operations.
Remove the limitation by using strbuf instead.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git status", even though it is a read-only operation, tries to
update the index with refreshed lstat(2) info to optimize future
accesses to the working tree opportunistically, but this could
race with a "read-write" operation that modify the index while it
is running. Detect such a race and avoid overwriting the index.
* ym/fix-opportunistic-index-update-race:
read-cache.c: verify index file before we opportunistically update it
wrapper.c: add xpread() similar to xread()
We often represent our strings as a counted string, i.e. a pair of
the pointer to the beginning of the string and its length, and the
string may not be NUL terminated to that length.
To compare a pair of such counted strings, unpack-trees.c and
read-cache.c implement their own name_compare() functions
identically. In addition, the cache_name_compare() function in
read-cache.c is nearly identical. The only difference is when one
string is the prefix of the other string, in which case
name_compare() returns -1/+1 to show which one is longer, and
cache_name_compare() returns the difference of the lengths to show
the same information.
Unify these three functions by using the implementation from
cache_name_compare(). This does not make any difference to the
existing and future callers, as they must be paying attention only
to the sign of the returned value (and not the magnitude) because
the original implementations of these two functions return values
returned by memcmp(3) when the one string is not a prefix of the
other string, and the only thing memcmp(3) guarantees its callers is
the sign of the returned value, not the magnitude.
Signed-off-by: Jeremiah Mahler <jmmahler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Also include direct dependencies (strbuf.h and git-compat-util.h for
__attribute__) so that trace.h can be used independently of cache.h, e.g.
in test programs.
Signed-off-by: Karsten Blees <blees@dcon.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If you have a large work tree but only make changes in a subset, then
$GIT_DIR/index's size should be stable after a while. If you change
branches that touch something else, $GIT_DIR/index's size may grow
large that it becomes as slow as the unified index. Do --split-index
again occasionally to force all changes back to the shared index and
keep $GIT_DIR/index small.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We know the positions of replaced entries via the replace bitmap in
"link" extension, so the "name" path does not have to be stored (it's
still in the shared index). With this, we also have a way to
distinguish additions vs replacements at load time and can catch
broken "link" extensions.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We are sure that after merge_base_index() is done. cache-tree can
still be used with the final index. So don't destroy cache tree.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The large part of this patch just follows CE_ENTRY_CHANGED
marks. replace_index_entry() is updated to update
split_index->base->cache[] as well so base->cache[] does not reference
to a freed entry.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This split-index mode is designed to keep write cost proportional to
the number of changes the user has made, not the size of the work
tree. (Read cost is another matter, to be dealt separately.)
This mode stores index info in a pair of $GIT_DIR/index and
$GIT_DIR/sharedindex.<SHA-1>. sharedindex is large and unchanged over
time while "index" is smaller and updated often. Format details are in
index-format.txt, although not everything is implemented in this
patch.
Shared indexes are not automatically removed, because it's unclear if
the shared index is needed by any (even temporary) indexes by just
looking at it. After a while you'll collect stale shared indexes. The
good news is one shared index is useable for long, until
$GIT_DIR/index becomes too big and sluggish that the new shared index
must be created.
The safest way to clean shared indexes is to turn off split index
mode, so shared files are all garbage, delete them all, then turn on
split index mode again.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Also update SHA-1 after writing. If we do not do that, the second
read_index() will see "initialized" variable already set and not read
.git/index again, which is fine, except istate->sha1 now has a stale
value.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Other fill_stat_cache_info() is on new entries, which should set
CE_ENTRY_ADDED in cache_changed, so we're safe.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
cache entry additions, removals and modifications are separated
out. The rest of changes are still in the catch-all flag
SOMETHING_CHANGED, which would be more specific later.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We're running out of room for in-memory flags. But since b60e188
(Strip namelen out of ce_flags into a ce_namelen field - 2012-07-11),
we copy the namelen (first 12 bits) to ce_namelen field. So those bits
are free to use. Just make sure we do not accidentally write any
in-memory flags back.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function is now only used by write_locked_index(). Move it to
read-cache.c (because read-cache.c will need to be aware of
alternate_index_output later) and unexport it.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
[efl: moved MinGW-specific part to compat/]
[jes: fixed compilation on non-Windows]
Eric Sunshine fixed mingw_offset_1st_component() to return
consistently "foo" for UNC "//machine/share/foo", cf
http://groups.google.com/group/msysgit/browse_thread/thread/c0af578549b5dda0
Author: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Cezary Zawadka <czawadka@gmail.com>
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Stepan Kasal <kasal@ucw.cz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jk/commit-date-approxidate:
commit: accept more date formats for "--date"
commit: print "Date" line when the user has set date
pretty: make show_ident_date public
commit: use split_ident_line to compare author/committer
Read-only operations such as "git status" that internally refreshes
the index write out the refreshed index to the disk to optimize
future accesses to the working tree, but this could race with a
"read-write" operation that modify the index while it is running.
