refs: unregister reference stores from "chdir_notify"
When creating reference stores we register them with the "chdir_notify"
subsystem. This is required because some of the paths we track may be
relative paths, so we have to reparent them in case the current working
directory changes.
But while we register the reference stores, we never unregister them.
This can have multiple outcomes:
- For a repository's main reference database we essentially keep the
pointer alive. We never free that database, either, and our leak
checker doesn't notice because it's still registered.
- For submodule and worktree reference databases we do eventually free
them in `repo_clear()`, so we may keep pointers to free'd memory
registered. We never notice though as we don't tend to chdir around
in the middle of the process.
We never noticed either of these symptoms, but they are obviously bad.
Partially fix those issues by unregistering the reference stores when
releasing them. The leak of the main reference database will be fixed in
a subsequent commit.
Note that this requires us to use `chdir_notify_register()` instead of
`chdir_notify_reparent()`, as there is no infrastructure to unregister the
latter.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
setup: don't apply "GIT_REFERENCE_BACKEND" without a repository
When discovering a repository we eventually also apply the
"GIT_REFERENCE_BACKEND" environment variable to the repository. There's
two problems with that:
- We do this unconditionally, which is rather pointless: we really
only have to configure the repository when we have found one.
- We have already applied the repository format at that point in time,
so we need to manually reapply it.
Move the logic around so that we only apply the environment variable
when a repository was discovered. This also allows us to drop the
explcit call to `repo_set_ref_storage_format()` because we now adjust
the format before we apply it via `apply_repository_format()`.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
When discovering the repository in "setup.c" we apply the final
repository format multiple times:
- Once via `repository_format_configure()`, where we apply the hash
algorithm and ref storage format to both `struct repository_format`
and `struct repository`.
- And once via `apply_repository_format()`, where we apply these two
settings from `struct repository_format` to `struct repository`.
With the current flow both of these are in fact necessary. But this is
only because we call `repository_format_configure()` after we have
called `apply_repository_format()`. Consequently, if we only changed the
repository format in `repository_format_configure()` it would never
propagate to the repository.
Refactor the code so that we first configure the repository format
before applying it to the repository so that we can stop setting the
hash and reference storage format multiple times.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
We have two callsites of `check_and_apply_repository_format()`. In a
subsequent commit we'll want to adapt one of those callsites to change
the order in which we read and apply the repository format, at which
point the helper function will not really be a good fit for us anymore.
Inline the function to both of the callsites.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
Junio C Hamano [Thu, 11 Jun 2026 12:09:19 +0000 (05:09 -0700)]
Merge branch 'ps/setup-centralize-odb-creation' into ps/refs-onbranch-fixes
* ps/setup-centralize-odb-creation:
setup: construct object database in `apply_repository_format()`
repository: stop reading loose object map twice on repo init
setup: stop initializing object database without repository
setup: stop creating the object database in `setup_git_env()`
repository: stop initializing the object database in `repo_set_gitdir()`
setup: deduplicate logic to apply repository format
setup: drop `setup_git_env()`
t0001: plug test gaps for git-init(1) with GIT_OBJECT_DIRECTORY
Junio C Hamano [Tue, 9 Jun 2026 01:04:51 +0000 (10:04 +0900)]
Merge branch 'ib/doc-push-default-simple'
The documentation for `push.default = simple` has been clarified to
better explain its behavior, making it clear that it pushes the
current branch to a same-named branch on the remote, and detailing
the upstream requirements for centralized workflows.
Junio C Hamano [Tue, 9 Jun 2026 01:04:50 +0000 (10:04 +0900)]
Merge branch 'ua/push-remote-group'
"git push" learned to take a "remote group" name to push to, which
causes pushes to multiple places, just like "git fetch" would do.
* ua/push-remote-group:
push: support pushing to a remote group
remote: move remote group resolution to remote.c
remote: fix sign-compare warnings in push_cas_option
Junio C Hamano [Tue, 9 Jun 2026 01:04:49 +0000 (10:04 +0900)]
Merge branch 'th/promisor-quiet-per-repo'
The "promisor.quiet" configuration variable was not used from
relevant submodules when commands like "grep --recurse-submodules"
triggered a lazy fetch, which has been corrected.
* th/promisor-quiet-per-repo:
promisor-remote: fix promisor.quiet to use the correct repository
Junio C Hamano [Tue, 9 Jun 2026 01:04:49 +0000 (10:04 +0900)]
Merge branch 'tb/bitmap-build-performance'
Reachability bitmap generation has been significantly optimized. By
reordering tree traversal, caching object positions, and refining how
pseudo-merge bitmaps are constructed, the performance of "git repack
--write-midx-bitmaps" is improved, especially for large repositories
and when using pseudo-merges.
* tb/bitmap-build-performance:
pack-bitmap: build pseudo-merge bitmaps after regular bitmaps
pack-bitmap: remember pseudo-merge parents
pack-bitmap: sort bitmaps before XORing
pack-bitmap: cache object positions during fill
pack-bitmap: consolidate `find_object_pos()` success path
pack-bitmap: reuse stored selected bitmaps
pack-bitmap: check subtree bits before recursing
pack-bitmap: pass object position to `fill_bitmap_tree()`
Junio C Hamano [Sun, 7 Jun 2026 14:58:25 +0000 (23:58 +0900)]
Merge branch 'ja/doc-synopsis-style-again'
A batch of documentation pages has been updated to use the modern
synopsis style.
* ja/doc-synopsis-style-again:
doc: convert git-imap-send synopsis and options to new style
doc: convert git-apply synopsis and options to new style
doc: convert git-am synopsis and options to new style
doc: convert git-grep synopsis and options to new style
doc: git bisect: clarify the usage of the synopsis vs actual command
doc: convert git-bisect to synopsis style
Junio C Hamano [Sun, 7 Jun 2026 14:58:25 +0000 (23:58 +0900)]
Merge branch 'kk/commit-reach-optim'
The check for non-stale commits in the priority queue used by
`paint_down_to_common` and `ahead_behind` has been optimized by
replacing an O(N) scan with an O(1) counter, yielding performance
improvements in repositories with wide histories.
* kk/commit-reach-optim:
commit-reach: replace queue_has_nonstale() scan with O(1) tracking
commit-reach: deduplicate queue entries in paint_down_to_common
object.h: fix stale entries in object flag allocation table
Junio C Hamano [Sun, 7 Jun 2026 14:58:24 +0000 (23:58 +0900)]
Merge branch 'ds/restore-sparse-index'
'git restore --staged' has been optimized to avoid unnecessarily expanding
the sparse index when operating on paths within the sparse checkout
definition, by handling sparse directory entries at the tree level.
* ds/restore-sparse-index:
restore: avoid sparse index expansion
t1092: test 'git restore' with sparse index
Junio C Hamano [Sun, 7 Jun 2026 14:58:24 +0000 (23:58 +0900)]
Merge branch 'ar/receive-pack-worktree-env'
The GIT_WORK_TREE variable prepared to invoke the push-to-checkout
hook was leaking into the environment even when there was no hook
used and broke the default push-to-deploy (i.e., let "git checkout"
update the working tree only when the working tree is clean).
* ar/receive-pack-worktree-env:
receive-pack: fix updateInstead with core.worktree
setup: construct object database in `apply_repository_format()`
With the preceding changes we now always construct the repository's
object database before applying the repository format. Remove this
duplication by constructing it in `apply_repository_format()` instead.
Note that we create the object database _after_ having set up the
repository's hash algorithm, but _before_ setting the compat hash
algorithm. This is intentional:
- Constructing the object database may require knowledge of its
intended object format.
