Regression fix for 2.36 where "git name-rev" started to sometimes
reference strings after they are freed.
* rs/name-rev-fix-free-after-use:
Revert "name-rev: release unused name strings"
"diff-tree --stdin" has been broken for about a year, but 2.36
release broke it even worse by breaking running the command with
<pathspec>, which in turn broke "gitk" and got noticed. This has
been corrected by aligning its behaviour to that of "log".
* jc/diff-tree-stdin-fix:
2.36 gitk/diff-tree --stdin regression fix
"git submodule update" without pathspec should silently skip an
uninitialized submodule, but it started to become noisy by mistake.
* gc/submodule-update-part2:
submodule--helper: fix initialization of warn_if_uninitialized
The description of 'safe.directory' mentions that it's respected in
the system and global configs, and ignored in the repository config
and on the command line, but it doesn't mention whether it's respected
or ignored when specified via environment variables (nor does the
commit message adding 'safe.directory' [1]).
Clarify that 'safe.directory' is ignored when specified in the
environment, and add tests to make sure that it remains so.
[1] 8959555cee (setup_git_directory(): add an owner check for the
top-level directory, 2022-03-02)
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
According to the documentation 'safe.directory' "is only respected
when specified in a system or global config, not when it is specified
in a repository config or via the command line option -c
safe.directory=<path>".
Add tests to check that 'safe.directory' in the repository config or
on the command line is indeed ignored.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 8959555cee (setup_git_directory(): add an owner check for the
top-level directory, 2022-03-02) when git finds itself in a repository
owned by someone else, it aborts with a "fatal: unsafe repository
(<repo path>)" error message and an advice about how to set the
'safe.directory' config variable to mark that repository as safe.
't0033-safe-directory.sh' contains tests that check that this feature
and handling said config work as intended. To ensure that git dies
for the right reason, several of those tests check that its standard
error contains the name of that config variable, but:
- it only appears in the advice part, not in the actual error
message.
- it is interpreted as a regexp by 'grep', so, because of the dot,
it matches the name of the test script and the path of the trash
directory as well. Consequently, these tests could be fooled by
any error message that would happen to include the path of the
test repository.
Tighten these checks to look for "unsafe repository" instead.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It is not obvious that the 'git rev-parse' builtin would use the sparse
index, but it is possible to parse paths out of the index using the
":<path>" syntax. The 'git rev-parse' output is only the OID of the
object found at that location, but otherwise behaves similarly to 'git
show :<path>'. This includes the failure conditions on directories and
the error messages depending on whether a path is in the worktree or
not.
The only code change required is to change the
command_requires_full_index setting in builtin/rev-parse.c, and we can
re-use many existing 'git show' tests for the rev-parse case.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When running 'git show :<path>' where '<path>' is a directory, then
there is a subtle difference between a full checkout and a sparse
checkout. The error message from diagnose_invalid_index_path() reports
whether the path is on disk or not. The full checkout will have the
directory on disk, but the path will not be in the index. The sparse
checkout could have the directory not exist, specifically when that
directory is outside of the sparse-checkout cone.
In the case of a sparse index, we have yet another state: the path can
be a sparse directory in the index. In this case, the error message from
diagnose_invalid_index_path() would erroneously say "path '<path>' is in
the index, but not at stage 0", which is false.
Add special casing around sparse directory entries so we get to the
correct error message. This requires two checks in order to get parity
with the normal sparse-checkout case.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_oid_with_context_1() method is used when parsing revision
arguments. One particular case is to take a ":<path>" string and search
the index for the given path.
In the case of a sparse index, this might find a sparse directory entry,
in which case the contained object is a tree. In the case of a full
index, this search within the index would fail.
In order to maintain identical return state as in a full index, inspect
the discovered cache entry to see if it is a sparse directory and reject
it. This requires being careful around the only_to_die option to be sure
we die only at the correct time.
This changes the behavior of 'git show :<sparse-dir>', but does not
bring it entirely into alignment with a full index case. It specifically
hits the wrong error message within diagnose_invalid_index_path(). That
error message will be corrected in a future change.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The 'git show' command can take an input to request the state of an
object in the index. This can lead to parsing the index in order to load
a specific file entry. Without the change presented here, a sparse index
would expand to a full one, taking much longer than usual to access a
simple file.
