The file 'dir/subdir/file' can only be modified if its leading
directories 'dir' and 'dir/subdir' are modified as well.
So when checking modified path Bloom filters looking for commits
modifying a path with multiple path components, then check not only
the full path in the Bloom filters, but all its leading directories as
well. Take care to check these paths in "deepest first" order,
because it's the full path that is least likely to be modified, and
the Bloom filter queries can short circuit sooner.
This can significantly reduce the average false positive rate, by
about an order of magnitude or three(!), and can further speed up
pathspec-limited revision walks. The table below compares the average
false positive rate and runtime of
git rev-list HEAD -- "$path"
before and after this change for 5000+ randomly* selected paths from
each repository:
Average false Average Average
positive rate runtime runtime
before after before after difference
------------------------------------------------------------------
git 3.220% 0.7853% 0.0558s 0.0387s -30.6%
linux 2.453% 0.0296% 0.1046s 0.0766s -26.8%
tensorflow 2.536% 0.6977% 0.0594s 0.0420s -29.2%
*Path selection was done with the following pipeline:
git ls-tree -r --name-only HEAD | sort -R | head -n 5000
The improvements in runtime are much smaller than the improvements in
average false positive rate, as we are clearly reaching diminishing
returns here. However, all these timings depend on that accessing
tree objects is reasonably fast (warm caches). If we had a partial
clone and the tree objects had to be fetched from a promisor remote,
e.g.:
$ git clone --filter=tree:0 --bare file://.../webkit.git webkit.notrees.git
$ git -C webkit.git -c core.modifiedPathBloomFilters=1 \
commit-graph write --reachable
$ cp webkit.git/objects/info/commit-graph webkit.notrees.git/objects/info/
$ git -C webkit.notrees.git -c core.modifiedPathBloomFilters=1 \
rev-list HEAD -- "$path"
then checking all leading path component can reduce the runtime from
over an hour to a few seconds (and this is with the clone and the
promisor on the same machine).
This adjusts the tracing values in t4216-log-bloom.sh, which provides a
concrete way to notice the improvement.
Helped-by: Taylor Blau <me@ttaylorr.com>
Helped-by: René Scharfe <l.s.r@web.de>
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The prepare_to_use_bloom_filter() method was not intended to be called
on an empty pathspec. However, 'git log -- .' and 'git log' are subtly
different: the latter reports all commits while the former will simplify
commits that do not change the root tree.
This means that the path used to construct the bloom_key might be empty,
and that value is not added to the Bloom filter during construction.
That means that the results are likely incorrect!
To resolve the issue, be careful about the length of the path and stop
filling Bloom filters. To be completely sure we do not use them, drop
the pointer to the bloom_filter_settings from the commit-graph. That
allows our test to look at the trace2 logs to verify no Bloom filter
statistics are reported.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Here, four spaces were used instead of tab characters.
Reported-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_bloom_filter() method is a bit complicated in some parts where
it does not need to be. In particular, it needs to return a NULL filter
only when compute_if_not_present is zero AND the filter data cannot be
loaded from a commit-graph file. This currently happens by accident
because the commit-graph does not load changed-path Bloom filters from
an existing commit-graph when writing a new one. This will change in a
later patch.
Also clean up some style issues while we are here.
One side-effect of returning a NULL filter is that the filters that are
reported as "too large" will now be reported as NULL insead of length
zero. This case was not properly covered before, so add a test. Further,
remote the counting of the zero-length filters from revision.c and the
trace2 logs.
Helped-by: René Scharfe <l.s.r@web.de>
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
ll_diff_tree_oid() has only ever returned 0 [1], so it's return value
is basically useless. It's only caller diff_tree_oid() has only ever
returned the return value of ll_diff_tree_oid() as-is [2], so its
return value is just as useless. Most of diff_tree_oid()'s callers
simply ignore its return value, except:
- diff_root_tree_oid() is a thin wrapper around diff_tree_oid() and
returns with its return value, but all of diff_root_tree_oid()'s
callers ignore its return value.
- rev_compare_tree() and rev_same_tree_as_empty() do look at the
return value in a condition, but, since the return value is always
0, the former's < 0 condition is never fulfilled, while the
latter's >= 0 condition is always fulfilled.
So let's drop the return value of ll_diff_tree_oid(), diff_tree_oid()
and diff_root_tree_oid(), and drop those conditions from
rev_compare_tree() and rev_same_tree_as_empty() as well.
[1] ll_diff_tree_oid() and its ancestors have been returning only 0
ever since it was introduced as diff_tree() in 9174026cfe (Add
"diff-tree" program to show which files have changed between two
trees., 2005-04-09).
[2] diff_tree_oid() traces back to diff-tree.c:main() in 9174026cfe as
well.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous changes to the line-log machinery focused on making the
first result appear faster. This was achieved by no longer walking the
entire commit history before returning the early results. There is still
another way to improve the performance: walk most commits much faster.
Let's use the changed-path Bloom filters to reduce time spent computing
diffs.
Since the line-log computation requires opening blobs and checking the
content-diff, there is still a lot of necessary computation that cannot
be replaced with changed-path Bloom filters. The part that we can reduce
is most effective when checking the history of a file that is deep in
several directories and those directories are modified frequently. In
this case, the computation to check if a commit is TREESAME to its first
parent takes a large fraction of the time. That is ripe for improvement
with changed-path Bloom filters.