- Setting up the compatibility hash requires the object database to be
initialized already, because we immediately read the loose object
map.
The first point is sensible, the second maybe a little less so. Ideally,
it should be the responsibility of the object database itself to
initialize any data structures required for the compatibility hash. But
this would require further changes, so this is kept as-is for now.
Further note that this requires us to move handling of the environment
variables GIT_OBJECT_DIRECTORY and GIT_ALTERNATE_OBJECT_DIRECTORIES into
the repository format, as well. This allows the caller more flexibility
around whether or not those environment variables are being honored, as
we want to respect them in "setup.c", but not in "repository.c".
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
repository: stop reading loose object map twice on repo init
When initializing a repository via `repo_init()` we end up reading the
loose object map twice:
- `apply_repository_format()` calls `repo_set_compat_hash_algo()`,
which in turn calls `repo_read_loose_object_map()` if we have a
compatibility hash configured.
- `repo_init()` calls `repo_read_loose_object_map()` directly a second
time.
Drop the second read of the loose object map in `repo_init()`.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
setup: stop initializing object database without repository
The function `setup_git_directory_gently()` is responsible for
discovering and setting up a Git repository based on various environment
variables and the current working directory. The result is thus a fully
usable Git repository.
One oddity of this function is that we may set up the object database
even in the case where we don't have a repository, namely in the case
where the `GIT_DIR_EXPLICIT` environment variable is set but points to a
non-existent repository. If so, we call `setup_git_env_internal()` with
the value of the environment variable so that the repository's Git
directory is configured, even if it points to a non-existent directory.
Historically though, this function didn't only configure the repository,
but also initialized the object database. We retained this behaviour
from a preceding commit, even though it really doesn't make much sense
in the first place -- there is no repository, so we don't have an object
database either. There seemingly isn't much of a reason to construct the
object database, as we typically won't try to read objects when we don't
have an object database.
There's one exception though: git-index-pack(1) may run outside of a
repository, which can be used to perform consistency checks for a
packfile. The code path is _almost_ working: we already know to call
`parse_object_buffer()`, which can read objects without an object
database being available. And that works for all object types except for
commits, because `parse_commit_buffer()` calls `parse_commit_graph()`,
and that function doesn't handle the case where we don't have an object
database.
Fix this instance to check for the object database instead of checking
for the Git directory having been initialized. With this fixed, we can
now stop constructing an object database completely.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
setup: stop creating the object database in `setup_git_env()`
In the preceding commit we have stopped creating the object database in
`repo_set_gitdir()`. But the logic is still somewhat confusing as we
still end up creating it conditionally in `setup_git_dir()`, which is
called multiple times.
Drop the conditional logic and instead create the object database in all
places where we have discovered and configured a repository.
This leads to even more duplication than we already had in the preceding
commit, but an alert reader may notice that we now (almost) always call
`odb_new()` directly before having called `apply_repository_format()`.
The only exception to this is `setup_git_directory_gently()`, where we
also call the function when _not_ applying the repository format. This
will be fixed in the next commit, and once that's done we can then unify
creation of the object database into `apply_repository_format()`.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
repository: stop initializing the object database in `repo_set_gitdir()`
The function `repo_set_gitdir()` obviously sets the Git directory for a
given repository. Less obviously though, the function also configures a
couple of auxiliary settings.
One such thing is that we create the object database in this function.
This logic only happens conditionally though, as `set_git_dir()` may be
called multiple times during repository setup, and we don't want to
create the object database multiple times. This is somewhat tangled and
hard to follow.
Remove the logic from `repo_set_gitdir()` and instead initialize the
object database outside of it. This leads to some duplication right now,
but that duplication will be removed in a subsequent step where we will
start initializing the object database as part of applying the repo's
format.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
setup: deduplicate logic to apply repository format
After having discovered the repository format we then apply it to the
repository so that it knows to use the proper repository extensions. The
logic to apply the format is duplicated across three callsites, which
makes it rather painfull to add new extensions.
Introduce a new function `apply_repository_format()` that takes a repo
and applies a given format to it and adapt all callsites to use it.
This function is also the new caller of `verify_repository_format()` so
that we can ensure that we never apply an invalid repository format.
The verification we have in `read_and_verify_repository_format()` is
thus redundant now and dropped.
Rename `read_and_verify_repository_format()` accordingly. While at it,
also rename `check_repository_format()` to clarify that it doesn't only
_check_ the format, but that it also applies it.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
The `setup_git_env()` function is a trivial wrapper around
`setup_git_env_internal()` and has a single call site only. Drop the
function.
While at it, drop stale documentation in "environment.h" that points to
this function, even though it hasn't been exposed to callers outside of
"setup.c" since 43ad1047a9 (setup: stop using `the_repository` in
`setup_git_env()`, 2026-03-27) anymore.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
t0001: plug test gaps for git-init(1) with GIT_OBJECT_DIRECTORY
In subsequent commits we'll rework how we set up the repository. This is
a somewhat intricate and thus fragile sequence; there's many things that
can go subtly wrong, and there are lots of interesting interactions that
one can discover.
One such discovered edge case was the interaction between git-init(1)
and the "GIT_OBJECT_DIRECTORY" environment variable. When set, the
behaviour is that the object directory should be created at the path
that the variable points to. This behaviour is documented as such in
its man page:
If the object storage directory is specified via the
GIT_OBJECT_DIRECTORY environment variable then the sha1 directories
are created underneath; otherwise, the default $GIT_DIR/objects
directory is used.
Curiously enough though we don't seem to have any tests that exercise
this directly, and thus a subsequent commit inadvertently would have
broken this expectation.
Plug this test gap.
Signed-off-by: Patrick Steinhardt <redacted> Signed-off-by: Junio C Hamano <redacted>
Junio C Hamano [Tue, 2 Jun 2026 07:15:29 +0000 (16:15 +0900)]
Merge branch 'ds/path-walk-filters'
The "git pack-objects --path-walk" traversal has been integrated
with several object filters, including blobless and sparse filters.
* ds/path-walk-filters:
path-walk: support `combine` filter
path-walk: support `object:type` filter
path-walk: support `tree:0` filter
t6601: tag otherwise-unreachable trees
pack-objects: support sparse:oid filter with path-walk
path-walk: add pl_sparse_trees to control tree pruning
path-walk: support blob size limit filter
backfill: die on incompatible filter options
path-walk: support blobless filter
path-walk: always emit directly-requested objects
t/perf: add pack-objects filter and path-walk benchmark
pack-objects: pass --objects with --path-walk
t5620: make test work with path-walk var
Junio C Hamano [Tue, 2 Jun 2026 07:15:29 +0000 (16:15 +0900)]
Merge branch 'ta/approxidate-noon-fix'
"Friday noon" asked in the morning on Sunday was parsed to be one
day before the specified time, which has been corrected.
* ta/approxidate-noon-fix:
approxidate: use deferred mday adjustments for "specials"
approxidate: make "specials" respect fixed day-of-month
t0006: add support for approxidate test date adjustment
approxidate: make "today" wrap to midnight
Junio C Hamano [Sun, 31 May 2026 01:00:38 +0000 (10:00 +0900)]
Merge branch 'jk/commit-graph-lazy-load-fallback'
The logic to lazy-load trees from the commit-graph has been made
more robust by falling back to reading the commit object when
the commit-graph is no longer available.
* jk/commit-graph-lazy-load-fallback:
commit: fall back to full read when maybe_tree is NULL
Junio C Hamano [Sun, 31 May 2026 01:00:38 +0000 (10:00 +0900)]
Merge branch 'pt/fsmonitor-linux'
The fsmonitor daemon has been implemented for Linux.