Detect such a race and avoid overwriting the index.
Duy raised a good point that we may need to do the same for the
normal writeout codepath, not just the "opportunistic" update
codepath. While that is true, nobody sane would be running two
simultaneous operations that are clearly write-oriented competing
with each other against the same index file. So in that sense that
can be done as a less urgent follow-up for this topic.
* ym/fix-opportunistic-index-update-race:
read-cache.c: verify index file before we opportunistically update it
wrapper.c: add xpread() similar to xread()
Instead of running N pair-wise diff-trees when inspecting a
N-parent merge, find the set of paths that were touched by walking
N+1 trees in parallel. These set of paths can then be turned into
N pair-wise diff-tree results to be processed through rename
detections and such. And N=2 case nicely degenerates to the usual
2-way diff-tree, which is very nice.
* ks/tree-diff-nway:
mingw: activate alloca
combine-diff: speed it up, by using multiparent diff tree-walker directly
tree-diff: rework diff_tree() to generate diffs for multiparent cases as well
Portable alloca for Git
tree-diff: reuse base str(buf) memory on sub-tree recursion
tree-diff: no need to call "full" diff_tree_sha1 from show_path()
tree-diff: rework diff_tree interface to be sha1 based
tree-diff: diff_tree() should now be static
tree-diff: remove special-case diff-emitting code for empty-tree cases
tree-diff: simplify tree_entry_pathcmp
tree-diff: show_path prototype is not needed anymore
tree-diff: rename compare_tree_entry -> tree_entry_pathcmp
tree-diff: move all action-taking code out of compare_tree_entry()
tree-diff: don't assume compare_tree_entry() returns -1,0,1
tree-diff: consolidate code for emitting diffs and recursion in one place
tree-diff: show_tree() is not needed
tree-diff: no need to pass match to skip_uninteresting()
tree-diff: no need to manually verify that there is no mode change for a path
combine-diff: move changed-paths scanning logic into its own function
combine-diff: move show_log_first logic/action out of paths scanning
When core.commentChar is "auto", the comment char starts with '#' as
in default but if it's already in the prepared message, find another
char in a small subset. This should stop surprises because git strips
some lines unexpectedly.
Note that git is not smart enough to recognize '#' as the comment char
in custom templates and convert it if the final comment char is
different. It thinks '#' lines in custom templates as part of the
commit message. So don't use this with custom templates.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit e208f9c converted error() into a macro to make its
constant return value more apparent to calling code. Commit
5ded807 prevents us using this macro with clang, since
clang's -Wunused-value is smart enough to realize that the
constant "-1" is useless in some contexts.
However, since the last commit puts the constant behind an
inline function call, this is enough to prevent the
-Wunused-value warning on both modern gcc and clang. So we
can now re-enable the macro when compiling with clang.
Tested with clang 3.3, 3.4, and 3.5.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit e208f9c introduced a macro to turn error() calls
into:
(error(), -1)
to make the constant return value more visible to the
calling code (and thus let the compiler make better
decisions about the code).
This works well for code like:
return error(...);
but the "-1" is superfluous in code that just calls error()
without caring about the return value. In older versions of
gcc, that was fine, but gcc 4.9 complains with -Wunused-value.
We can work around this by encapsulating the constant return
value in a static inline function, as gcc specifically
avoids complaining about unused function returns unless the
function has been specifically marked with the
warn_unused_result attribute.
We also use the same trick for config_error_nonbool and
opterror, which learned the same error technique in a469a10.
Reported-by: Felipe Contreras <felipe.contreras@gmail.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We use this function internally to format "Date" lines in
commit logs, but other parts of the code will want it, too.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Before we proceed to opportunistically update the index (often done
by an otherwise read-only operation like "git status" and "git diff"
that internally refreshes the index), we must verify that the
current index file is the same as the one that we read earlier
before we took the lock on it, in order to avoid a possible race.
In the example below git-status does "opportunistic update" and
git-rebase updates the index, but the race can happen in general.
1. process A calls git-rebase (or does anything that uses the index)
2. process A applies 1st commit
3. process B calls git-status (or does anything that updates the index)
4. process B reads index
5. process A applies 2nd commit
6. process B takes the lock, then overwrites process A's changes.
7. process A applies 3rd commit
As an end result the 3rd commit will have a revert of the 2nd commit.
When process B takes the lock, it needs to make sure that the index
hasn't changed since step 4.
Signed-off-by: Yiannis Marangos <yiannis.marangos@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously diff_tree(), which is now named ll_diff_tree_sha1(), was
generating diff_filepair(s) for two trees t1 and t2, and that was
usually used for a commit as t1=HEAD~, and t2=HEAD - i.e. to see changes
a commit introduces.
