The bulk-checkin subsystem provides various functions to manage ODB
transactions. Apart from {begin,end}_odb_transaction(), these functions
are only used by the object-file subsystem to manage aspects of a
transaction implementation specific to the files object source.
Relocate all the transaction code in bulk-checkin to object-file. This
simplifies the exposed transaction interface by reducing it to only
{begin,end}_odb_transaction(). Function and type names are adjusted in
the subsequent commit to better fit the new location.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Object database transactions can be explicitly flushed via
flush_odb_transaction() without actually completing the transaction.
This makes the provided transactional interface a bit awkward. Now that
there are no longer any flush_odb_transaction() call sites, drop the
function to simplify the interface and further ensure that a transaction
is only finalized when end_odb_transaction() is invoked.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With 23a3a303 (update-index: use the bulk-checkin infrastructure,
2022-04-04), object database transactions were added to
git-update-index(1) to facilitate writing objects in bulk. With
transactions, newly added objects are instead written to a temporary
object directory and migrated to the primary object database upon
transaction commit.
When the --verbose option is specified, the subsequent set of objects
written are explicitly flushed via flush_odb_transaction() prior to
reporting the update. Flushing the object database transaction migrates
pending objects to the primary object database without marking the
transaction as complete. This is done so objects are immediately visible
to git-update-index(1) callers using the --verbose option and that rely
on parsing verbose output to know when objects are written.
Due to how git-update-index(1) parses arguments, options that come after
a filename are not considered during the object update. Therefore, it
may not be known ahead of time whether the --verbose option is present
and thus object writes are considered transactional by default until a
--verbose option is parsed.
Flushing a transaction after individual object writes negates the
benefit of writing objects to a transaction in the first place.
Furthermore, the mechanism to flush a transaction without actually
committing is rather awkward. Drop the call to flush_odb_transaction()
in favor of ending the transaction altogether when the --verbose flag is
encountered. Subsequent object writes occur outside of a transaction and
are therefore immediately visible which matches the current behavior.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
ODB transactions support being nested. Only the outermost
{begin,end}_odb_transaction() start and finish a transaction. This
allows internal object write codepaths to be optimized with ODB
transactions without worrying about whether a transaction is already
active. When {begin,end}_odb_transaction() is invoked during an active
transaction, these operations are essentially treated as no-ops. This
can make the interface a bit awkward to use, as calling
end_odb_transaction() does not guarantee that a transaction is actually
ended. Thus, in situations where a transaction needs to be explicitly
flushed, flush_odb_transaction() must be used.
To remove the need for an explicit transaction flush operation via
flush_odb_transaction() and better clarify transaction semantics, drop
the transaction nesting mechanism in favor of begin_odb_transaction()
returning a NULL transaction value to signal it was a no-op, and
end_odb_transaction() behaving as a no-op when a NULL transaction value
is passed. This is safe for existing callers as the transaction value
wired to end_odb_transaction() already comes from
begin_odb_transaction() and thus continues the same no-op behavior when
a transaction is already pending. With this model, passing a pending
transaction to end_odb_transaction() ensures it is committed at that
point in time.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* jt/de-global-bulk-checkin:
bulk-checkin: use repository variable from transaction
bulk-checkin: require transaction for index_blob_bulk_checkin()
bulk-checkin: remove global transaction state
bulk-checkin: introduce object database transaction structure
A new command "git last-modified" has been added to show the closest
ancestor commit that touched each path.
* tc/last-modified:
last-modified: use Bloom filters when available
t/perf: add last-modified perf script
last-modified: new subcommand to show when files were last modified
"git ls-files <pathspec>..." should not necessarily have to expand
the index fully if a sparsified directory is excluded by the
pathspec; the code is taught to expand the index on demand to avoid
this.
* ds/ls-files-lazy-unsparse:
ls-files: conditionally leave index sparse
"git repack --path-walk" lost objects in some corner cases, which
has been corrected.
* ds/path-walk-repack-fix:
path-walk: create initializer for path lists
path-walk: fix setup of pending objects
Inspired by Ezekiel's recent effort to showcase Rust interface, the
hash function implementation used to hash lines have been updated
to the one used for ELF symbol lookup by Glibc.
* am/xdiff-hash-tweak:
xdiff: optimize xdl_hash_record_verbatim
xdiff: refactor xdl_hash_record()
Makefile tried to run multiple "cargo build" which would not work
very well; serialize their execution to work it around.
* da/cargo-serialize:
Makefile: build libgit-rs and libgit-sys serially
Update the instruction to use of GGG in the MyFirstContribution
document to say that a GitHub PR could be made against `git/git`
instead of `gitgitgadget/git`.
* ds/doc-ggg-pr-fork-clarify:
doc: clarify which remotes can be used with GitGitGadget
Fix missing single-quote pairs in a documentation page.
