When iterating over records with the merged iterator we put the records
into a priority queue before yielding them to the caller. This means
that we need to allocate the contents of these records before we can
pass them over to the caller.
The handover to the caller is quite inefficient though because we first
deallocate the record passed in by the caller and then copy over the new
record, which requires us to reallocate memory.
Refactor the code to instead transfer ownership of the new record to the
caller. So instead of reallocating all contents, we now release the old
record and then copy contents of the new record into place.
The following benchmark of `git show-ref --quiet` in a repository with
around 350k refs shows a clear improvement. Before:
HEAP SUMMARY:
in use at exit: 21,163 bytes in 193 blocks
total heap usage: 708,058 allocs, 707,865 frees, 36,783,255 bytes allocated
After:
HEAP SUMMARY:
in use at exit: 21,163 bytes in 193 blocks
total heap usage: 357,007 allocs, 356,814 frees, 24,193,602 bytes allocated
This shows that we now have roundabout a single allocation per record
that we're yielding from the iterator. Ideally, we'd also get rid of
this allocation so that the number of allocations doesn't scale with the
number of refs anymore. This would require some larger surgery though
because the memory is owned by the priority queue before transferring it
over to the caller.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 829231dc20 (reftable/merged: reuse buffer to compute record keys,
2023-12-11), we have refactored the merged iterator to reuse a pair of
long-living strbufs by relying on the fact that `reftable_record_key()`
tries to reuse already allocated strbufs by calling `strbuf_reset()`,
which should give us significantly fewer reallocations compared to the
old code that used on-stack strbufs that are allocated for each and
every iteration. Unfortunately, we called `strbuf_release()` on these
long-living strbufs that we meant to reuse on each iteration, defeating
the optimization.
Fix this performance issue by not releasing those buffers on iteration
anymore, where we instead rely on `merged_iter_close()` to release the
buffers for us.
Using `git show-ref --quiet` in a repository with ~350k refs this leads
to a significant drop in allocations. Before:
HEAP SUMMARY:
in use at exit: 21,163 bytes in 193 blocks
total heap usage: 1,410,148 allocs, 1,409,955 frees, 61,976,068 bytes allocated
After:
HEAP SUMMARY:
in use at exit: 21,163 bytes in 193 blocks
total heap usage: 708,058 allocs, 707,865 frees, 36,783,255 bytes allocated
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Similar to the preceding commit, convert ref records of type "val2" to
store their object IDs in static arrays instead of allocating them for
every single record.
We're using the same benchmark as in the preceding commit, with `git
show-ref --quiet` in a repository with ~350k refs. This time around
though the effects aren't this huge. Before:
HEAP SUMMARY:
in use at exit: 21,163 bytes in 193 blocks
total heap usage: 1,419,040 allocs, 1,418,847 frees, 62,153,868 bytes allocated
After:
HEAP SUMMARY:
in use at exit: 21,163 bytes in 193 blocks
total heap usage: 1,410,148 allocs, 1,409,955 frees, 61,976,068 bytes allocated
This is because "val2"-type records are typically only stored for peeled
tags, and the number of annotated tags in the benchmark repository is
rather low. Still, it can be seen that this change leads to a reduction
of allocations overall, even if only a small one.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When reading ref records of type "val1", we store its object ID in an
allocated array. This results in an additional allocation for every
single ref record we read, which is rather inefficient especially when
iterating over refs.
Refactor the code to instead use an embedded array of `GIT_MAX_RAWSZ`
bytes. While this means that `struct ref_record` is bigger now, we
typically do not store all refs in an array anyway and instead only
handle a limited number of records at the same point in time.
Using `git show-ref --quiet` in a repository with ~350k refs this leads
to a significant drop in allocations. Before:
HEAP SUMMARY:
in use at exit: 21,098 bytes in 192 blocks
total heap usage: 2,116,683 allocs, 2,116,491 frees, 76,098,060 bytes allocated
After:
HEAP SUMMARY:
in use at exit: 21,098 bytes in 192 blocks
total heap usage: 1,419,031 allocs, 1,418,839 frees, 62,145,036 bytes allocated
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We're about to convert reftable records to stop storing their object IDs
as allocated hashes. Prepare for this refactoring by constifying some
parts of the interface that will be impacted by this.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Each reftable may contain multiple types of blocks for refs, objects and
reflog records, where each of these may have an index that makes it more
efficient to find the records. It was observed that the index for log
records can become corrupted under certain circumstances, where the
first entry of the index points into the object index instead of to the
log records.