* pt/fsmonitor-linux:
fsmonitor: convert shown khash to strset in do_handle_client
fsmonitor: add tests for Linux
fsmonitor: add timeout to daemon stop command
fsmonitor: close inherited file descriptors and detach in daemon
run-command: add close_fd_above_stderr option
fsmonitor: implement filesystem change listener for Linux
fsmonitor: rename fsm-settings-darwin.c to fsm-settings-unix.c
fsmonitor: rename fsm-ipc-darwin.c to fsm-ipc-unix.c
fsmonitor: use pthread_cond_timedwait for cookie wait
compat/win32: add pthread_cond_timedwait
fsmonitor: fix hashmap memory leak in fsmonitor_run_daemon
fsmonitor: fix khash memory leak in do_handle_client
t9210, t9211: disable GIT_TEST_SPLIT_INDEX for scalar clone tests
Junio C Hamano [Sun, 31 May 2026 01:00:37 +0000 (10:00 +0900)]
Merge branch 'ps/graph-lane-limit'
The graph output from commands like "git log --graph" can now be
limited to a specified number of lanes, preventing overly wide output
in repositories with many branches.
* ps/graph-lane-limit:
graph: add truncation mark to capped lanes
graph: add --graph-lane-limit option
graph: limit the graph width to a hard-coded max
Junio C Hamano [Sun, 31 May 2026 01:00:37 +0000 (10:00 +0900)]
Merge branch 'jr/bisect-custom-terms-in-output'
"git bisect" now uses the selected terms (e.g., old/new) more
consistently in its output.
* jr/bisect-custom-terms-in-output:
rev-parse: use selected alternate terms to look up refs
bisect: print bisect terms in single quotes
bisect: use selected alternate terms in status output
Junio C Hamano [Sun, 31 May 2026 01:00:37 +0000 (10:00 +0900)]
Merge branch 'kk/tips-reachable-from-bases-optim'
Revision traversal optimization.
* kk/tips-reachable-from-bases-optim:
t6600: add tests for duplicate tips in tips_reachable_from_bases()
commit-reach: use object flags for tips_reachable_from_bases()
Replace `free_commit_list` with `commit_list_free`. The former was
deprecated in 9f18d089 (commit: rename `free_commit_list()` to conform
to coding guidelines, 2026-01-15).
This allows us to remove all the deprecated functions in the
next commit:
Taylor Blau [Wed, 27 May 2026 19:56:11 +0000 (15:56 -0400)]
pack-bitmap: build pseudo-merge bitmaps after regular bitmaps
When generating bitmaps, `bitmap_builder_init()` starts with an initial
selection of commits to receive bitmap coverage, and then determines a
set of "maximal" commits based on its input.
Commit 089f751360f (pack-bitmap-write: build fewer intermediate bitmaps,
2020-12-08) has extensive details, but the gist is as follows:
Each selected commit starts with one commit_mask bit in its "commit
mask" bitmap. Then, we walk the first-parent history in topological
order and OR each commit's mask into its (first) parent. Whenever that
OR results in the parent having more bits set, the child is deemed to be
non-maximal, and the frontier is pushed further back along the first
parent history.
That approach works extremely well for ordinary selected commits, whose
first-parent histories often describe real sharing between the bitmaps
we are going to write.
It struggles, however, to efficiently generate pseudo-merge bitmaps.
Unlike ordinary commits for which the above algorithm is designed,
pseudo-merges don't represent any "real" commit in history, just a
grouping of non-bitmapped reference tips. In that sense, their first
parent is just a part of a larger set, and treating them like ordinary
selected commits imposes a significant slow-down when generating bitmaps
with pseudo-merges enabled.
Consider partitioning all non-bitmapped reference tips into eight
individual pseudo-merges via the following configuration:
Benchmark 1: pseudo-merges: true
Time (mean ± σ): 2.613 s ± 0.012 s [User: 2.308 s, System: 0.305 s]
Range (min … max): 2.594 s … 2.633 s 10 runs
Benchmark 2: pseudo-merges: false
Time (mean ± σ): 52.205 s ± 0.170 s [User: 51.500 s, System: 0.697 s]
Range (min … max): 51.956 s … 52.458 s 10 runs
Summary
pseudo-merges: true ran
19.98 ± 0.11 times faster than pseudo-merges: false
In other words, we pay a nearly ~5 minute penalty to generate
pseudo-merge bitmaps, but only save ~50 seconds during traversal.
The problem stems from injecting pseudo-merges into the bitmap builder
as if they were normal commits. The maximal commit selection algorithm
was simply not designed for that case, and performs predictably poorly.
The only reason we reused the maximal commit selection routine for
pseudo-merges alongside regular non-pseudo-merge commits is because we
represent them both as commit objects (where the pseudo-merge commits
just represent a made-up commit as opposed to one that actually exists
in a repository's object store).
Instead, build the regular selected commit bitmaps first, considering
only non-pseudo-merge commits in `bitmap_builder_init()`. Once those
bitmaps have been stored, build each pseudo-merge bitmap separately and
attach its parent and object bitmaps to the corresponding pseudo-merge
entry before writing the extension.
This keeps the regular bitmap build shaped like the no-pseudo-merge
case. The later pseudo-merge fill can still stop at stored selected
ancestor bitmaps, so it does not have to rewalk each pseudo-merge
closure from scratch.
When an existing bitmap has the same pseudo-merge parent set, reuse and
remap that whole pseudo-merge bitmap before falling back to
fill_bitmap_commit(). This preserves the benefit of stable pseudo-merges
while keeping the on-disk format and reader behavior unchanged.
As a result, the overhead cost for generating pseudo-merges in the above
configuration is much smaller:
+------------------+-----------------+---------------+-------------------+
| | no pseudo-merge | pseudo-merges | Delta |
| | | (HEAD) | |
+------------------+-----------------+---------------+-------------------+
| elapsed | 294.1 s | 328.4 s | +34.3 s (+11.7%) |
| cycles | 1,365.5 B | 1,529.3 B | +163.7 B (+12.0%) |
| instructions | 1,389.8 B | 1,552.8 B | +163.0 B (+11.7%) |
| CPI | 0.983 | 0.985 | +0.002 (+0.2%) |
+------------------+-----------------+---------------+-------------------+
Recall that at the start of this series, generating reachability bitmaps
took 612.5 seconds *without* pseudo-merges. With this commit, it is
still ~46.38% *faster* to generate reachability bitmaps *with*
pseudo-merges than it was to generate bitmaps wihtout them at the
beginning of this series.
The changes to implement this are mostly straightforward. We exclude
pseudo-merge commits from the existing bitmap generation, and walk over
them in a separate pass, by either reusing an existing on-disk
pseudo-merge, or passing the pseudo-merge commit itself back to the
existing routine in `fill_bitmap_commit()`.
(Note that the routine to build pseudo-merge bitmaps is the same both
before and after this change, the difference is only that we do not let
psuedo-merges participate in determining the set of maximal commits.)
The only wrinkle is that `fill_bitmap_commit()` must be taught to not
expect that all tree objects have been parsed, which is the case for any
portion of history reachable by one or more pseudo-merge(s), but not by
any non-pseudo-merge commit selected for bitmapping.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Wed, 27 May 2026 19:56:08 +0000 (15:56 -0400)]
pack-bitmap: remember pseudo-merge parents
write_pseudo_merges() currently builds an array of temporary bitmaps for
the parent set of each pseudo-merge, then serializes those bitmaps later
while writing the extension.