There is one behavioral change that happens here, though: we now can
find a sparse directory entry within the index! Commands that previously
failed because we could not find an entry in the worktree or index now
succeed because we _do_ find an entry in the index.
There might be more work to do to make other situations succeed when
looking for an indexed tree, perhaps by looking at or updating the
cache-tree extension as needed. These situations include having a full
index or asking for a directory that is within the sparse-checkout cone
(and hence is not a sparse directory entry in the index).
For now, we demonstrate how the sparse index integration is extremely
simple for files outside of the cone as well as directories within the
cone. A later change will resolve this behavior around sparse
directories.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The .warn_if_uninitialized member was introduced by 48308681
(git submodule update: have a dedicated helper for cloning,
2016-02-29) to submodule_update_clone struct and initialized to
false. When c9911c93 (submodule--helper: teach update_data more
options, 2022-03-15) moved it to update_data struct, it started
to initialize it to true but this change was not explained in
its log message.
The member is set to true only when pathspec was given, and is
used when a submodule that matched the pathspec is found
uninitialized to give diagnostic message. "submodule update"
without pathspec is supposed to iterate over all submodules
(i.e. without pathspec limitation) and update only the
initialized submodules, and finding uninitialized submodules
during the iteration is a totally expected and normal thing that
should not be warned.
[jc: added tests]
Signed-off-by: Orgad Shaneh <orgads@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This only surfaced as a regression after 2.36 release, but the
breakage was already there with us for at least a year.
The diff_free() call is to be used after we completely finished with
a diffopt structure. After "git diff A B" finishes producing
output, calling it before process exit is fine. But there are
commands that prepares diff_options struct once, compares two sets
of paths, releases resources that were used to do the comparison,
then reuses the same diff_option struct to go on to compare the next
two sets of paths, like "git log -p".
After "git log -p" finishes showing a single commit, calling it
before it goes on to the next commit is NOT fine. There is a
mechanism, the .no_free member in diff_options struct, to help "git
log" to avoid calling diff_free() after showing each commit and
instead call it just one. When the mechanism was introduced in
e900d494 (diff: add an API for deferred freeing, 2021-02-11),
however, we forgot to do the same to "diff-tree --stdin", which *is*
a moral equivalent to "git log".
During 2.36 release cycle, we started clearing the pathspec in
diff_free(), so programs like gitk that runs
git diff-tree --stdin -- <pathspec>
downstream of a pipe, processing one commit after another, started
showing irrelevant comparison outside the given <pathspec> from the
second commit. The same commit, by forgetting to teach the .no_free
mechanism, broke "diff-tree --stdin -I<regexp>" and nobody noticed
it for over a year, presumably because it is so seldom used an
option.
But <pathspec> is a different story. The breakage was very
prominently visible and was reported immediately after 2.36 was
released.
Fix this breakage by mimicking how "git log" utilizes the .no_free
member so that "diff-tree --stdin" behaves more similarly to "log".
Protect the fix with a few new tests.
Reported-by: Matthias Aßhauer <mha1993@live.de>
Helped-by: René Scharfe <l.s.r@web.de>
Helped-by: Phillip Wood <phillip.wood123@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The get_object_directory() method returns the exact string stored at
the_repository->objects->odb->path. The return type of "char *" implies
that the caller must keep track of the buffer and free() it when
complete. This causes significant problems later when the ODB is
accessed.
Use "const char *" as the return type to avoid this confusion. There are
no current callers that care about the non-const definition.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The --object-dir argument to 'git multi-pack-index' allows a user to
specify an alternate to use instead of the local $GITDIR. This is used
by third-party tools like VFS for Git to maintain the pack-files in a
"shared object cache" used by multiple clones.
On Windows, the user can specify a path using a Windows-style file path
with backslashes such as "C:\Path\To\ObjectDir". This same path style is
used in the .git/objects/info/alternates file, so it already matches the
path of that alternate. However, find_odb() converts these paths to
real-paths for the comparison, which use forward slashes. As of the
previous change, lookup_multi_pack_index() uses real-paths, so it
correctly finds the target multi-pack-index when given these paths.