As it turns out, this corruption can occur whenever we write a log index
as well as at least one additional index. Writing records and their index
is basically a two-step process:
1. We write all blocks for the corresponding record. Each block that
gets written is added to a list of blocks to index.
2. Once all blocks were written we finish the section. If at least two
blocks have been added to the list of blocks to index then we will
now write the index for those blocks and flush it, as well.
When we have a very large number of blocks then we may decide to write a
multi-level index, which is why we also keep track of the list of the
index blocks in the same way as we previously kept track of the blocks
to index.
Now when we have finished writing all index blocks we clear the index
and flush the last block to disk. This is done in the wrong order though
because flushing the block to disk will re-add it to the list of blocks
to be indexed. The result is that the next section we are about to write
will have an entry in the list of blocks to index that points to the
last block of the preceding section's index, which will corrupt the log
index.
Fix this corruption by clearing the index after having written the last
block.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 5c086453ff (reftable/stack: perform auto-compaction with
transactional interface, 2023-12-11), we fixed a bug where the
transactional interface to add changes to a reftable stack did not
perform auto-compaction by calling `reftable_stack_auto_compact()` in
`reftable_stack_addition_commit()`. While correct, this change may now
cause us to perform auto-compaction twice in the non-transactional
interface `reftable_stack_add()`:
- It performs auto-compaction by itself.
- It now transitively performs auto-compaction via the transactional
interface.
Remove the first instance so that we only end up doing auto-compaction
once.
Reported-by: Han-Wen Nienhuys <hanwenn@gmail.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In order to compact multiple stacks we iterate through the merged ref
and log records. When there is any error either when reading the records
from the old merged table or when writing the records to the new table
then we break out of the respective loops. When breaking out of the loop
for the ref records though the error code will be overwritten, which may
cause us to inadvertently skip over bad ref records. In the worst case,
this can lead to a compacted stack that is missing records.
Fix the code by using `goto done` instead so that any potential error
codes are properly returned to the caller.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
trace2 streams used to record the URLs that potentially embed
authentication material, which has been corrected.
* jh/trace2-redact-auth:
t0212: test URL redacting in EVENT format
t0211: test URL redacting in PERF format
trace2: redact passwords from https:// URLs by default
trace2: fix signature of trace2_def_param() macro
"git merge-file" learned to take the "--diff-algorithm" option to
use algorithm different from the default "myers" diff.
* ad/merge-file-diff-algo:
merge-file: add --diff-algorithm option
Clean-up code that handles combinations of incompatible options.
* rs/i18n-cannot-be-used-together:
i18n: factorize even more 'incompatible options' messages
Stale URLs have been updated to their current counterparts (or
archive.org) and HTTP links are replaced with working HTTPS links.
* js/update-urls-in-doc-and-comment:
doc: refer to internet archive
doc: update links for andre-simon.de
doc: switch links to https
doc: update links to current pages
Earlier we stopped relying on commit-graph that (still) records
information about commits that are lost from the object store,
which has negative performance implications. The default has been
flipped to disable this pessimization.
* ps/commit-graph-less-paranoid:
commit-graph: disable GIT_COMMIT_GRAPH_PARANOIA by default
Introduce "git replay", a tool meant on the server side without
working tree to recreate a history.
* cc/git-replay:
replay: stop assuming replayed branches do not diverge
replay: add --contained to rebase contained branches
replay: add --advance or 'cherry-pick' mode
replay: use standard revision ranges
replay: make it a minimal server side command
replay: remove HEAD related sanity check
replay: remove progress and info output
replay: add an important FIXME comment about gpg signing
replay: change rev walking options
replay: introduce pick_regular_commit()
replay: die() instead of failing assert()
replay: start using parse_options API
replay: introduce new builtin
t6429: remove switching aspects of fast-rebase
Simplify API implementation to delete references by eliminating
duplication.
* ps/ref-deletion-updates:
refs: remove `delete_refs` callback from backends
refs: deduplicate code to delete references
refs/files: use transactions to delete references
t5510: ensure that the packed-refs file needs locking
When iterating over entries in the block iterator we compute the key of
each of the entries and write it into a buffer. We do not reuse the
buffer though and thus re-allocate it on every iteration, which is
wasteful.