Move those parent bitmaps onto the corresponding bitmapped_commit
entries instead. This keeps the on-disk output unchanged, but gives the
parent bitmap the same lifetime and access pattern that later changes
will use when pseudo-merge object bitmaps are built before the write
step.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Wed, 27 May 2026 19:56:05 +0000 (15:56 -0400)]
pack-bitmap: sort bitmaps before XORing
Reachability bitmaps may be stored as XORs against nearby bitmaps, up to
10 away. However, when callers provide selected commits in an arbitrary
order, the writer may miss good ancestor/descendant pairs and produce
much larger bitmap files without changing query coverage.
Sort the selected bitmaps in date order (from oldest to newest) before
computing XOR offsets, leaving pseudo-merge bitmaps alone (which we will
deal with separately in following commits).
On our same testing repository from previous commits, this change shrunk
our selection of 1,261 bitmaps from ~635.46 MiB to 176.4 MiB for a
~72.24% reduction in the on-disk size of our *.bitmap file. The time to
generate the smaller bitmap file decreased by ~3.69 seconds, though this
is likely mostly noise.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Wed, 27 May 2026 19:56:02 +0000 (15:56 -0400)]
pack-bitmap: cache object positions during fill
The previous commits removed some redundant work from bitmap generation
by avoiding unnecessary tree recursion and by reusing selected bitmaps
that have already been computed.
Even with those changes in place, there is still an extremely hot path
from `fill_bitmap_commit()` and `fill_bitmap_tree()` to translate object
IDs into their corresponding bit positions in order to generate their
bitmaps.
In a small repository, this overhead is not significant. However, in a
very large repository (e.g., the one that we have been using as a
benchmark over the past several commits with ~57M total objects), the
overhead of locating object bit positions (often repeatedly) adds up
significantly.
Combat this by adding a small, direct-mapped cache to the bitmap writer
which maps object IDs to their corresponding bit positions. Size the
cache according to the number of objects being written, with fixed lower
and upper bounds so small repositories do not pay for a large table and
large repositories can avoid most repeated packlist and MIDX lookups.
On my machine with (a somewhat outdated) GCC 15.2.0, each entry in the
cache is 40 bytes wide:
, and we will allocate up to 2^21 entries for a maximum total of 80 MiB
of cache overhead.
In our example repository from above and in earlier commits, this
results in a ~9.4% reduction in runtime relative to the previous commit:
+------------------+-------------+-------------+---------------------+
| | HEAD^ | HEAD | Delta |
+------------------+-------------+-------------+---------------------+
| elapsed | 324.8 s | 294.1 s | -30.7 s (-9.4%) |
| cycles | 1,508.6 B | 1,365.5 B | -143.0 B (-9.5%) |
| instructions | 1,436.6 B | 1,389.8 B | -46.9 B (-3.3%) |
| CPI | 1.050 | 0.983 | -0.068 (-6.4%) |
+------------------+-------------+-------------+---------------------+
When generating bitmaps on this repository (to produce the above
timings), the cache grew to its maximum size of 80 MiB, and resulted in
1.024B cache hits and 59.957M cache misses.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Both sides of `find_object_pos()` report success in the same way by
setting the optional `found` out-parameter and return the resolved
bitmap position.
Prepare for adding more bookkeeping around object-position lookups by
storing the result in a local `pos` variable and sharing the success
return path between the packlist and MIDX cases.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Wed, 27 May 2026 19:55:56 +0000 (15:55 -0400)]
pack-bitmap: reuse stored selected bitmaps
When `fill_bitmap_commit()` reaches an ancestor that was selected for
its own bitmap and processed earlier, its object closure is already
stored in `writer->bitmaps` as an EWAH bitmap. As a result, walking
through that commit's tree and parents again is redundant.
Teach `fill_bitmap_commit()` to notice that case. For non-root commits in
the walk, look for a stored selected bitmap and OR it into the bitmap
being built. If one exists, skip the commit, its tree, and its parents.
Building bitmaps from scratch on the same test repository from the
previous commits yields a significant speed-up:
+------------------+-------------+-------------+---------------------+
| | HEAD^ | HEAD | Delta |
+------------------+-------------+-------------+---------------------+
| elapsed | 562.8 s | 324.8 s | -237.9 s (-42.3%) |
| cycles | 2,621.3 B | 1,508.6 B | -1,112.7 B (-42.4%) |
| instructions | 2,348.9 B | 1,436.6 B | -912.3 B (-38.8%) |
| CPI | 1.116 | 1.050 | -0.066 (-5.9%) |
+------------------+-------------+-------------+---------------------+
In our testing repository, there are 1,261 commits selected for bitmap
coverage, and 1,382 maximal commits induced as a result of that. Of the
1,382 calls made to `fill_bitmap_commit()` (one per maximal commit), 131
of them can be short-circuited at some point during their traversal as a
consequence of this change.
In large repositories where the cost of filling the bitmap for any
individual commit is large, being able to short-circuit even ~9.5% of
the calls to `fill_bitmap_commit()` results in a significant savings.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Wed, 27 May 2026 19:55:53 +0000 (15:55 -0400)]
pack-bitmap: check subtree bits before recursing
In the previous commit, we adjusted the callers of `fill_bitmap_tree()`
to pass in the bit position of the tree they wish to fill.
This commit makes use of that information at the call site to avoid
setting up a stack frame for fill_bitmap_tree() entirely whenever a
tree's bit position is already set.
Since this is such a hot path, the avoided cost of setting up and
tearing down stack frames for each noop'd call to `fill_bitmap_tree()`
is significant:
+--------------+-------------+-------------+-------------------+
| | HEAD^ | HEAD | Delta |
+--------------+-------------+-------------+-------------------+
| elapsed | 582.4 s | 562.8 s | -19.6 s (-3.4%) |
| cycles | 2,713.3 B | 2,621.3 B | -92.0 B (-3.4%) |
| instructions | 2,415.5 B | 2,348.9 B | -66.6 B (-2.8%) |
| CPI | 1.123 | 1.116 | -0.007 (-0.7%) |
+--------------+-------------+-------------+-------------------+
In the same repository as in the previous commit, our timings dropped
from ~582.4 seconds down to ~562.77 seconds.
While the cycles-per-instruction ratio is basically unchanged, we
execute significantly fewer instructions, and correspondingly fewer
cycles.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Wed, 27 May 2026 19:55:50 +0000 (15:55 -0400)]
pack-bitmap: pass object position to `fill_bitmap_tree()`
In the following commit, callers of `fill_bitmap_tree()` will be
required to check the bit corresponding to their tree before calling
that function. That change will reduce the overhead of setting up and
tearing down stack frames for trees whose bits are already set.
To prepare for that change, have callers pass in the tree's bit position
in `fill_bitmap_tree()`, which will make the next commit easier to read.
In the meantime, this change has a surprising and measurable benefit
during bitmap generation, particularly on very large repositories.
When processing sub-trees within `fill_bitmap_tree()`, the preimage of
this patch did the following:
while (tree_entry(&desc, entry)) {
switch (object_type(entry.mode)) {
case OBJ_TREE:
if (fill_bitmap_tree(writer, bitmap,
lookup_tree(writer->repo,
&entry.oid)) < 0) {
/* ... */
}
/* ... */
}
}
, first performing the object lookup via `lookup_tree()`, and then
locating its bit position within the recursive call. This patch
effectively reorders those two calls so that we first discover the
sub-tree's bit position, *then* load its tree.