We must ensure that prepare_to_use_bloom_filters() is called in
revision.c so that the bloom_filter_settings are loaded into the struct
rev_info from the commit-graph. Of course, some cases are still
forbidden, but in the line-log case the pathspec is provided in a
different way than normal.
Since multiple paths and segments could be requested, we compute the
struct bloom_key data dynamically during the commit walk. This could
likely be improved, but adds code complexity that is not valuable at
this time.
There are two cases to care about: merge commits and "ordinary" commits.
Merge commits have multiple parents, but if we are TREESAME to our first
parent in every range, then pass the blame for all ranges to the first
parent. Ordinary commits have the same condition, but each is done
slightly differently in the process_ranges_[merge|ordinary]_commit()
methods. By checking if the changed-path Bloom filter can guarantee
TREESAME, we can avoid that tree-diff cost. If the filter says "probably
changed", then we need to run the tree-diff and then the blob-diff if
there was a real edit.
The Linux kernel repository is a good testing ground for the performance
improvements claimed here. There are two different cases to test. The
first is the "entire history" case, where we output the entire history
to /dev/null to see how long it would take to compute the full line-log
history. The second is the "first result" case, where we find how long
it takes to show the first value, which is an indicator of how quickly a
user would see responses when waiting at a terminal.
To test, I selected the paths that were changed most frequently in the
top 10,000 commits using this command (stolen from StackOverflow [1]):
git log --pretty=format: --name-only -n 10000 | sort | \
uniq -c | sort -rg | head -10
which results in
121 MAINTAINERS
63 fs/namei.c
60 arch/x86/kvm/cpuid.c
59 fs/io_uring.c
58 arch/x86/kvm/vmx/vmx.c
51 arch/x86/kvm/x86.c
45 arch/x86/kvm/svm.c
42 fs/btrfs/disk-io.c
42 Documentation/scsi/index.rst
(along with a bogus first result). It appears that the path
arch/x86/kvm/svm.c was renamed, so we ignore that entry. This leaves the
following results for the real command time:
| | Entire History | First Result |
| Path | Before | After | Before | After |
|------------------------------|--------|--------|--------|--------|
| MAINTAINERS | 4.26 s | 3.87 s | 0.41 s | 0.39 s |
| fs/namei.c | 1.99 s | 0.99 s | 0.42 s | 0.21 s |
| arch/x86/kvm/cpuid.c | 5.28 s | 1.12 s | 0.16 s | 0.09 s |
| fs/io_uring.c | 4.34 s | 0.99 s | 0.94 s | 0.27 s |
| arch/x86/kvm/vmx/vmx.c | 5.01 s | 1.34 s | 0.21 s | 0.12 s |
| arch/x86/kvm/x86.c | 2.24 s | 1.18 s | 0.21 s | 0.14 s |
| fs/btrfs/disk-io.c | 1.82 s | 1.01 s | 0.06 s | 0.05 s |
| Documentation/scsi/index.rst | 3.30 s | 0.89 s | 1.46 s | 0.03 s |
It is worth noting that the least speedup comes for the MAINTAINERS file
which is
* edited frequently,
* low in the directory heirarchy, and
* quite a large file.
All of those points lead to spending more time doing the blob diff and
less time doing the tree diff. Still, we see some improvement in that
case and significant improvement in other cases. A 2-4x speedup is
likely the more typical case as opposed to the small 5% change for that
file.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous patch made it possible to perform line-level filtering
during history traversal instead of in an expensive preprocessing
step, but it still requires some simpler preprocessing steps, notably
topo-ordering. However, nowadays we have commit-graphs storing
generation numbers, which make it possible to incrementally traverse
the history in topological order, without the preparatory limit_list()
and sort_in_topological_order() steps; see b45424181e (revision.c:
generation-based topo-order algorithm, 2018-11-01).
This patch combines the two, so we can do both the topo-ordering and
the line-level filtering during history traversal, eliminating even
those simpler preprocessing steps, and thus further reducing the delay
before showing the first commit modifying the given line range.
The 'revs->limited' flag plays the central role in this, because, due
to limitations of the current implementation, the generation
number-based topo-ordering is only enabled when this flag remains
unset. Line-level log, however, always sets this flag in
setup_revisions() ever since the feature was introduced in 12da1d1f6f
(Implement line-history search (git log -L), 2013-03-28). The reason
for setting 'limited' is unclear, though, because the line-level log
itself doesn't directly depend on it, and it doesn't affect how the
limit_list() function limits the revision range. However, there is an
indirect dependency: the line-level log requires topo-ordering, and
the "traditional" sort_in_topological_order() requires an already
limited commit list since e6c3505b44 (Make sure we generate the whole
commit list before trying to sort it topologically, 2005-07-06). The
new, generation numbers-based topo-ordering doesn't require a limited
commit list anymore.
So don't set 'revs->limited' for line-level log, unless it is really
necessary, namely:
- The user explicitly requested parent rewriting, because that is
still done in the line_log_filter() preprocessing step (see
previous patch), which requires sort_in_topological_order() and in
turn limit_list() as well.
- A commit-graph file is not available or it doesn't yet contain
generation numbers. In these cases we had to fall back on
sort_in_topological_order() and in turn limit_list(). The
existing condition with generation_numbers_enabled() has already
ensured that the 'limited' flag is set in these cases; this patch
just makes sure that the line-level log sets 'revs->topo_order'
before that condition.