Some commands such as 'git multi-pack-index repack' call child processes
using the object_dir value, so it can be helpful to convert the path to
the real-path before sending it to those locations.
Add a callback to convert the real path immediately upon parsing the
argument. We need to be careful that we don't store the exact value out
of get_object_directory() and free it, or we could corrupt a later use
of the_repository->objects->odb->path.
We don't use get_object_directory() for the initial instantiation in
cmd_multi_pack_index() because we need 'git multi-pack-index -h' to work
without a Git repository.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This helper looks for a parsed multi-pack-index whose object directory
matches the given object_dir. Before going into the loop over the parsed
multi-pack-indexes, it calls find_odb() to ensure that the given
object_dir is actually a known object directory.
However, find_odb() uses real-path manipulations to compare the input to
the alternate directories. This same real-path comparison is not used in
the loop, leading to potential issues with the strcmp().
Update the method to use the real-path values instead.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When cloning directly from a local repository, we load a list of refs
based on scanning the $GIT_DIR/refs/ directory of the "server"
repository. If files exist in that directory that do not parse as
hexadecimal hashes, then the ref array used by write_remote_refs()
ends up with some entries with null OIDs. This causes us to hit a BUG()
statement in ref_transaction_create():
BUG: create called without valid new_oid
This BUG() call used to be a die() until 033abf97f (Replace all
die("BUG: ...") calls by BUG() ones, 2018-05-02). Before that, the die()
was added by f04c5b552 (ref_transaction_create(): check that new_sha1 is
valid, 2015-02-17).
The original report for this bug [1] mentioned that this problem did not
exist in Git 2.27.0. The failure bisects unsurprisingly to 968f12fda
(refs: turn on GIT_REF_PARANOIA by default, 2021-09-24). When
GIT_REF_PARANOIA is enabled, this case always fails as far back as I am
able to successfully compile and test the Git codebase.
[1] https://github.com/git-for-windows/git/issues/3781
There are two approaches to consider here. One would be to remove this
BUG() statement in favor of returning with an error. There are only two
callers to ref_transaction_create(), so this would have a limited
impact.
The other approach would be to add special casing in 'git clone' to
avoid this faulty input to the method.
While I originally started with changing 'git clone', I decided that
modifying ref_transaction_create() was a more complete solution. This
prevents failing with a BUG() statement when we already have a good way
to report an error (including a reason for that error) within the
method. Both callers properly check the return value and die() with the
error message, so this is an appropriate direction.
The added test helps check against a regression, but does check that our
intended error message is handled correctly.
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This reverts commit 2d53975488.
3656f84278 (name-rev: prefer shorter names over following merges,
2021-12-04) broke the assumption of 2d53975488 (name-rev: release unused
name strings, 2020-02-04) that a better name for a child is a better
name for all of its ancestors as well, because it added a penalty for
generation > 0. This leads to strings being free(3)'d that are still
needed.
079f970971 (name-rev: sort tip names before applying, 2020-02-05)
already reduced the number of free(3) calls for the use case that
motivated the original patch (name-rev --all in the Chromium repository)
from ca. 44000 to 5, and 3656f84278 eliminated even those few. So this
revert won't affect name-rev's performance on that particular repo.
Reported-by: Thomas Hurst <tom@hur.st>
Helped-by: Elijah Newren <newren@gmail.com>
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "--since=<time>" option of "git log" limits the commits displayed by
the command by stopping the traversal once it sees a commit whose
timestamp is older than the given time and not digging further into its
parents.
This is OK in a history where a commit always has a newer timestamp than
any of its parents'. Once you see a commit older than the given <time>,
all ancestor commits of it are even older than the time anyway. It
poses, however, a problem when there is a commit with a wrong timestamp
that makes it appear older than its parents. Stopping traversal at the
"incorrectly old" commit will hide its ancestors that are newer than
that wrong commit and are newer than the cut-off time given with the
--since option. --max-age and --after being the synonyms to --since,
they share the same issue.