By reordering these two operations, we spend fewer CPU cycles per
instruction, likely due to improved CPU dependency/cache/pipeline
behavior. Comparing the results of: running `perf stat` before and after
this commit, we have:
+--------------+-------------+-------------+-------------------+
| | HEAD^ | HEAD | Delta |
+--------------+-------------+-------------+-------------------+
| elapsed | 612.5 s | 582.4 s | -30.1 s (-4.9%) |
| cycles | 2,857.3 B | 2,713.3 B | -144.0 B (-5.0%) |
| instructions | 2,413.2 B | 2,415.5 B | +2.3 B (+0.1%) |
| CPI | 1.184 | 1.123 | -0.061 (-5.1%) |
+--------------+-------------+-------------+-------------------+
In a large repository with ~4.8M commit, and ~37.1M tree objects this
change improves timing from ~612.5 seconds down to ~582.4 seconds, or a
~4.9% improvement. More importantly, the number of CPU cycles spent
dropped off significantly as a result of this commit, lowering our
cycles-per-instruction ratio by about ~5.1%.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Junio C Hamano <redacted>
Junio C Hamano [Wed, 27 May 2026 05:15:46 +0000 (14:15 +0900)]
Merge branch 'ps/setup-wo-the-repository'
Many uses of the_repository has been updated to use a more
appropriate struct repository instance in setup.c codepath.
* ps/setup-wo-the-repository:
setup: stop using `the_repository` in `init_db()`
setup: stop using `the_repository` in `create_reference_database()`
setup: stop using `the_repository` in `initialize_repository_version()`
setup: stop using `the_repository` in `check_repository_format()`
setup: stop using `the_repository` in `upgrade_repository_format()`
setup: stop using `the_repository` in `setup_git_directory()`
setup: stop using `the_repository` in `setup_git_directory_gently()`
setup: stop using `the_repository` in `setup_git_env()`
setup: stop using `the_repository` in `set_git_work_tree()`
setup: stop using `the_repository` in `setup_work_tree()`
setup: stop using `the_repository` in `enter_repo()`
setup: stop using `the_repository` in `verify_non_filename()`
setup: stop using `the_repository` in `verify_filename()`
setup: stop using `the_repository` in `path_inside_repo()`
setup: stop using `the_repository` in `prefix_path()`
setup: stop using `the_repository` in `is_inside_work_tree()`
setup: stop using `the_repository` in `is_inside_git_dir()`
setup: replace use of `the_repository` in static functions
Junio C Hamano [Wed, 27 May 2026 05:15:46 +0000 (14:15 +0900)]
Merge branch 'ps/odb-in-memory'
Add a new odb "in-memory" source that is meant to only hold
tentative objects (like the virtual blob object that represents the
working tree file used by "git blame").
* ps/odb-in-memory:
t/unit-tests: add tests for the in-memory object source
odb: generic in-memory source
odb/source-inmemory: stub out remaining functions
odb/source-inmemory: implement `freshen_object()` callback
odb/source-inmemory: implement `count_objects()` callback
odb/source-inmemory: implement `find_abbrev_len()` callback
odb/source-inmemory: implement `for_each_object()` callback
odb/source-inmemory: convert to use oidtree
oidtree: add ability to store data
cbtree: allow using arbitrary wrapper structures for nodes
odb/source-inmemory: implement `write_object_stream()` callback
odb/source-inmemory: implement `write_object()` callback
odb/source-inmemory: implement `read_object_stream()` callback
odb/source-inmemory: implement `read_object_info()` callback
odb: fix unnecessary call to `find_cached_object()`
odb/source-inmemory: implement `free()` callback
odb: introduce "in-memory" source
Junio C Hamano [Wed, 27 May 2026 05:15:45 +0000 (14:15 +0900)]
Merge branch 'tb/incremental-midx-part-3.3'
The repacking code has been refactored and compaction of MIDX layers
have been implemented, and incremental strategy that does not require
all-into-one repacking has been introduced.
* tb/incremental-midx-part-3.3:
repack: allow `--write-midx=incremental` without `--geometric`
repack: introduce `--write-midx=incremental`
repack: implement incremental MIDX repacking
packfile: ensure `close_pack_revindex()` frees in-memory revindex
builtin/repack.c: convert `--write-midx` to an `OPT_CALLBACK`
repack-geometry: prepare for incremental MIDX repacking
repack-midx: extract `repack_fill_midx_stdin_packs()`
repack-midx: factor out `repack_prepare_midx_command()`
midx: expose `midx_layer_contains_pack()`
repack: track the ODB source via existing_packs
midx: support custom `--base` for incremental MIDX writes
midx: introduce `--no-write-chain-file` for incremental MIDX writes
midx: use `strvec` for `keep_hashes`
midx: build `keep_hashes` array in order
midx: use `strset` for retained MIDX files
midx-write: handle noop writes when converting incremental chains
Junio C Hamano [Wed, 27 May 2026 05:15:45 +0000 (14:15 +0900)]
Merge branch 'ds/fetch-negotiation-options'
The negotiation tip options in "git fetch" have been reworked to
allow requiring certain refs to be sent as "have" lines, and to
restrict negotiation to a specific set of refs.
* ds/fetch-negotiation-options:
send-pack: pass negotiation config in push
remote: add remote.*.negotiationInclude config
fetch: add --negotiation-include option for negotiation
negotiator: add have_sent() interface
remote: add remote.*.negotiationRestrict config
transport: rename negotiation_tips
fetch: add --negotiation-restrict option
t5516: fix test order flakiness
Junio C Hamano [Wed, 27 May 2026 05:15:44 +0000 (14:15 +0900)]
Merge branch 'kn/refs-fsck-skip-lock-files'
The consistency checks for the files reference backend have been updated
to skip lock files earlier, avoiding unnecessary parsing of
intermediate files.
* kn/refs-fsck-skip-lock-files:
refs/files: skip lock files during consistency checks
Junio C Hamano [Wed, 27 May 2026 05:15:43 +0000 (14:15 +0900)]
Merge branch 'en/batch-prefetch'
In a lazy clone, "git cherry" and "git grep" often fetch necessary
blob objects one by one from promisor remotes. It has been corrected
to collect necessary object names and fetch them in bulk to gain
reasonable performance.
Junio C Hamano [Wed, 27 May 2026 05:15:43 +0000 (14:15 +0900)]
Merge branch 'pb/doc-diff-format-updates'
Doc updates.
* pb/doc-diff-format-updates:
diff-format.adoc: mode and hash are 0* for unmerged paths from index only
diff-format.adoc: 'git diff-files' prints two lines for unmerged files
diff-format.adoc: remove mention of diff-tree specific output
Junio C Hamano [Wed, 27 May 2026 05:15:43 +0000 (14:15 +0900)]
Merge branch 'kk/limit-list-optim'
The limit_list() function that is one of the core part of the
revision traversal infrastructure has been optimized by replacing
its use of linear list with priority queue.
* kk/limit-list-optim:
revision: use priority queue in limit_list()
Derrick Stolee [Tue, 26 May 2026 20:26:34 +0000 (20:26 +0000)]
restore: avoid sparse index expansion
Teach update_some() to handle sparse directory entries at the tree
level rather than expanding the entire sparse index. When iterating a
source tree during checkout/restore operations:
- If a directory matches a sparse directory entry with the same OID,
skip it entirely (no change needed).
- If the OID differs and we are in non-overlay mode (e.g., restore
--staged), update the sparse directory entry's OID in place. This
is semantically correct because non-overlay mode removes paths not
in the source tree anyway.
- In overlay mode (e.g., checkout <tree> -- .), fall through to
recursive descent so individual file entries are preserved
correctly.