While the reduced delay before showing the first commit is measurable
in git.git, it takes a bigger repository to make it clearly noticable.
In both cases below the line ranges were chosen so that they were
modified rather close to the starting revisions, so the effect of this
change is most noticable.
# git.git
$ time git --no-pager log -L:read_alternate_refs:sha1-file.c -1 v2.23.0
Before:
real 0m0.107s
user 0m0.091s
sys 0m0.013s
After:
real 0m0.058s
user 0m0.050s
sys 0m0.005s
# linux.git
$ time git --no-pager log \
-L:build_restore_work_registers:arch/mips/mm/tlbex.c -1 v5.2
Before:
real 0m1.129s
user 0m1.061s
sys 0m0.069s
After:
real 0m0.096s
user 0m0.087s
sys 0m0.009s
Additional testing by Derrick Stolee: Since this patch improves
the performance for the first result, I repeated the experiment
from the previous patch on the Linux kernel repository, reporting
real time here:
Command: git log -L 100,200:MAINTAINERS -n 1 >/dev/null
Before: 0.71 s
After: 0.05 s
Now, we have dropped the full topo-order of all ~910,000 commits
before reporting the first result. The remaining performance
improvements then are:
1. Update the parent-rewriting logic to be incremental similar to
how "git log --graph" behaves.
2. Use changed-path Bloom filters to reduce the time spend in the
tree-diff to see if the path(s) changed.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The current line-level log implementation performs a preprocessing
step in prepare_revision_walk(), during which the line_log_filter()
function filters and rewrites history to keep only commits modifying
the given line range. This preprocessing affects both responsiveness
and correctness:
- Git doesn't produce any output during this preprocessing step.
Checking whether a commit modified the given line range is
somewhat expensive, so depending on the size of the given revision
range this preprocessing can result in a significant delay before
the first commit is shown.
- Limiting the number of displayed commits (e.g. 'git log -3 -L...')
doesn't limit the amount of work during preprocessing, because
that limit is applied during history traversal. Alas, by that
point this expensive preprocessing step has already churned
through the whole revision range to find all commits modifying the
revision range, even though only a few of them need to be shown.
- It rewrites parents, with no way to turn it off. Without the user
explicitly requesting parent rewriting any parent object ID shown
should be that of the immediate parent, just like in case of a
pathspec-limited history traversal without parent rewriting.
However, after that preprocessing step rewrote history, the
subsequent "regular" history traversal (i.e. get_revision() in a
loop) only sees commits modifying the given line range.
Consequently, it can only show the object ID of the last ancestor
that modified the given line range (which might happen to be the
immediate parent, but many-many times it isn't).
This patch addresses both the correctness and, at least for the common
case, the responsiveness issues by integrating line-level log
filtering into the regular revision walking machinery:
- Make process_ranges_arbitrary_commit(), the static function in
'line-log.c' deciding whether a commit modifies the given line
range, public by removing the static keyword and adding the
'line_log_' prefix, so it can be called from other parts of the
revision walking machinery.
- If the user didn't explicitly ask for parent rewriting (which, I
believe, is the most common case):
- Call this now-public function during regular history traversal,
namely from get_commit_action() to ignore any commits not
modifying the given line range.
Note that while this check is relatively expensive, it must be
performed before other, much cheaper conditions, because the
tracked line range must be adjusted even when the commit will
end up being ignored by other conditions.
- Skip the line_log_filter() call, i.e. the expensive
preprocessing step, in prepare_revision_walk(), because, thanks
to the above points, the revision walking machinery is now able
to filter out commits not modifying the given line range while
traversing history.
This way the regular history traversal sees the unmodified
history, and is therefore able to print the object ids of the
immediate parents of the listed commits. The eliminated
preprocessing step can greatly reduce the delay before the first
commit is shown, see the numbers below.
- However, if the user did explicitly ask for parent rewriting via
'--parents' or a similar option, then stick with the current
implementation for now, i.e. perform that expensive filtering and
history rewriting in the preprocessing step just like we did
before, leaving the initial delay as long as it was.
I tried to integrate line-level log filtering with parent rewriting
into the regular history traversal, but, unfortunately, several
subtleties resisted... :) Maybe someday we'll figure out how to do
that, but until then at least the simple and common (i.e. without
parent rewriting) 'git log -L:func:file' commands can benefit from the
reduced delay.
This change makes the failing 'parent oids without parent rewriting'
test in 't4211-line-log.sh' succeed.
The reduced delay is most noticable when there's a commit modifying
the line range near the tip of a large-ish revision range:
# no parent rewriting requested, no commit-graph present
$ time git --no-pager log -L:read_alternate_refs:sha1-file.c -1 v2.23.0
Before:
real 0m9.570s
user 0m9.494s
sys 0m0.076s
After:
real 0m0.718s
user 0m0.674s
sys 0m0.044s
A significant part of the remaining delay is spent reading and parsing
commit objects in limit_list(). With the help of the commit-graph we
can eliminate most of that reading and parsing overhead, so here are
the timing results of the same command as above, but this time using
the commit-graph:
Before:
real 0m8.874s
user 0m8.816s
sys 0m0.057s
After:
real 0m0.107s
user 0m0.091s
sys 0m0.013s
The next patch will further reduce the remaining delay.