* kh/doc-interpret-trailers-markup-fix:
doc: interpret-trailers: close all pairs of single quotes
"git describe <blob>" misbehaves and/or crashes in some corner
cases, which has been taught to exit with failure gracefully.
* jk/describe-blob:
describe: pass commit to describe_commit()
describe: handle blob traversal with no commits
describe: catch unborn branch in describe_blob()
describe: error if blob not found
describe: pass oid struct by const pointer
"git fetch" can clobber a symref that is dangling when the
remote-tracking HEAD is set to auto update, which has been
corrected.
* jk/no-clobber-dangling-symref-with-fetch:
refs: do not clobber dangling symrefs
t5510: prefer "git -C" to subshell for followRemoteHEAD tests
t5510: stop changing top-level working directory
t5510: make confusing config cleanup more explicit
Discord has been added to the first contribution documentation as
another way to ask for help.
* ds/doc-community-discord:
doc: add discord to ways of getting help
Code clean-ups.
* ps/reftable-libgit2-cleanup:
refs/reftable: always reload stacks when creating lock
reftable: don't second-guess errors from flock interface
reftable/stack: handle outdated stacks when compacting
reftable/stack: allow passing flags to `reftable_stack_add()`
reftable/stack: fix compiler warning due to missing braces
reftable/stack: reorder code to avoid forward declarations
reftable/writer: drop Git-specific `QSORT()` macro
reftable/writer: fix type used for number of records
Our 'git last-modified' performs a revision walk, and computes a diff at
each point in the walk to figure out whether a given revision changed
any of the paths it considers interesting.
When changed-path Bloom filters are available, we can avoid computing
many such diffs. Before computing a diff, we first check if any of the
remaining paths of interest were possibly changed at a given commit by
consulting its Bloom filter. If any of them are, we are resigned to
compute the diff.
If none of those queries returned "maybe", we know that the given commit
doesn't contain any changed paths which are interesting to us. So, we
can avoid computing it in this case.
Comparing the perf test results on git.git:
Test HEAD~ HEAD
------------------------------------------------------------------------------------
8020.1: top-level last-modified 4.49(4.34+0.11) 2.22(2.05+0.09) -50.6%
8020.2: top-level recursive last-modified 5.64(5.45+0.11) 5.62(5.30+0.11) -0.4%
8020.3: subdir last-modified 0.11(0.06+0.04) 0.07(0.03+0.04) -36.4%
Based-on-patch-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Toon Claes <toon@iotcl.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This just runs some simple last-modified commands. We already test
correctness in the regular suite, so this is just about finding
performance regressions from one version to another.
Based-on-patch-by: Jeff King <peff@peff.net>
Signed-off-by: Toon Claes <toon@iotcl.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Similar to git-blame(1), introduce a new subcommand
git-last-modified(1). This command shows the most recent modification to
paths in a tree. It does so by expanding the tree at a given commit,
taking note of the current state of each path, and then walking
backwards through history looking for commits where each path changed
into its final commit ID.
Based-on-patch-by: Jeff King <peff@peff.net>
Improved-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Toon Claes <toon@iotcl.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The compatObjectFormat extension is used to hide an incomplete
feature that is not yet usable for any purpose other than
developing the feature further. Document it as such to discourage
its use by mere mortals.
* bc/doc-compat-object-format-not-working:
docs: note that extensions.compatobjectformat is incomplete
Under a race against another process that is repacking the
repository, especially a partially cloned one, "git fetch" may
mistakenly think some objects we do have are missing, which has
been corrected.
* jk/fetch-check-graph-objects-fix:
fetch-pack: re-scan when double-checking graph objects
"git log -L..." compared trees of multiple parents with the tree of the
merge result in an unnecessarily inefficient way.
* sg/line-log-merge-optim:
line-log: simplify condition checking for merge commits
line-log: initialize diff queue in process_ranges_ordinary_commit()
line-log: get rid of the parents array in process_ranges_merge_commit()
line-log: avoid unnecessary tree diffs when processing merge commits
The start_delayed_progress() function in the progress eye-candy API
did not clear its internal state, making an initial delay value
larger than 1 second ineffective, which has been corrected.
* js/progress-delay-fix:
progress: pay attention to (customized) delay time
Documentation for "git rebase" has been updated.
* je/doc-rebase:
doc: git-rebase: update discussion of internals
doc: git-rebase: move --onto explanation down
doc: git rebase: clarify arguments syntax
doc: git rebase: dedup merge conflict discussion
doc: git-rebase: start with an example
When running 'git ls-files' with a pathspec, the index entries get
filtered according to that pathspec before iterating over them in
show_files(). In 78087097b8 (ls-files: add --sparse option,
2021-12-22), this iteration was prefixed with a check for the '--sparse'
option which allows the command to output directory entries; this
created a pre-loop call to ensure_full_index().