Also switch from index_name_pos() to index_name_pos_sparse() for
individual file lookups to avoid triggering ensure_full_index() when
the file is already individually tracked in the index.
Update the test expectation in t1092 to assert that 'restore --staged'
no longer expands the sparse index.
Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Derrick Stolee [Tue, 26 May 2026 20:26:33 +0000 (20:26 +0000)]
t1092: test 'git restore' with sparse index
A user reported that 'git restore --staged .' causes the sparse index to
expand. This is somewhat natural because the '.' pathspec means 'check
every path'. However, the restore will not update paths marked with the
SKIP_WORKTREE bit, so we shouldn't need to process such entries.
For now, establish the current behavior, including the sparse index
expansion, in the t1092 test case as a baseline.
Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Ivan Baluta [Tue, 26 May 2026 03:58:07 +0000 (03:58 +0000)]
doc: clarify push.default=simple behavior
The documentation for the 'simple' push mode currently singles out
the centralized workflow, which can cause confusion about its
behavior in other scenarios, such as triangular workflows.
Clarify that 'simple' always pushes the current branch to a branch
of the same name, but only enforces the strict upstream tracking
requirement when pushing back to the same remote being pulled from.
Suggested-by: Junio C Hamano <redacted> Signed-off-by: Ivan Baluta <redacted> Signed-off-by: Junio C Hamano <redacted>
Alyssa Ross [Mon, 25 May 2026 16:23:12 +0000 (18:23 +0200)]
receive-pack: fix updateInstead with core.worktree
Before a8cc594333 (hooks: fix an obscure TOCTOU "did we just run a
hook?" race, 2022-03-07), when receive.denyCurrentBranch is set to
updateInstead, only one of push_to_checkout() or push_to_deploy()
was called. That commit changed to always call push_to_checkout(),
and then to call push_to_deploy() if push_to_checkout() didn't run
anything.
This change didn't take into account that push_to_checkout() had a
side effect of modifying env, and that modified env broke updating
the worktree in push_to_deploy() if core.worktree was configured.
To fix this, only mutate the environment used inside
push_to_commit(), rather than the environment that might later be
passed to push_to_deploy().
Signed-off-by: Alyssa Ross <redacted> Signed-off-by: Junio C Hamano <redacted>
commit-reach: replace queue_has_nonstale() scan with O(1) tracking
paint_down_to_common() and ahead_behind() call queue_has_nonstale()
on every iteration to decide whether to continue the walk.
queue_has_nonstale() performs a linear scan of the priority queue,
making the overall walk O(n*m) where n is the number of commits
walked and m is the queue size.
Introduce 'struct nonstale_queue', a thin wrapper around prio_queue
that maintains a 'max_nonstale' pointer — the lowest-priority
(oldest) non-stale commit seen so far. When this commit is popped,
every remaining queue entry is known to be stale, so the walk can
stop. This reduces the per-iteration termination check from O(m)
to O(1).
Uses <= 0 (not < 0) when comparing priorities so that among distinct
commits with equal priority (same generation and timestamp) the
last-enqueued one is tracked. Since prio_queue breaks ties by
insertion order, this ensures max_nonstale is always the last in its
priority class to be popped, making pointer equality on pop
sufficient for correctness.
The previous commit's ENQUEUED deduplication guarantees each commit
appears at most once in the queue, which is required for the pointer
equality check to be unambiguous.
On a large monorepo (3.7M commits), this yields ~2x end-to-end
speedup for merge-base calculations on deep import branches.
Profiling shows paint_down_to_common() drops from 50% to 4% of
total runtime (~27x faster), with the remaining time in commit
graph lookups and heap operations:
commit-reach: deduplicate queue entries in paint_down_to_common
paint_down_to_common() can enqueue the same commit multiple times
when it is reached through different parents with different flag
combinations. Add an ENQUEUED flag to track whether a commit is
currently in the priority queue, and skip it if already present.
Introduce prio_queue_put_dedup() and prio_queue_get_dedup()
wrappers that manage the ENQUEUED flag on enqueue and dequeue.
This change is performance-neutral on its own: the O(n)
queue_has_nonstale() scan still dominates the per-iteration cost.
However, the deduplication guarantee (each commit appears in the
queue at most once) is a prerequisite for the next commit, which
replaces that scan with O(1) tracking.
Signed-off-by: Kristofer Karlsson <redacted> Signed-off-by: Junio C Hamano <redacted>
object.h: fix stale entries in object flag allocation table
Update three stale entries found during an audit of the flag
allocation table:
- sha1-name.c was renamed to object-name.c
- builtin/show-branch.c uses bits 0 and 2-28, not 0-26
(REV_SHIFT=2, MAX_REVS=FLAG_BITS-REV_SHIFT=27)
- negotiator/skipping.c uses bits 2-5 like negotiator/default.c
(ADVERTISED on bit 3 instead of COMMON_REF)
Signed-off-by: Kristofer Karlsson <redacted> Signed-off-by: Junio C Hamano <redacted>
Jean-Noël Avila [Mon, 25 May 2026 10:28:27 +0000 (10:28 +0000)]
doc: convert git-imap-send synopsis and options to new style
Convert git-imap-send from [verse]/single-quote style to the modern
synopsis-block style:
- Replace [verse] with [synopsis] in SYNOPSIS block
- Backtick-quote all OPTIONS terms
- Backtick-quote all config keys in config/imap.adoc
- Backtick-quote bare config key references in prose
Signed-off-by: Jean-Noël Avila <redacted> Signed-off-by: Junio C Hamano <redacted>
Jean-Noël Avila [Mon, 25 May 2026 10:28:23 +0000 (10:28 +0000)]
doc: git bisect: clarify the usage of the synopsis vs actual command
The difference between a synopsis and an actual command is that the synopsis
is a more abstract representation of the command, which may include
placeholders for arguments and options. The actual command is the specific
instance of the command with all the arguments and options filled in.
The formatting of an actual command is a code block, with the command
prefixed by a dollar sign ($) to indicate that it is a command to be run in
the terminal. It can also include comments with a hash sign (#) to explain
the command or provide additional information, just like in a regular
terminal session.
Signed-off-by: Jean-Noël Avila <redacted> Signed-off-by: Junio C Hamano <redacted>
Jean-Noël Avila [Mon, 25 May 2026 10:28:22 +0000 (10:28 +0000)]
doc: convert git-bisect to synopsis style
Convert Documentation/git-bisect.adoc to the modern synopsis style.
- Replace [verse] with [synopsis] in the SYNOPSIS block
- Remove single quotes around command names in the synopsis
- Use backticks for inline commands, options, refs, and special values
- Apply [synopsis] attribute to in-body command-form code blocks
- Format OPTIONS entries with backtick-quoted terms and direct
- Add synopsis-style formatting to listing blocks
- Format man page references as `command`(N)
Signed-off-by: Jean-Noël Avila <redacted> Signed-off-by: Junio C Hamano <redacted>
Junio C Hamano [Mon, 25 May 2026 02:58:48 +0000 (11:58 +0900)]
SubmittingPatches: proactively monitor GHCI pages
Even those contributors who do not come from GGG and do not first
push their changes to their repositories on GitHub with CI enabled,
can still monitor the CI runs triggered by integration of their
topic to 'seen' and other branches to notice a breakage their topic
caused to the system.
Encourage them to help the project by keeping an eye on these CI
runs.
Junio C Hamano [Mon, 25 May 2026 00:40:08 +0000 (09:40 +0900)]
Merge branch 'jk/dumb-http-alternate-fix'
The HTTP walker misinterpreted the alternates file that gives an
absolute path when the server URL does not have the final slash
(i.e., "https://example.com" not "https://example.com/").