Add a new "--since-as-filter" option that is a variant of
"--since=<time>". Instead of stopping the traversal to hide an old
enough commit and its all ancestors, exclude commits with an old
timestamp from the output but still keep digging the history.
Without other traversal stopping options, this will force the command in
"git log" family to dig down the history to the root. It may be an
acceptable cost for a small project with short history and many commits
with screwy timestamps.
It is quite unlikely for us to add traversal stopper other than since,
so have this as a --since-as-filter option, rather than a separate
--as-filter, that would be probably more confusing.
Signed-off-by: Miklos Vajna <vmiklos@vmiklos.hu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix a regression in 707d2f2fe8 (CI: use "$runs_on_pool", not
"$jobname" to select packages & config, 2021-11-23).
In that commit I changed CC=gcc from CC=gcc-9, but on OSX the "gcc" in
$PATH points to clang, we need to use gcc-9 instead. Likewise for the
linux-gcc job CC=gcc-8 was changed to the implicit CC=gcc, which would
select GCC 9.4.0 instead of GCC 8.4.0.
Furthermore in 25715419bf (CI: don't run "make test" twice in one
job, 2021-11-23) when the "linux-TEST-vars" job was split off from
"linux-gcc" the "cc_package: gcc-8" line was copied along with
it, so its "cc_package" line wasn't working as intended either.
As a table, this is what's changed by this commit, i.e. it only
affects the linux-gcc, linux-TEST-vars and osx-gcc jobs:
|-------------------+-----------+-------------------+-------+-------|
| jobname | vector.cc | vector.cc_package | old | new |
|-------------------+-----------+-------------------+-------+-------|
| linux-clang | clang | - | clang | clang |
| linux-sha256 | clang | - | clang | clang |
| linux-gcc | gcc | gcc-8 | gcc | gcc-8 |
| osx-clang | clang | - | clang | clang |
| osx-gcc | gcc | gcc-9 | clang | gcc-9 |
| linux-gcc-default | gcc | - | gcc | gcc |
| linux-TEST-vars | gcc | gcc-8 | gcc | gcc-8 |
|-------------------+-----------+-------------------+-------+-------|
Reported-by: Carlo Arenas <carenas@gmail.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
--keep-base rebases onto the merge base of the given upstream and the
current HEAD regardless of whether a branch is given. This is contrary
to the documentation and to the option's intended purpose. Instead,
rebase onto the merge base of the given upstream and the given branch.
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Git uses temporary files to pass the contents of blobs to external diff
programs and textconv filters. It calls mks_tempfile_ts() to create
them, which puts them all in the same directory. This requires adding
a random name prefix.
Use mks_tempfile_dt() instead, which allows the files to have arbitrary
names, each in their own separate temporary directory. This way they
can have the same basename as the original blob, which looks nicer in
graphical diff programs.
The test in t4020 to check the prettiness of the temporary paths was
neutered by 5476bdf0e8 (diff tests: don't ignore "git diff" exit code in
"read" loop, 2022-03-07), which removed its grep check without replacing
it with an equivalent test_cmp check. Add one that only checks the
basename of the temporary file and nothing else.
And make the test more robust while at it, by using test_when_finished
to get rid of the added file even if the test fails.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a function to create a temporary file with a certain name in a
temporary directory created using mkdtemp(3). Its result is more
sightly than the paths created by mks_tempfile_ts(), which include
a random prefix. That's useful for files passed to a program that
displays their name, e.g. an external diff tool.
Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There are two reasons that we could return NULL early within
load_commit_graph_chain():
1. The file does not exist, so the file pointer is NULL.
2. The file exists, but is too small to contain a single hash.
These were grouped together when the function was first written in
5c84b3396 (commit-graph: load commit-graph chains, 2019-06-18) in order
to simplify how the 'chain_name' string is freed. However, the current
code leaves a narrow window where the file pointer is not closed when
the file exists, but is rejected for being too small.
Split out these cases separately to ensure we close the file in this
case.