To be clear: this patch doesn't actually optimize the line-level log,
but merely moves most of the work from the preprocessing step to the
history traversal, so the commits modifying the line range can be
shown as soon as they are processed, and the traversal can be
terminated as soon as the given number of commits are shown.
Consequently, listing the full history of a line range, potentially
all the way to the root commit, will take the same time as before (but
at least the user might start reading the output earlier).
Furthermore, if the most recent commit modifying the line range is far
away from the starting revision, then that initial delay will still be
significant.
Additional testing by Derrick Stolee: In the Linux kernel repository,
the MAINTAINERS file was changed ~3,500 times across the ~915,000
commits. In addition to that edit frequency, the file itself is quite
large (~18,700 lines). This means that a significant portion of the
computation is taken up by computing the patch-diff of the file. This
patch improves the real time it takes to output the first result quite
a bit:
Command: git log -L 100,200:MAINTAINERS -n 1 >/dev/null
Before: 3.88 s
After: 0.71 s
If we drop the "-n 1" in the command, then there is no change in
end-to-end process time. This is because the command still needs to
walk the entire commit history, which negates the point of this
patch. This is expected.
As a note for future reference, the ~4.3 seconds in the old code
spends ~2.6 seconds computing the patch-diffs, and the rest of the
time is spent walking commits and computing diffs for which paths
changed at each commit. The changed-path Bloom filters could improve
the end-to-end computation time (i.e. no "-n 1" in the command).
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The changed-path Bloom filters work only when we can compute an
explicit Bloom filter key in advance. When a pathspec is given
that allows case-insensitive checks or wildcard matching, we
must disable the Bloom filter performance checks.
By checking the pathspec in prepare_to_use_bloom_filters(), we
avoid setting up the Bloom filter data and thus revert to the
usual logic.
Before this change, the following tests would fail*:
t6004-rev-list-path-optim.sh (Tests 6-7)
t6130-pathspec-noglob.sh (Tests 3-6)
t6131-pathspec-icase.sh (Tests 3-5)
*These tests would fail when using GIT_TEST_COMMIT_GRAPH and
GIT_TEST_COMMIT_GRAPH_BLOOM_FILTERS except that the latter
environment variable was not set up correctly to write the changed-
path Bloom filters in the test suite. That will be fixed in the
next change.
Helped-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add trace2 statistics around Bloom filter usage and behavior
for 'git log -- path' commands that are hoping to benefit from
the presence of computed changed paths Bloom filters.
These statistics are great for performance analysis work and
for formal testing, which we will see in the commit following
this one.
Helped-by: Derrick Stolee <dstolee@microsoft.com
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Helped-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Garima Singh <garima.singh@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Revision walk will now use Bloom filters for commits to speed up
revision walks for a particular path (for computing history for
that path), if they are present in the commit-graph file.
We load the Bloom filters during the prepare_revision_walk step,
currently only when dealing with a single pathspec. Extending
it to work with multiple pathspecs can be explored and built on
top of this series in the future.
While comparing trees in rev_compare_trees(), if the Bloom filter
says that the file is not different between the two trees, we don't
need to compute the expensive diff. This is where we get our
performance gains. The other response of the Bloom filter is '`:maybe',
in which case we fall back to the full diff calculation to determine
if the path was changed in the commit.
We do not try to use Bloom filters when the '--walk-reflogs' option
is specified. The '--walk-reflogs' option does not walk the commit
ancestry chain like the rest of the options. Incorporating the
performance gains when walking reflog entries would add more
complexity, and can be explored in a later series.
Performance Gains:
We tested the performance of `git log -- <path>` on the git repo, the linux
and some internal large repos, with a variety of paths of varying depths.
On the git and linux repos:
- we observed a 2x to 5x speed up.
On a large internal repo with files seated 6-10 levels deep in the tree:
- we observed 10x to 20x speed ups, with some paths going up to 28 times
faster.
Helped-by: Derrick Stolee <dstolee@microsoft.com
Helped-by: SZEDER Gábor <szeder.dev@gmail.com>
Helped-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Garima Singh <garima.singh@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git format-patch" can take a set of configured format.notes values
to specify which notes refs to use in the log message part of the
output. The behaviour of this was not consistent with multiple
--notes command line options, which has been corrected.
* dl/format-patch-notes-config-fixup:
notes.h: fix typos in comment
notes: break set_display_notes() into smaller functions
config/format.txt: clarify behavior of multiple format.notes
format-patch: move git_config() before repo_init_revisions()
format-patch: use --notes behavior for format.notes
notes: extract logic into set_display_notes()
notes: create init_display_notes() helper
notes: rename to load_display_notes()
In 8164c961e1 (format-patch: use --notes behavior for format.notes,
2019-12-09), we introduced set_display_notes() which was a monolithic
function with three mutually exclusive branches. Break the function up
into three small and simple functions that each are only responsible for
one task.
This family of functions accepts an `int *show_notes` instead of
returning a value suitable for assignment to `show_notes`. This is for
two reasons. First of all, this guarantees that the external
`show_notes` variable changes in lockstep with the
`struct display_notes_opt`. Second, this prompts future developers to be
careful about doing something meaningful with this value. In fact, a
NULL check is intentionally omitted because causing a segfault here
would tell the future developer that they are meant to use the value for
something meaningful.