Refactor the code to reuse the buffer.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are a bunch of locations where we initialize members of `struct
block_iter`, which makes it harder than necessary to expand this struct
to have additional members. Unify the logic via a new `BLOCK_ITER_INIT`
macro that initializes all members.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When iterating over entries in the merged iterator's queue, we compute
the key of each of the entries and write it into a buffer. We do not
reuse the buffer though and thus re-allocate it on every iteration,
which is wasteful given that we never transfer ownership of the
allocated bytes outside of the loop.
Refactor the code to reuse the buffer. This also fixes a potential
memory leak when `merged_iter_advance_subiter()` returns an error.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When writing a new reftable stack, Git will first create the stack with
a random suffix so that concurrent updates will not try to write to the
same file. This random suffix is computed via a call to rand(3P). But we
never seed the function via srand(3P), which means that the suffix is in
fact always the same.
Fix this bug by using `git_rand()` instead, which does not need to be
initialized. While this function is likely going to be slower depending
on the platform, this slowness should not matter in practice as we only
use it when writing a new reftable stack.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When starting a transaction via `reftable_stack_init_addition()`, we
create a lockfile for the reftable stack itself which we'll write the
new list of tables to. But if we terminate abnormally e.g. via a call to
`die()`, then we do not remove the lockfile. Subsequent executions of
Git which try to modify references will thus fail with an out-of-date
error.
Fix this bug by registering the lock as a `struct tempfile`, which
ensures automatic cleanup for us.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In `reftable_stack_reload_once()` we iterate over all the tables added
to the stack in order to figure out whether any of the tables needs to
be reloaded. We use a set of buffers in this context to compute the
paths of these tables, but discard those buffers on every iteration.
This is quite wasteful given that we do not need to transfer ownership
of the allocated buffer outside of the loop.
Refactor the code to instead reuse the buffers to reduce the number of
allocations we need to do. Note that we do not have to manually reset
the buffer because `stack_filename()` does this for us already.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Whenever updating references or reflog entries in the reftable stack, we
need to add a new table to the stack, thus growing the stack's length by
one. The stack can grow to become quite long rather quickly, leading to
performance issues when trying to read records. But besides performance
issues, this can also lead to exhaustion of file descriptors very
rapidly as every single table requires a separate descriptor when
opening the stack.
While git-pack-refs(1) fixes this issue for us by merging the tables, it
runs too irregularly to keep the length of the stack within reasonable
limits. This is why the reftable stack has an auto-compaction mechanism:
`reftable_stack_add()` will call `reftable_stack_auto_compact()` after
its added the new table, which will auto-compact the stack as required.
But while this logic works alright for `reftable_stack_add()`, we do not
do the same in `reftable_addition_commit()`, which is the transactional
equivalent to the former function that allows us to write multiple
updates to the stack atomically. Consequentially, we will easily run
into file descriptor exhaustion in code paths that use many separate
transactions like e.g. non-atomic fetches.
Fix this issue by calling `reftable_stack_auto_compact()`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
While we have several tests that check whether we correctly perform
auto-compaction when manually calling `reftable_stack_auto_compact()`,
we don't have any tests that verify whether `reftable_stack_add()` does
call it automatically. Add one.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are calls to write(3P) where we don't properly handle interrupts.
Convert them to use `write_in_full()`.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are calls to pread(3P) and read(3P) where we don't properly handle
interrupts. Convert them to use `pread_in_full()` and `read_in_full()`,
respectively.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The `EXPECT` macros used by the reftable test framework are all using a
single `if` statement with the actual condition. This results in weird
syntax when using them in if/else statements like the following:
```
if (foo)
EXPECT(foo == 2)
else
EXPECT(bar == 2)
```
Note that there need not be a trailing semicolon. Furthermore, it is not
immediately obvious whether the else now belongs to the `if (foo)` or
whether it belongs to the expanded `if (foo == 2)` from the macro.
Fix this by wrapping the macros in a do/while loop.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Newer versions of Getopt::Long started giving warnings against our
(ab)use of it in "git send-email". Bump the minimum version
requirement for Perl to 5.8.1 (from September 2002) to allow
simplifying our implementation.
* tz/send-email-negatable-options:
send-email: avoid duplicate specification warnings
perl: bump the required Perl version to 5.8.1 from 5.8.0
"git rebase --autosquash" is now enabled for non-interactive rebase,
but it is still incompatible with the apply backend.
* ak/rebase-autosquash:
rebase: rewrite --(no-)autosquash documentation
rebase: support --autosquash without -i
rebase: fully ignore rebase.autoSquash without -i
"git for-each-ref --no-sort" still sorted the refs alphabetically
which paid non-trivial cost. It has been redefined to show output
in an unspecified order, to allow certain optimizations to take
advantage of.