Signed-off-by: Kleber Tarcísio <klebertarcisio@yahoo.com.br>
Signed-off-by: Derrick Stolee <derrickstolee@github.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There is an if statement where both if and else have the same
assignment of options.type to REBASE_MERGE. Simplify
it by getting that assigmnent out of the if.
Signed-off-by: Edmundo Carmona Antoranz <eantoranz@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A couple of work around for CI breaking warnings from gcc 12.
* cb/buggy-gcc-12-workaround:
config.mak.dev: alternative workaround to gcc 12 warning in http.c
config.mak.dev: workaround gcc 12 bug affecting "pedantic" CI job
This provides a "no code change needed" option to the "fix" currently
queued as part of ab/http-gcc-12-workaround and therefore should be
reverted once that gets merged.
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Originally noticed by Peff[1], but yet to be corrected[2] and planned to
be released with Fedora 36 (scheduled for Apr 19).
dir.c: In function ‘git_url_basename’:
dir.c:3085:13: error: ‘memchr’ specified bound [9223372036854775808, 0] exceeds maximum object size 9223372036854775807 [-Werror=stringop-overread]
3085 | if (memchr(start, '/', end - start) == NULL
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Fedora is used as part of the CI, and therefore that release will trigger
failures, unless the version of the image used is locked to an older
release, as an alternative.
Restricting the flag to the affected source file, as well as implementing
an independent facility to track these workarounds was specifically punted
to minimize the risk of introducing problems so close to a release.
This change should be reverted once the underlying gcc bug is solved and
which should be visible by NOT triggering a warning, otherwise.
[1] https://lore.kernel.org/git/YZQhLh2BU5Hquhpo@coredump.intra.peff.net/
[2] https://bugzilla.redhat.com/show_bug.cgi?id=2075786
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
1214aa841b (reftable: add blocksource, an abstraction for random
access reads, 2021-10-07), makes the assumption that it is ok to
free a reftable_block pointing to NULL if the size is also set to
0, but implements that using a memset call that at least in glibc
based system will trigger a runtime exception if called with a
NULL pointer as its first parameter.
Avoid doing so by adding a conditional to check for the size in all
three identically looking functions that were affected, and therefore,
still allow memset to help catch callers that might incorrectly pass
a NULL pointer with a non zero size, but avoiding the exception for
the valid cases.
Signed-off-by: Carlo Marcelo Arenas Belón <carenas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Revert the "deletion of a ref should not trigger transaction events
for loose and packed ref backends separately" that regresses the
behaviour when a ref is not modified since it was packed.
* jc/revert-ref-transaction-hook-changes:
RelNotes: revert the description on the reverted topics
Revert "fetch: increase test coverage of fetches"
Revert "Merge branch 'ps/avoid-unnecessary-hook-invocation-with-packed-refs'"
* update the following words translations: commit, untracked, stage,
cache, stash, work..., index, reset, label, check..., tags, graft,
alternate object, amend, ancestor, cherry-pick, bisect, blame, chain,
cache, bug, chunk, branch, bundle, clean, clone, commit-graph, commit
object, commit-ish, committer, cover letter, conflict, dangling,
detach, dir, dumb, fast-forward, file system, fixup, fork, fetch, Git
archive, gitdir, graft, replace ref
* correct some mispellings
* git-po-helper update
* remove some obsolete lines
* unfuzzy entries
* random translation updates
* update contact in pt_PT.po
* add the following words to the translation table: override, recurse,
print, offset, unbundle, mirror repository, multi-pack, bad,
whitespace, batch
* remove the following words of the translation table: core Git
* change the following words on the translation table: dry-run, apply,
patch, replay, blame, chain, gitdir, file system, fork, unset, handle
* some translation to the first person
* update copyright text
* word 'utilização:' to 'uso:'
* word 'pai' to 'parente'
Signed-off-by: Daniel Santos <dacs.git@brilhante.top>
We do not have to guess how common the mistake the change targets is
when describing it. Such an argument may be good while proposing a
change, but does not quite belong in the record of what has already
happened, i.e. a release note.
Helped-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This reverts commit 2a0cafd464,
as it expects a working "a ref deletion must produce a single
transaction, not one for loose and another for packed" topic,
which we do not have.
Junio C Hamano