One alternative was making the family of functions accept a
`struct rev_info *` instead of the `struct display_notes_opt *`, since
the former contains the `show_notes` field as well. This does not work
because we have to call git_config() before repo_init_revisions().
However, if we had a `struct rev_info`, we'd need to initialize it before
it gets assigned values from git_config(). As a result, we break the
circular dependency by having standalone `int show_notes` and
`struct display_notes_opt notes_opt` variables which temporarily hold
values from git_config() until the values are copied over to `rev`.
To implement this change, we need to get a pointer to
`rev_info::show_notes`. Unfortunately, this is not possible with
bitfields and only direct-assignment is possible. Change
`rev_info::show_notes` to a non-bitfield int so that we can get its
address.
Signed-off-by: Denton Liu <liu.denton@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git log" family learned "--pretty=reference" that gives the name
of a commit in the format that is often used to refer to it in log
messages.
* dl/pretty-reference:
SubmittingPatches: use `--pretty=reference`
pretty: implement 'reference' format
pretty: add struct cmt_fmt_map::default_date_mode_type
pretty: provide short date format
t4205: cover `git log --reflog -z` blindspot
pretty.c: inline initalize format_context
revision: make get_revision_mark() return const pointer
completion: complete `tformat:` pretty format
SubmittingPatches: remove dq from commit reference
pretty-formats.txt: use generic terms for hash
SubmittingPatches: use generic terms for hash
Instead of open coding the logic that tweaks the variables in
`struct display_notes_opt` within handle_revision_opt(), abstract away the
logic into set_display_notes() so that it can be reused.
Signed-off-by: Denton Liu <liu.denton@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We currently open code the initialization for revs->notes_opt. Abstract
this away into a helper function so that the logic can be reused in a
future commit.
This is slightly wasteful as we memset the struct twice but this is only
run once so it shouldn't have any major effect.
Signed-off-by: Denton Liu <liu.denton@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The revision walking machinery uses resources like per-object flag
bits that need to be reset before a new iteration of walking
begins, but the resources related to topological walk were not
cleared correctly, which has been corrected.
* mh/clear-topo-walk-upon-reset:
revision: free topo_walk_info before creating a new one in init_topo_walk
revision: clear the topo-walk flags in reset_revision_walk
init_topo_walk doesn't reuse an existing topo_walk_info, and currently
leaks the one that might exist on the current rev_info if it was already
used for a topo walk beforehand.
Signed-off-by: Mike Hommey <mh@glandium.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Not doing so can lead to wrong topo-walks when using the revision walk API
consecutively.
Signed-off-by: Mike Hommey <mh@glandium.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
get_revision_mark() used to return a `char *`, even though all of the
strings it was returning were string literals. Make get_revision_mark()
return a `const char *` so that callers won't be tempted to modify the
returned string.
Signed-off-by: Denton Liu <liu.denton@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A few simplification and bugfixes to PCRE interface.
* ab/pcre-jit-fixes:
grep: under --debug, show whether PCRE JIT is enabled
grep: do not enter PCRE2_UTF mode on fixed matching
grep: stess test PCRE v2 on invalid UTF-8 data
grep: create a "is_fixed" member in "grep_pat"
grep: consistently use "p->fixed" in compile_regexp()
grep: stop using a custom JIT stack with PCRE v1
grep: stop "using" a custom JIT stack with PCRE v2
grep: remove overly paranoid BUG(...) code
grep: use PCRE v2 for optimized fixed-string search
grep: remove the kwset optimization
grep: drop support for \0 in --fixed-strings <pattern>
grep: make the behavior for NUL-byte in patterns sane
grep tests: move binary pattern tests into their own file
grep tests: move "grep binary" alongside the rest
grep: inline the return value of a function call used only once
t4210: skip more command-line encoding tests on MinGW
grep: don't use PCRE2?_UTF8 with "log --encoding=<non-utf8>"
log tests: test regex backends in "--encode=<enc>" tests
"git log --decorate-refs-exclude=<pattern>" was incorrectly
overruled when the "--simplify-by-decoration" option is used, which
has been corrected.
* rs/simplify-by-deco-with-deco-refs-exclude:
log-tree: call load_ref_decorations() in get_name_decoration()
log: test --decorate-refs-exclude with --simplify-by-decoration
Since these macros already take a `keyvar' pointer of a known type,
we can rely on OFFSETOF_VAR to get the correct offset without
relying on non-portable `__typeof__' and `offsetof'.
Argument order is also rearranged, so `keyvar' and `member' are
sequential as they are used as: `keyvar->member'
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
While we cannot rely on a `__typeof__' operator being portable
to use with `offsetof'; we can calculate the pointer offset
using an existing pointer and the address of a member using
pointer arithmetic for compilers without `__typeof__'.
This allows us to simplify usage of hashmap iterator macros
by not having to specify a type when a pointer of that type
is already given.
In the future, list iterator macros (e.g. list_for_each_entry)
may also be implemented using OFFSETOF_VAR to save hackers the
trouble of using container_of/list_entry macros and without
relying on non-portable `__typeof__'.
v3: use `__typeof__' to avoid clang warnings
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
`hashmap_free_entries' behaves like `container_of' and passes
the offset of the hashmap_entry struct to the internal
`hashmap_free_' function, allowing the function to free any
struct pointer regardless of where the hashmap_entry field
is located.