However, when a user runs 'git ls-files' where the pathspec matches
directories that are recursively matched in the sparse-checkout, there
are not any sparse directories that match the pathspec so they would not
be written to the output. The expansion in this case is just a
performance drop for no behavior difference.
Replace this global check to expand the index with a check inside the
loop for a matched sparse directory. If we see one, then expand the
index and continue from the current location. This is safe since the
previous entries in the index did not have any sparse directories and
thus would remain stable in this expansion.
A test in t1092 confirms that this changes the behavior.
Signed-off-by: Derrick Stolee <stolee@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"make -JN" with INCLUDE_LIBGIT_RS enabled causes cargo lock warnings
and can trigger ld errors during the build.
The build errors are caused by two inner "make" invocations getting
triggered concurrently: once inside of libgit-sys and another inside of
libgit-rs.
Make libgit-rs depend on libgit-sys so that "make" prevents them
from running concurrently. Apply the same logic to the test invocations.
Use cargo's "--manifest-path" option instead of "cd" in the recipes.
Signed-off-by: David Aguilar <davvid@gmail.com>
Acked-by: Kyle Lippincott <spectral@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The compatibility object format is only implemented for loose objects,
not packed objects, so anyone attempting to push or fetch data into a
repository with this option will likely not see it work as expected. In
addition, the underlying storage of loose object mapping is likely to
change because the current format is inefficient and does not handle
important mapping information such as that of submodules.
It would have been preferable to initially document that this was not
yet ready for prime time, but we did not do so. We hinted at the fact
that this functionality is incomplete in the description, but did not
say so explicitly. Let's do so now: indicate that this feature is
incomplete and subject to change and that the option is not designed to
be used by end users.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Using one of the start_delayed_*() functions, clients of the progress
API can request that a progress meter is only shown after some time.
To do that, the implementation intends to count down the number of
seconds stored in struct progress by observing flag progress_update,
which the timer interrupt handler sets when a second has elapsed. This
works during the first second of the delay. But the code forgets to
reset the flag to zero, so that subsequent calls of display_progress()
think that another second has elapsed and decrease the count again
until zero is reached. Due to the frequency of the calls, this happens
without an observable delay in practice, so that the effective delay is
always just one second.
This bug has been with us since the inception of the feature. Despite
having been touched on various occasions, such as 8aade107dd
(progress: simplify "delayed" progress API), 9c5951cacf (progress:
drop delay-threshold code), and 44a4693bfc (progress: create
GIT_PROGRESS_DELAY), the short delay went unnoticed.
Copy the flag state into a local variable and reset the global flag
right away so that we can detect the next clock tick correctly.
Since we have not had any complaints that the delay of one second is
too short nor that GIT_PROGRESS_DELAY is ignored, people seem to be
comfortable with the status quo. Therefore, set the default to 1 to
keep the current behavior.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A new subcommand "git repo" gives users a way to grab various
repository characteristics.
* lo/repo-info:
repo: add the --format flag
repo: add the field layout.shallow
repo: add the field layout.bare
repo: add the field references.format
repo: declare the repo command
Remove dependency on the_repository and other globals from the
commit-graph code, and other changes unrelated to de-globaling.
* ps/commit-graph-wo-globals:
commit-graph: stop passing in redundant repository
commit-graph: stop using `the_repository`
commit-graph: stop using `the_hash_algo`
commit-graph: refactor `parse_commit_graph()` to take a repository
commit-graph: store the hash algorithm instead of its length
commit-graph: stop using `the_hash_algo` via macros
Test clean-up.
* dk/t7005-editor-updates:
t7005: sanitize test environment for subsequent tests
t7005: stop abusing --exec-path
t7005: use modern test style
Doc lint updates to encourage the newer and easier-to-use
`synopsis` format, with fixes to a handful of existing uses.
* ja/doc-lint-sections-and-synopsis:
doc lint: check that synopsis manpages have synopsis inlines
doc:git-for-each-ref: fix styling and typos
doc: check for absence of the form --[no-]parameter
doc: check for absence of multiple terms in each entry of desc list
doc: check well-formedness of delimited sections
doc: test linkgit macros for well-formedness
"git cmd --help-all" now works outside repositories.
* dk/help-all:
builtin: also setup gently for --help-all
parse-options: refactor flags for usage_with_options_internal
Revert back to “Git's” which was used before d30c5cc459 (doc: convert
git-mergetool options to new synopsis style, 2025-05-25) accidentally
changed it.
Signed-off-by: Kristoffer Haugsbakk <code@khaugsbakk.name>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The fetch code tries to avoid asking the remote side for an object we
already have. It does this by traversing recent commits reachable from
our refs looking for matches. Commit 5d4cc78f72 (fetch-pack: die if in
commit graph but not obj db, 2024-11-05) introduced an extra check
there: if we think we have an object because it's in the commit graph,
we double-check that we actually have it in our object database with a
call to odb_has_object().