In Git, however, we have fundamentally built flexibility in that a
commit can have many parents - 1 for a plain commit, 2 for a simple merge,
but also more than 2 for merging several heads at once.
For merges there is a so called combine-diff, which shows diff, a merge
introduces by itself, omitting changes done by any parent. That works
through first finding paths, that are different to all parents, and then
showing generalized diff, with separate columns for +/- for each parent.
The code lives in combine-diff.c .
There is an impedance mismatch, however, in that a commit could
generally have any number of parents, and that while diffing trees, we
divide cases for 2-tree diffs and more-than-2-tree diffs. I mean there
is no special casing for multiple parents commits in e.g.
revision-walker .
That impedance mismatch *hurts* *performance* *badly* for generating
combined diffs - in "combine-diff: optimize combine_diff_path
sets intersection" I've already removed some slowness from it, but from
the timings provided there, it could be seen, that combined diffs still
cost more than an order of magnitude more cpu time, compared to diff for
usual commits, and that would only be an optimistic estimate, if we take
into account that for e.g. linux.git there is only one merge for several
dozens of plain commits.
That slowness comes from the fact that currently, while generating
combined diff, a lot of time is spent computing diff(commit,commit^2)
just to only then intersect that huge diff to almost small set of files
from diff(commit,commit^1).
That's because at present, to compute combine-diff, for first finding
paths, that "every parent touches", we use the following combine-diff
property/definition:
D(A,P1...Pn) = D(A,P1) ^ ... ^ D(A,Pn) (w.r.t. paths)
where
D(A,P1...Pn) is combined diff between commit A, and parents Pi
and
D(A,Pi) is usual two-tree diff Pi..A
So if any of that D(A,Pi) is huge, tracting 1 n-parent combine-diff as n
1-parent diffs and intersecting results will be slow.
And usually, for linux.git and other topic-based workflows, that
D(A,P2) is huge, because, if merge-base of A and P2, is several dozens
of merges (from A, via first parent) below, that D(A,P2) will be diffing
sum of merges from several subsystems to 1 subsystem.
The solution is to avoid computing n 1-parent diffs, and to find
changed-to-all-parents paths via scanning A's and all Pi's trees
simultaneously, at each step comparing their entries, and based on that
comparison, populate paths result, and deduce we could *skip*
*recursing* into subdirectories, if at least for 1 parent, sha1 of that
dir tree is the same as in A. That would save us from doing significant
amount of needless work.
Such approach is very similar to what diff_tree() does, only there we
deal with scanning only 2 trees simultaneously, and for n+1 tree, the
logic is a bit more complex:
D(T,P1...Pn) calculation scheme
-------------------------------
D(T,P1...Pn) = D(T,P1) ^ ... ^ D(T,Pn) (regarding resulting paths set)
D(T,Pj) - diff between T..Pj
D(T,P1...Pn) - combined diff from T to parents P1,...,Pn
We start from all trees, which are sorted, and compare their entries in
lock-step:
T P1 Pn
- - -
|t| |p1| |pn|
|-| |--| ... |--| imin = argmin(p1...pn)
| | | | | |
|-| |--| |--|
|.| |. | |. |
. . .
. . .
at any time there could be 3 cases:
1) t < p[imin];
2) t > p[imin];
3) t = p[imin].
Schematic deduction of what every case means, and what to do, follows:
1) t < p[imin] -> ∀j t ∉ Pj -> "+t" ∈ D(T,Pj) -> D += "+t"; t↓
2) t > p[imin]
2.1) ∃j: pj > p[imin] -> "-p[imin]" ∉ D(T,Pj) -> D += ø; ∀ pi=p[imin] pi↓
2.2) ∀i pi = p[imin] -> pi ∉ T -> "-pi" ∈ D(T,Pi) -> D += "-p[imin]"; ∀i pi↓
3) t = p[imin]
3.1) ∃j: pj > p[imin] -> "+t" ∈ D(T,Pj) -> only pi=p[imin] remains to investigate
3.2) pi = p[imin] -> investigate δ(t,pi)
|
|
v
3.1+3.2) looking at δ(t,pi) ∀i: pi=p[imin] - if all != ø ->
⎧δ(t,pi) - if pi=p[imin]
-> D += ⎨
⎩"+t" - if pi>p[imin]
in any case t↓ ∀ pi=p[imin] pi↓
~
For comparison, here is how diff_tree() works:
D(A,B) calculation scheme
-------------------------
A B
- -
|a| |b| a < b -> a ∉ B -> D(A,B) += +a a↓
|-| |-| a > b -> b ∉ A -> D(A,B) += -b b↓
| | | | a = b -> investigate δ(a,b) a↓ b↓
|-| |-|
|.| |.|
. .
. .