`hashmap_free' no longer takes any arguments aside from
the hashmap itself.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Inspired by list_for_each_entry in the Linux kernel.
Once again, these are somewhat compromised usability-wise
by compilers lacking __typeof__ support.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Another step in eliminating the requirement of hashmap_entry
being the first member of a struct.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update callers to use hashmap_get_entry, hashmap_get_entry_from_hash
or container_of as appropriate.
This is another step towards eliminating the requirement of
hashmap_entry being the first field in a struct.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is less error-prone than "void *" as the compiler now
detects invalid types being passed.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is less error-prone than "const void *" as the compiler
now detects invalid types being passed.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
C compilers do type checking to make life easier for us. So
rely on that and update all hashmap_entry_init callers to take
"struct hashmap_entry *" to avoid future bugs while improving
safety and readability.
Signed-off-by: Eric Wong <e@80x24.org>
Reviewed-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Load a default set of ref name decorations at the first lookup. This
frees direct and indirect callers from doing so. They can still do it
if they want to use a filter or are interested in full decorations
instead of the default short ones -- the first load_ref_decorations()
call wins.
This means that the load in builtin/log.c::cmd_log_init_finish() is
respected even if --simplify-by-decoration is given, as the previously
dominating earlier load in handle_revision_opt() is gone. So a filter
given with --decorate-refs-exclude is used for simplification in that
case, as expected.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a function for accessing the ID of the object referenced by a tag
safely, i.e. without causing a segfault when encountering a broken tag
where ->tagged is NULL.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There's currently no robust way to tell Git that a particular option is
meant to be a revision, and not an option. So if you have a branch
"refs/heads/--foo", you cannot just say:
git rev-list --foo
You can say:
git rev-list refs/heads/--foo
But that breaks down if you don't know the refname, and in particular if
you're a script passing along a value from elsewhere. In most programs,
you can use "--" to end option parsing, like this:
some-prog -- "$revision"
But that doesn't work for the revision parser, because "--" is already
meaningful there: it separates revisions from pathspecs. So we need some
other marker to separate options from revisions.
This patch introduces "--end-of-options", which serves that purpose:
git rev-list --oneline --end-of-options "$revision"
will work regardless of what's in "$revision" (well, if you say "--" it
may fail, but it won't do something dangerous, like triggering an
unexpected option).
The name is verbose, but that's probably a good thing; this is meant to
be used for scripted invocations where readability is more important
than terseness.
One alternative would be to introduce an explicit option to mark a
revision, like:
git rev-list --oneline --revision="$revision"
That's slightly _more_ informative than this commit (because it makes
even something silly like "--" unambiguous). But the pattern of using a
separator like "--" is well established in git and in other commands,
and it makes some scripting tasks simpler like:
git rev-list --end-of-options "$@"
There's no documentation in this patch, because it will make sense to
describe the feature once it is available everywhere (and support will
be added in further patches).
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The tips of refs from the alternate object store can be used as
starting point for reachability computation now.
* jk/check-connected-with-alternates:
check_everything_connected: assume alternate ref tips are valid
object-store.h: move for_each_alternate_ref() from transport.h
When we receive a remote ref update to sha1 "X", we want to check that
we have all of the objects needed by "X". We can assume that our
repository is not currently corrupted, and therefore if we have a ref
pointing at "Y", we have all of its objects. So we can stop our
traversal from "X" as soon as we hit "Y".
If we make the same non-corruption assumption about any repositories we
use to store alternates, then we can also use their ref tips to shorten
the traversal.
This is especially useful when cloning with "--reference", as we
otherwise do not have any local refs to check against, and have to
traverse the whole history, even though the other side may have sent us
few or no objects. Here are results for the included perf test (which
shows off more or less the maximal savings, getting one new commit and
sharing the whole history):
Test HEAD^ HEAD
--------------------------------------------------------------------
[on git.git]
5600.3: clone --reference 2.94(2.86+0.08) 0.09(0.08+0.01) -96.9%
[on linux.git]
5600.3: clone --reference 45.74(45.34+0.41) 0.36(0.30+0.08) -99.2%
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix a bug introduced in 18547aacf5 ("grep/pcre: support utf-8",
2016-06-25) that was missed due to a blindspot in our tests, as
discussed in the previous commit. I then blindly copied the same bug
in 94da9193a6 ("grep: add support for PCRE v2", 2017-06-01) when
adding the PCRE v2 code.
We should not tell PCRE that we're processing UTF-8 just because we're
dealing with non-ASCII. In the case of e.g. "log --encoding=<...>"
under is_utf8_locale() the haystack might be in ISO-8859-1, and the
needle might be in a non-UTF-8 encoding.
Maybe we should be more strict here and die earlier? Should we also be
converting the needle to the encoding in question, and failing if it's
not a string that's valid in that encoding? Maybe.
But for now matching this as non-UTF8 at least has some hope of
producing sensible results, since we know that our default heuristic
of assuming the text to be matched is in the user locale encoding
isn't true when we've explicitly encoded it to be in a different
encoding.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When updating the topo-order walk in b454241 (revision.c: generation-based
topo-order algorithm, 2018-11-01), the logic was a huge rewrite of the
walk logic. In that massive change, we accidentally included the
UNINTERESTING commits in expand_topo_walk(). This means that a simple
query like
git rev-list --topo-order HEAD~1..HEAD
will expand the topo walk for all commits reachable from HEAD, and not
just one commit.