* vd/for-each-ref-unsorted-optimization:
t/perf: add perf tests for for-each-ref
ref-filter.c: use peeled tag for '*' format fields
for-each-ref: clean up documentation of --format
ref-filter.c: filter & format refs in the same callback
ref-filter.c: refactor to create common helper functions
ref-filter.c: rename 'ref_filter_handler()' to 'filter_one()'
ref-filter.h: add functions for filter/format & format-only
ref-filter.h: move contains caches into filter
ref-filter.h: add max_count and omit_empty to ref_format
ref-filter.c: really don't sort when using --no-sort
Test and shell scripts clean-up.
* ps/ban-a-or-o-operator-with-test:
Makefile: stop using `test -o` when unlinking duplicate executables
contrib/subtree: convert subtree type check to use case statement
contrib/subtree: stop using `-o` to test for number of args
global: convert trivial usages of `test <expr> -a/-o <expr>`
"git format-patch --encode-email-headers" ignored the option when
preparing the cover letter, which has been corrected.
* ss/format-patch-use-encode-headers-for-cover-letter:
format-patch: fix ignored encode_email_headers for cover letter
Update ref-related tests.
* ps/ref-tests-update:
t: mark several tests that assume the files backend with REFFILES
t7900: assert the absence of refs via git-for-each-ref(1)
t7300: assert exact states of repo
t4207: delete replace references via git-update-ref(1)
t1450: convert tests to remove worktrees via git-worktree(1)
t: convert tests to not access reflog via the filesystem
t: convert tests to not access symrefs via the filesystem
t: convert tests to not write references via the filesystem
t: allow skipping expected object ID in `ref-store update-ref`
"git add" and "git stash" learned to support the ":(attr:...)"
magic pathspec.
* jw/git-add-attr-pathspec:
attr: enable attr pathspec magic for git-add and git-stash
Code clean-up for jk/chunk-bounds topic.
* jk/chunk-bounds-more:
commit-graph: mark chunk error messages for translation
commit-graph: drop verify_commit_graph_lite()
commit-graph: check order while reading fanout chunk
commit-graph: use fanout value for graph size
commit-graph: abort as soon as we see a bogus chunk
commit-graph: clarify missing-chunk error messages
commit-graph: drop redundant call to "lite" verification
midx: check consistency of fanout table
commit-graph: handle overflow in chunk_size checks
The way CI testing used "prove" could lead to running the test
suite twice needlessly, which has been corrected.
* js/ci-discard-prove-state:
ci: avoid running the test suite _twice_
Add support for GitLab CI.
* ps/ci-gitlab:
ci: add support for GitLab CI
ci: install test dependencies for linux-musl
ci: squelch warnings when testing with unusable Git repo
ci: unify setup of some environment variables
ci: split out logic to set up failed test artifacts
ci: group installation of Docker dependencies
ci: make grouping setup more generic
ci: reorder definitions for grouping functions
Update the base topic to work with CMake builds.
* js/doc-unit-tests-with-cmake:
cmake: handle also unit tests
cmake: use test names instead of full paths
cmake: fix typo in variable name
artifacts-tar: when including `.dll` files, don't forget the unit-tests
unit-tests: do show relative file paths
unit-tests: do not mistake `.pdb` files for being executable
cmake: also build unit tests
Process to add some form of low-level unit tests has started.
* js/doc-unit-tests:
ci: run unit tests in CI
unit tests: add TAP unit test framework
unit tests: add a project plan document
Continue the work of 12909b6b8a (i18n: turn "options are incompatible"
into "cannot be used together", 2022-01-05) and a699367bb8 (i18n:
factorize more 'incompatible options' messages, 2022-01-31) to use the
same parameterized error message for reporting incompatible command line
options. This reduces the number of strings to translate and makes the
UI slightly more consistent.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Releasing strbuf and string_list just before exiting is not strictly
necessary, but it gets rid of false positives reported by leak checkers,
which can then be more easily used to show that the column-printing
machinery behind print_columns() are free of leaks.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The replay command is able to replay multiple branches but when some of
them are based on other replayed branches, their commit should be
replayed onto already replayed commits.
For this purpose, let's store the replayed commit and its original
commit in a key value store, so that we can easily find and reuse a
replayed commit instead of the original one.
Co-authored-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Let's add a `--contained` option that can be used along with
`--onto` to rebase all the branches contained in the <revision-range>
argument.
Co-authored-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