~~~~~~~~
This patch generalizes diff tree-walker to work with arbitrary number of
parents as described above - i.e. now there is a resulting tree t, and
some parents trees tp[i] i=[0..nparent). The generalization builds on
the fact that usual diff
D(A,B)
is by definition the same as combined diff
D(A,[B]),
so if we could rework the code for common case and make it be not slower
for nparent=1 case, usual diff(t1,t2) generation will not be slower, and
multiparent diff tree-walker would greatly benefit generating
combine-diff.
What we do is as follows:
1) diff tree-walker ll_diff_tree_sha1() is internally reworked to be
a paths generator (new name diff_tree_paths()), with each generated path
being `struct combine_diff_path` with info for path, new sha1,mode and for
every parent which sha1,mode it was in it.
2) From that info, we can still generate usual diff queue with
struct diff_filepairs, via "exporting" generated
combine_diff_path, if we know we run for nparent=1 case.
(see emit_diff() which is now named emit_diff_first_parent_only())
3) In order for diff_can_quit_early(), which checks
DIFF_OPT_TST(opt, HAS_CHANGES))
to work, that exporting have to be happening not in bulk, but
incrementally, one diff path at a time.
For such consumers, there is a new callback in diff_options
introduced:
->pathchange(opt, struct combine_diff_path *)
which, if set to !NULL, is called for every generated path.
(see new compat ll_diff_tree_sha1() wrapper around new paths
generator for setup)
4) The paths generation itself, is reworked from previous
ll_diff_tree_sha1() code according to "D(A,P1...Pn) calculation
scheme" provided above:
On the start we allocate [nparent] arrays in place what was
earlier just for one parent tree.
then we just generalize loops, and comparison according to the
algorithm.
Some notes(*):
1) alloca(), for small arrays, is used for "runs not slower for
nparent=1 case than before" goal - if we change it to xmalloc()/free()
the timings get ~1% worse. For alloca() we use just-introduced
xalloca/xalloca_free compatibility wrappers, so it should not be a
portability problem.
2) For every parent tree, we need to keep a tag, whether entry from that
parent equals to entry from minimal parent. For performance reasons I'm
keeping that tag in entry's mode field in unused bit - see S_IFXMIN_NEQ.
Not doing so, we'd need to alloca another [nparent] array, which hurts
performance.
3) For emitted paths, memory could be reused, if we know the path was
processed via callback and will not be needed later. We use efficient
hand-made realloc-style path_appendnew(), that saves us from ~1-1.5%
of potential additional slowdown.
4) goto(s) are used in several places, as the code executes a little bit
faster with lowered register pressure.
Also
- we should now check for FIND_COPIES_HARDER not only when two entries
names are the same, and their hashes are equal, but also for a case,
when a path was removed from some of all parents having it.
The reason is, if we don't, that path won't be emitted at all (see
"a > xi" case), and we'll just skip it, and FIND_COPIES_HARDER wants
all paths - with diff or without - to be emitted, to be later analyzed
for being copies sources.
The new check is only necessary for nparent >1, as for nparent=1 case
xmin_eqtotal always =1 =nparent, and a path is always added to diff as
removal.
~~~~~~~~
Timings for
# without -c, i.e. testing only nparent=1 case
`git log --raw --no-abbrev --no-renames`
before and after the patch are as follows:
navy.git linux.git v3.10..v3.11
before 0.611s 1.889s
after 0.619s 1.907s
slowdown 1.3% 0.9%
This timings show we did no harm to usual diff(tree1,tree2) generation.
From the table we can see that we actually did ~1% slowdown, but I think
I've "earned" that 1% in the previous patch ("tree-diff: reuse base
str(buf) memory on sub-tree recursion", HEAD~~) so for nparent=1 case,
net timings stays approximately the same.
The output also stayed the same.
(*) If we revert 1)-4) to more usual techniques, for nparent=1 case,
we'll get ~2-2.5% of additional slowdown, which I've tried to avoid, as
"do no harm for nparent=1 case" rule.
For linux.git, combined diff will run an order of magnitude faster and
appropriate timings will be provided in the next commit, as we'll be
taking advantage of the new diff tree-walker for combined-diff
generation there.
P.S. and combined diff is not some exotic/for-play-only stuff - for
example for a program I write to represent Git archives as readonly
filesystem, there is initial scan with
`git log --reverse --raw --no-abbrev --no-renames -c`
to extract log of what was created/changed when, as a result building a
map
{} sha1 -> in which commit (and date) a content was added
that `-c` means also show combined diff for merges, and without them, if
a merge is non-trivial (merges changes from two parents with both having
separate changes to a file), or an evil one, the map will not be full,
i.e. some valid sha1 would be absent from it.
That case was my initial motivation for combined diffs speedup.
Signed-off-by: Kirill Smelkov <kirr@mns.spb.ru>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