This change should speed up these cases, but there is still a need
for corrected commit-date for some A..B queries.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a commit-graph exists with computed generation numbers, then a
'git rev-list --topo-order -n <N> <rev>' query will use those generation
numbers to reduce the number of commits walked before writing N commits.
One caveat put in b454241 (revision.c: generation-based topo-order
algorithm, 2018-11-01) was to not enable the new algorithm for queries
with a revision range "A..B". The logic was placed to walk from "A" and
mark those commits as uninteresting, but the performance was actually
worse than the existing logic in some cases.
The root cause of this performance degradation is that generation
numbers _increase_ the number of commits we walk relative to the
existing heuristic of walking by commit date. While generation numbers
actually guarantee that the algorithm is correct, the existing logic
is very rarely wrong and that added requirement is not worth the cost.
This motivates the planned "corrected commit date" to replace
generation numbers in a future version of Git.
The current change enables the logic to use whatever reachability
index is currently in the commit-graph (generation numbers or
corrected commit date).
The limited flag in struct rev_info forces a full walk of the
commit history (after discovering the A..B range). Previosuly, it
is enabled whenever we see an uninteresting commit. We prevent
enabling the parameter when we are planning to use the reachability
index for a topo-order.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Performance fix for "rev-list --parents -- pathspec".
* jk/revision-rewritten-parents-in-prio-queue:
revision: use a prio_queue to hold rewritten parents
Code cleanup.
* jk/unused-params-even-more:
parse_opt_ref_sorting: always use with NONEG flag
pretty: drop unused strbuf from parse_padding_placeholder()
pretty: drop unused "type" parameter in needs_rfc2047_encoding()
parse-options: drop unused ctx parameter from show_gitcomp()
fetch_pack(): drop unused parameters
report_path_error(): drop unused prefix parameter
unpack-trees: drop unused error_type parameters
unpack-trees: drop name_entry from traverse_by_cache_tree()
test-date: drop unused "now" parameter from parse_dates()
update-index: drop unused prefix_length parameter from do_reupdate()
log: drop unused "len" from show_tagger()
log: drop unused rev_info from early output
revision: drop some unused "revs" parameters
"git log -L<from>,<to>:<path>" with "-s" did not suppress the patch
output as it should. This has been corrected.
* jk/line-log-with-patch:
line-log: detect unsupported formats
line-log: suppress diff output with "-s"
This patch fixes a quadratic list insertion in rewrite_one() when
pathspec limiting is combined with --parents. What happens is something
like this:
1. We see that some commit X touches the path, so we try to rewrite
its parents.
2. rewrite_one() loops forever, rewriting parents, until it finds a
relevant parent (or hits the root and decides there are none). The
heavy lifting is done by process_parent(), which uses
try_to_simplify_commit() to drop parents.
3. process_parent() puts any intermediate parents into the
&revs->commits list, inserting by commit date as usual.
So if commit X is recent, and then there's a large chunk of history that
doesn't touch the path, we may add a lot of commits to &revs->commits.
And insertion by commit date is O(n) in the worst case, making the whole
thing quadratic.
We tried to deal with this long ago in fce87ae538 (Fix quadratic
performance in rewrite_one., 2008-07-12). In that scheme, we cache the
oldest commit in the list; if the new commit to be added is older, we
can start our linear traversal there. This often works well in practice
because parents are older than their descendants, and thus we tend to
add older and older commits as we traverse.
But this isn't guaranteed, and in fact there's a simple case where it is
not: merges. Imagine we look at the first parent of a merge and see a
very old commit (let's say 3 years old). And on the second parent, as we
go back 3 years in history, we might have many commits. That one
first-parent commit has polluted our oldest-commit cache; it will remain
the oldest while we traverse a huge chunk of history, during which we
have to fall back to the slow, linear method of adding to the list.
Naively, one might imagine that instead of caching the oldest commit,
we'd start at the last-added one. But that just makes some cases faster
while making others slower (and indeed, while it made a real-world test
case much faster, it does quite poorly in the perf test include here).
Fundamentally, these are just heuristics; our worst case is still
quadratic, and some cases will approach that.
Instead, let's use a data structure with better worst-case performance.
Swapping out revs->commits for something else would have repercussions
all over the code base, but we can take advantage of one fact: for the
rewrite_one() case, nobody actually needs to see those commits in
revs->commits until we've finished generating the whole list.
That leaves us with two obvious options:
1. We can generate the list _unordered_, which should be O(n), and
then sort it afterwards, which would be O(n log n) total. This is
"sort-after" below.
2. We can insert the commits into a separate data structure, like a
priority queue. This is "prio-queue" below.
I expected that sort-after would be the fastest (since it saves us the
extra step of copying the items into the linked list), but surprisingly
the prio-queue seems to be a bit faster.