* jk/dumb-http-alternate-fix:
http: handle absolute-path alternates from server root
Junio C Hamano [Mon, 25 May 2026 00:40:08 +0000 (09:40 +0900)]
Merge branch 'jk/pretty-no-strbuf-presizing'
Remove ineffective strbuf presizing that would have computed an
allocation that would not have fit in the available memory anyway,
or too small due to integer wraparound to cause immediate automatic
growing.
* jk/pretty-no-strbuf-presizing:
pretty: drop strbuf pre-sizing from add_rfc2047()
Junio C Hamano [Mon, 25 May 2026 00:40:07 +0000 (09:40 +0900)]
Merge branch 'mm/diff-U-takes-no-negative-values'
The command line parser for "git diff" learned a few options take
only non-negative integers.
* mm/diff-U-takes-no-negative-values:
parse-options: clarify what "negated" means for PARSE_OPT_NONEG
xdiff: guard against negative context lengths
diff: reject negative values for -U/--unified
diff: reject negative values for --inter-hunk-context
Adam Johnson [Fri, 22 May 2026 23:12:25 +0000 (23:12 +0000)]
stash: reuse cached index entries in --patch temporary index
`git stash -p` prepares the interactive selection by creating a
temporary index at HEAD, switching `GIT_INDEX_FILE` to it, and then
running the `add -p` machinery.
That temporary index was created by running `git read-tree HEAD`. The
resulting index had no useful cached stat data or fsmonitor-valid bits
from the real index. When `run_add_p()` refreshed that temporary index
before showing the first prompt, it could end up lstat(2)-ing every
tracked file, even in a repository where `git diff` and `git restore -p`
can use fsmonitor to avoid that work.
Create the temporary index in-process instead. Use `unpack_trees()` to
reset the real index contents to HEAD while writing the result to the
temporary index path. For paths whose index entries already match HEAD,
`oneway_merge()` reuses the existing cache entries, preserving their
cached stat data and `CE_FSMONITOR_VALID` state.
This makes the refresh performed by `run_add_p()` behave like the one
used by `git restore -p`: unchanged paths can be skipped via fsmonitor
instead of being scanned again.
In a 206k file repository with `core.fsmonitor` enabled and a one-line
change in one file, time to first prompt dropped from 34.774 seconds to
0.659 seconds. The new perf test file demonstrates similar improvements,
with maen times for without- and with-fsmonitor cases dropping from 6.90
and 6.83 seconds to 0.55 and 0.28 seconds, respectively.
Signed-off-by: Adam Johnson <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Fri, 22 May 2026 18:24:37 +0000 (18:24 +0000)]
path-walk: support `combine` filter
The `combine` filter takes the intersection of its children, that is:
objects are shown only when all child filters would admit the object.
The preceding patches added support for many individual filter types.
Enable users to compose these filters by implementing support for the
`combine` filter type.
Mapping intersection onto path_walk_info works because every supported
child filter is a monotonic restriction:
- `blob:none`, `tree:0` unconditionally clear `info->blobs` and (for
`tree:0`) `info->trees`; clearing an already-cleared flag is a
no-op.
- `object:type=X` is now expressed as an AND of each type flag with the
filtered type, so applying multiple such filters only refines the
existing set rather than overwrites it.
- `blob:limit=N` has to compose too: the intersection of "size < L1"
and "size < L2" is "size < min(L1, L2)".
Update the `LOFC_BLOB_LIMIT` handler to take the running minimum when
`info->blob_limit` is already set, so a combined filter with, e.g.,
both "blob:limit=10" and "blob:limit=5" produces a limit of 5
regardless of ordering.
- `sparse:oid` is left unchanged. A `combine` filter that includes a
`sparse:oid` is allowed at most once, since the existing handler
refuses to overwrite `info->pl`. Two `sparse:oid` filters in a single
`combine` would be unusual and are rejected with a warning, matching
the standalone `sparse:oid` behavior.
Implementation-wise, the existing `prepare_filters()` called
`list_objects_filter_release()` inside each case branch. That works fine for
top-level filters, but `combine` filters need to recurse over its child
filters without releasing each one in turn (since the parent's release
iterates the sub array). Split `prepare_filters()` into a recursive helper
that performs only the mutation, plus a thin wrapper that calls the helper
and then releases the top-level filter once.
The `LOFC_COMBINE` case in the helper just walks `sub_nr` and recurses;
child filters are released by the wrapper's single
`list_objects_filter_release()` call on the parent (which itself recursively
releases each sub-filter, the same way it always has).
If any sub-filter is unsupported (e.g. "tree:1", "sparse:<path>", or a
not-yet-supported choice), the recursion bubbles a failure up and the
existing pack-objects/backfill fallback paths kick in.
Add coverage in t6601:
- "combine:blob:none+tree:0" collapses to "tree:0"
- "combine:object:type=blob+blob:limit=3" yields only the blobs
smaller than three bytes
- "combine:object:type=blob+object:type=tree" intersects to empty
- "combine:tree:1+blob:none" reports the "tree:1" error.
Update Documentation/git-pack-objects.adoc to add combine to the
list of supported --filter forms.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Fri, 22 May 2026 18:24:36 +0000 (18:24 +0000)]
path-walk: support `object:type` filter
The `object:type` filter accepts only objects of a single type; it is
the second member of the object-info-only filter family that bitmap
traversal already supports.
Like `blob:none` and `tree:0`, it can be evaluated with nothing more
than the object's type, which is exactly the granularity path-walk's
existing info->{commits,trees,blobs,tags} flags already control.
Map `LOFC_OBJECT_TYPE` in `prepare_filters()` by AND-ing each flag
against the filtered type. A single `object:type=X` filter
applied to the default info (all flags = 1) leaves `info->X = 1` and
all the others 0, which is what we want.
Using an AND rather than straight assignment prepares us for a
subsequent change to implement combined object filters.
The path-walk machinery is mostly already wired for the per-type
distinction:
- `walk_path()` calls `path_fn` for a batch only when the corresponding
`info->X` flag is set, so unwanted types are silently not reported.
- `add_tree_entries()` skips tree entries of type `OBJ_BLOB` when
`info->blobs` is unset, so we don't even allocate paths for them.
- The commit-walk loop short-circuits the root-tree fetch when
`!info->trees && !info->blobs`, so commit-only filters don't descend
into trees at all.
But there are a couple of side effects of the "trees off, blobs on" case
that need fixing:
1. 'setup_pending_objects()' previously skipped pending trees as soon
as `info->trees` was zero. For 'object:type=blob' the call site
needs those pending trees: a lightweight tag pointing to a tree, or
an annotated tag whose peeled target is a tree, can both reach
blobs that are otherwise unreachable from any commit's root tree.
Loosen the gate to "if (!info->trees && !info->blobs) continue" and
similarly retrieve the root_tree_list whenever either trees or
blobs are wanted.
2. The revision machinery's `handle_commit()` drops pending trees when
`revs->tree_objects` is zero (see the 'OBJ_TREE' handler in
revision.c), so by the time path-walk sees the pending list
after `prepare_revision_walk()` the tree-bearing pendings would
already be gone. Fix this by setting
revs->tree_objects = info->trees || info->blobs
so pending trees survive `prepare_revision_walk()` whenever we
need to walk into them. Path-walk still resets tree_objects to
zero immediately after `prepare_revision_walk()` returns, so the
rev-walk itself never enumerates trees redundantly with
path-walk's own descent.