But that call does not pass any flags, and so the function won't call
reprepared_packed_git() if it does not find the object. That opens us up
to the usual race against some other process repacking the odb:
1. We scan the list of packs in objects/pack but haven't yet opened them.
2. Somebody else packs the object into a new pack (which we don't know
about), and deletes the old pack it was in.
3. Our odb_has_object() calls tries to open that old pack, but finds it
is gone. We declare that we don't have the object.
And this causes us to erroneously complain and abort the fetch, thinking
our commit-graph and object database are out of sync. Instead, we should
pass HAS_OBJECT_RECHECK_PACKED, which will add a new step:
4. We re-scan the pack directory again, find the new pack, and locate
the object.
Often the fetch code tries to avoid these kinds of re-scans if it's
likely that we won't have the object. If the other side has told us
about object X and we want to know if we have it, we'll skip the re-scan
(to avoid spending a lot of effort when there are many such objects). We
can accept the racy false negative in that case because the worst case
is that we ask the other side to send us the object.
But this is not one of those cases. These are objects which are
accessible from _our_ refs, and which we already found in the commit
graph file. We should have them, and if we don't, we'll die()
immediately. So the performance impact is negligible, and getting the
right answer is important.
There's no test here because it's inherently racy. In fact, I had
trouble even developing a minimal test. The problem seen in the wild can
be produced like this:
# Any git.git mirror which supports partial clones; I think this
# should work with any repo that contains submodules, but note that
# $obj below is specific to this repo
url=https://github.com/git/git.git
# This is a commit that is not at the tip of any branches (so after
# we have it, we'll still have some commits to fetch).
obj=cf6f63ea6bf35173e02e18bdc6a4ba41288acff9
git init
git fetch --filter=tree:0 $url $obj:refs/heads/foo
git checkout foo
git commit-graph write --reachable
git fetch $url
What happens here is that the initial fetch grabs that older commit (and
its ancestors) but no trees or blobs, and the subsequent checkout grabs
the necessary trees and blobs just for that commit. The final fetch
spawns a long sequence of child fetches due to fetch_submodules(), which
wants to check whether there have been any gitlink modifications which
should trigger a fetch of the related submodule (we'll leave aside the
irony that we did not even check out any submodules yet).
That series of fetches causes us to accumulate packs, which eventually
triggers background maintenance to run. That repacks all-into-one, and
the pack containing $obj goes away in favor of a new pack. And then the
fetch eventually fails with:
fatal: You are attempting to fetch cf6f63ea6b, which is in the commit graph file but
not in the object database.
In the scenario above, the race becomes likely because of the long
series of quick fetches. But I _think_ the bug is independent of partial
clones entirely, and you could run into the same thing with a single
fetch, some other process running "git repack" simultaneously, and a bit
of bad luck. I haven't been able to reproduce, though. I'm not sure if
that's because there's some mis-analysis above, or if the race window is
just small enough that it's hard to trigger.
At any rate, re-scanning here seems like an obviously correct thing to
do with no downside, and it does fix the partial-clone case shown above.
Reported-by: Дилян Палаузов <dilyan.palauzov@aegee.org>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are three tips how to compose a non-line-wrapped patch with
Thunderbird. The first one suggests use of an add-on. The one
referenced has long been superseded by a different one. Update the
link to the new one. Mention that additional configuration is
required to make the add-on work.
Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The bulk-checkin subsystem depends on `the_repository`. Adapt functions
and call sites to access the repository through `struct odb_transaction`
instead. The `USE_THE_REPOSITORY_VARIBALE` is still required as the
`pack_compression_level` and `pack_size_limit_cfg` globals are still
used.
Also adapt functions using packfile state to instead access it through
the transaction. This makes some function parameters redundant and go
away.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The bulk-checkin subsystem provides a mechanism to write blobs directly
to a packfile via `index_blob_bulk_checkin()`. If there is an ongoing
transaction when invoked, objects written via this function are stored
in the same packfile. The packfile is not flushed until the transaction
itself is flushed. If there is no transaction, the single object is
written to a packfile and immediately flushed. This complicates
`index_blob_bulk_checkin()` as it cannot reliably use the provided
transaction to get the associated repository.
Update `index_blob_bulk_checkin()` to assume that a valid transaction is
always provided. Callers are now expected to ensure a transaction is set
up beforehand. With this simplification, `deflate_blob_bulk_checkin()`
is no longer needed as a standalone internal function and is combined
with `index_blob_bulk_checkin()`. The single call site in
`object-file.c:index_fd()` is updated accordingly. Due to how
`{begin,end}_odb_transaction()` handles nested transactions, a new
transaction is only created and committed if there is not already an
ongoing transaction.
Signed-off-by: Justin Tobler <jltobler@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