Here are timings for the new p0001.6 for all three techniques across a
few repositories, as compared to master:
master cache-last sort-after prio-queue
--------------------------------------------------------------------------------------------
GIT_PERF_REPO=git.git
0.52(0.50+0.02) 0.53(0.51+0.02) +1.9% 0.37(0.33+0.03) -28.8% 0.37(0.32+0.04) -28.8%
GIT_PERF_REPO=linux.git
20.81(20.74+0.07) 20.31(20.24+0.07) -2.4% 0.94(0.86+0.07) -95.5% 0.91(0.82+0.09) -95.6%
GIT_PERF_REPO=llvm-project.git
83.67(83.57+0.09) 4.23(4.15+0.08) -94.9% 3.21(3.15+0.06) -96.2% 2.98(2.91+0.07) -96.4%
A few items to note:
- the cache-list tweak does improve the bad case for llvm-project.git
that started my digging into this problem. But it performs terribly
on linux.git, barely helping at all.
- the sort-after and prio-queue techniques work well. They approach
the timing for running without --parents at all, which is what you'd
expect (see below for more data).
- prio-queue just barely outperforms sort-after. As I said, I'm not
really sure why this is the case, but it is. You can see it even
more prominently in this real-world case on llvm-project.git:
git rev-list --parents 07ef786652e7 -- llvm/test/CodeGen/Generic/bswap.ll
where prio-queue routinely outperforms sort-after by about 7%. One
guess is that the prio-queue may just be more efficient because it
uses a compact array.
There are three new perf tests:
- "rev-list --parents" gives us a baseline for running with --parents.
This isn't sped up meaningfully here, because the bad case is
triggered only with simplification. But it's good to make sure we
don't screw it up (now, or in the future).
- "rev-list -- dummy" gives us a baseline for just traversing with
pathspec limiting. This gives a lower bound for the next test (and
it's also a good thing for us to be checking in general for
regressions, since we don't seem to have any existing tests).
- "rev-list --parents -- dummy" shows off the problem (and our fix)
Here are the timings for those three on llvm-project.git, before and
after the fix:
Test master prio-queue
------------------------------------------------------------------------------
0001.3: rev-list --parents 2.24(2.12+0.12) 2.22(2.11+0.11) -0.9%
0001.5: rev-list -- dummy 2.89(2.82+0.07) 2.92(2.89+0.03) +1.0%
0001.6: rev-list --parents -- dummy 83.67(83.57+0.09) 2.98(2.91+0.07) -96.4%
Changes in the first two are basically noise, and you can see we
approach our lower bound in the final one.
Note that we can't fully get rid of the list argument from
process_parents(). Other callers do have lists, and it would be hard to
convert them. They also don't seem to have this problem (probably
because they actually remove items from the list as they loop, meaning
it doesn't grow so large in the first place). So this basically just
drops the "cache_ptr" parameter (which was used only by the one caller
we're fixing here) and replaces it with a prio_queue. Callers are free
to use either data structure, depending on what they're prepared to
handle.
Reported-by: Björn Pettersson A <bjorn.a.pettersson@ericsson.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are several internal helpers that take a rev_info struct but don't
actually look at it. While one could argue that all helpers in
revision.c should take a rev_info struct for consistency, dropping the
unused parameter makes it clear that they don't actually depend on any
other rev options.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If you use "log -L" with an output format like "--raw" or "--stat",
we'll silently ignore the format and just output the normal patch.
Let's detect and complain about this, which at least tells the user
what's going on.
The tests here aren't exhaustive over the set of all formats, but it
should at least let us know if somebody breaks the format-checking.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Output from "diff --cc" did not show the original paths when the
merge involved renames. A new option adds the paths in the
original trees to the output.
* en/combined-all-paths:
log,diff-tree: add --combined-all-paths option
The combined diff format for merges will only list one filename, even if
rename or copy detection is active. For example, with raw format one
might see:
::100644 100644 100644 fabadb8cc95eb04866510 MM describe.c
::100755 100755 100755 52b7a2d 6d1ac04 d2ac7d7 RM bar.sh
::100644 100644 100644 e07d6c5 9042e82 ee91881 RR phooey.c
This doesn't let us know what the original name of bar.sh was in the
first parent, and doesn't let us know what either of the original names
of phooey.c were in either of the parents. In contrast, for non-merge
commits, raw format does provide original filenames (and a rename score
to boot). In order to also provide original filenames for merge
commits, add a --combined-all-paths option (which must be used with
either -c or --cc, and is likely only useful with rename or copy
detection active) so that we can print tab-separated filenames when
renames are involved. This transforms the above output to:
::100644 100644 100644 fabadb8cc95eb04866510 MM desc.c desc.c desc.c
::100755 100755 100755 52b7a2d 6d1ac04 d2ac7d7 RM foo.sh bar.sh bar.sh
::100644 100644 100644 e07d6c5 9042e82 ee91881 RR fooey.c fuey.c phooey.c
Further, in patch format, this changes the from/to headers so that
instead of just having one "from" header, we get one for each parent.
For example, instead of having
--- a/phooey.c
+++ b/phooey.c
we would see
--- a/fooey.c
--- a/fuey.c
+++ b/phooey.c
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git pack-objects" learned another algorithm to compute the set of
objects to send, that trades the resulting packfile off to save
traversal cost to favor small pushes.
* ds/push-sparse-tree-walk:
pack-objects: create GIT_TEST_PACK_SPARSE
pack-objects: create pack.useSparse setting
revision: implement sparse algorithm
list-objects: consume sparse tree walk
revision: add mark_tree_uninteresting_sparse