Add coverage in t6601 for each of the four `object:type` values. The
'object:type=blob' test in particular asserts that file2 and child/file
(both reachable only through tag-pointed trees) show up in the output,
exercising the pending-tree fix.
Update Documentation/git-pack-objects.adoc to add object:type to
the list of supported --filter forms.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Taylor Blau [Fri, 22 May 2026 18:24:35 +0000 (18:24 +0000)]
path-walk: support `tree:0` filter
The `tree:0` object filter omits all trees and blobs from the result,
keeping only commits and tags. Consequently, this filter type should
has a fairly straightforward integration with path-walk, as the decision
to include an object depends only on its type and does not depend on any
path-sensitive state.
Mapping it onto `path_walk_info` is direct: set `info->trees = 0` and
`info->blobs = 0` in `prepare_filters()` when the `LOFC_TREE_DEPTH`
choice is requested with depth zero. The existing code already plumbs
those flags through the rest of the walk:
- 'walk_objects_by_path()' sets `revs->blob_objects = info->blobs` and
`revs->tree_objects = info->trees` before `prepare_revision_walk()`,
so the revision walk doesn't try to enumerate trees or blobs itself.
- The commit-walk loop short-circuits the root-tree fetch with
"if (!info->trees && !info->blobs) continue;", so we never even
look up the root tree, let alone descend into it.
- `setup_pending_objects()` skips pending trees and blobs based on
the same flags.
This means the path-walk doesn't allocate or expand any tree structures
at all under `tree:0`, which matches the intended behavior of the
filter.
However, this requires first fixing some issues with how the path-walk
API handles directly-requested trees _and_ trees requested through
lightweight tags. These changes create substantial updates to
t6601-path-walk.sh, which the previous change highlighted as a problem
by tagging otherwise-unreachable trees and having them not appear in the
output.
Non-zero tree-depth filters are not supported. Those depend on the depth
at which a tree is visited, which is a path-walk concept the filter
machinery doesn't currently share with the path-walk API. Reject them in
`prepare_filters()` with a helpful error and let pack-objects fall back
to the regular traversal, the same way it already does for unsupported
filters.
Add coverage in t6601 for both `--all` and a single-branch case to
confirm that no trees or blobs are emitted, and a separate test that
`tree:1` is rejected with the expected error message. Place the new
tests before "setup sparse filter blob" so they run on the original set
of refs, before the orphan branch that the sparse-tree tests create.
Signed-off-by: Taylor Blau <redacted> Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Derrick Stolee [Fri, 22 May 2026 18:24:34 +0000 (18:24 +0000)]
t6601: tag otherwise-unreachable trees
The tests in t6601-path-walk.sh demonstrate the behavior of the
path-walk API under different conditions. One thing that I noticed while
updating the behavior of directly-requested objects is that we don't
actually emit tagged trees. This was previously not noticed due to those
tagged trees actually being reachable from commits that we are including
in the path-walk.
Update the test setup to have tree-tag and tree-tag2 point to trees that
are otherwise unreachable.
It is worth noting that this does not meaningfully change any of the
other test cases, demontrating the bug.
Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Derrick Stolee [Fri, 22 May 2026 18:24:33 +0000 (18:24 +0000)]
pack-objects: support sparse:oid filter with path-walk
The --filter=sparse:<oid> option to 'git pack-objects' allows focusing
an object set to a sparse-checkout definition. This reduces the set of
matching blobs while retaining all reachable trees. No server currently
supports fetching with this filter because it is expensive to compute
and reachability bitmaps do not help without a significant effort to
extend the bitmap feature to store bitmaps for each supported sparse-
checkout definition.
Without focusing on serving fetches and clones with these filters, there
are still benefits that could be realized by making this faster. With
the sparse index, it's more realistic now than ever to be able to
operate a local clone that was bootstrapped by a packfile created with
a sparse filter, because the missing trees are not needed to move a
sparse-checkout from one commit to another or to view the history of any
path in scope. Such clones could perhaps be bootstrapped by partial
bundles.
Previously, constructing these sparse packs has been incredibly
computationally inefficient. The revision walk that explores which
objects are in scope spends a lot of time checking each object to see if
it matches the sparse-checkout patterns, causing quadratic behavior
(number of objects times number of sparse-checkout patterns). This
improves somewhat when using cone-mode sparse-checkout patterns that can
use hashtables and prefix matches to determine containment. However, the
check per object is still too expensive for most cases.
This is where the path-walk feature comes in. We can proceed as normal
by placing objects in bins by path and _then_ check a group of objects
all at once. Since sparse:<oid> only restricts blobs, the path-walk must
include all reachable trees while using the cone-mode patterns to skip
blobs at paths outside the sparse scope. This establishes a baseline for
a potential future "treesparse:<oid>" filter that would also restrict
trees, but introducing such a new filter is deferred to a later change.
The implementation here is focused around loading the sparse-checkout
patterns from the provided object ID and checking that the patterns are
indeed cone-mode patterns. We can then load the correct pattern list
into the path walk context and use the logic that already exists from bff45557675 (backfill: add --sparse option, 2025-02-03), though that
feature loads sparse-checkout patterns from the worktree's local
settings and also restricts tree objects. We use a combination of errors
and warnings to signal problems during this load. The difference is that
errors are likely fatal for the non-path-walk version while the warnings
are probably just implementation details for the path-walk version and
the 'git pack-objects' command can fall back to the revision walk
version.
Now that the SEEN flag is deferred until after pattern checks (from the
previous commit), handle the case where a tree with a shared OID appears
at both an out-of-cone and in-cone path. When trees are not being pruned
(pl_sparse_trees == 0), the path-walk re-walks the tree at the in-cone
path so that in-cone blobs within it are discovered. The new tests in
t5317 and t6601 demonstrate this behavior and would fail without these
changes.
The performance test p5315 shows the impact of this change when using
sparse filters:
These performance tests were run on the Git repository. The --path-walk
feature shows meaningful space savings (14% smaller for sparse packs)
and dramatic time savings (60% faster) by leveraging the path-walk's
ability to skip blobs outside the sparse scope.
Co-authored-by: Taylor Blau <redacted> Signed-off-by: Taylor Blaue <redacted> Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
Derrick Stolee [Fri, 22 May 2026 18:24:32 +0000 (18:24 +0000)]
path-walk: add pl_sparse_trees to control tree pruning
The path-walk API prunes trees and blobs when a sparse-checkout pattern
list is provided, which is the correct behavior for 'git backfill
--sparse' since it only needs to fill in objects at paths within the
sparse cone.
However, a future change will use the path-walk API with a sparse:<oid>
filter that restricts only blobs while retaining all reachable trees.
To support both behaviors, add a 'pl_sparse_trees' flag to
path_walk_info. When set (as in 'git backfill --sparse' and the
--stdin-pl test helper mode), the sparse patterns prune both trees and
blobs. When unset, only blobs are filtered and all trees are walked and
reported.
Additionally, move the SEEN flag assignment in add_tree_entries() to
after the sparse pattern and pathspec checks. Previously, SEEN was set
immediately upon discovering an object, before checking whether its path
matched the sparse patterns. When the same object ID appeared at
multiple paths (e.g. sibling directories with identical contents), the
first path to be visited would mark the object as SEEN. If that path was
outside the sparse cone, the object would be skipped there but also
never discovered at its in-cone path.
By deferring the SEEN flag until after the checks pass, objects that are
skipped due to sparse filtering remain discoverable at other paths where
they may be in scope.
Signed-off-by: Derrick Stolee <redacted> Signed-off-by: Junio C Hamano <redacted>
git.99rst.org / git.git / log