In 8a2cd3f512 (stash: remove the stash.useBuiltin setting, 2020-03-03),
we removed `git-legacy-stash.sh`. But `git-sh-setup.sh` somehow still
thinks about it. Let's just not.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Extend the "describe your changes well" section to cover whom we are
trying to help by doing so in the first place.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We give a guidance for proposed log message to write problem
statement first, followed by the reasoning behind, and recipe for,
the solution. Clarify that we describe the situation _before_ the
proposed patch is applied in the present tense (not in the past
tense e.g. "we used to do X, but thanks to this commit we now do Y")
for consistency.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When opening a MIDX/pack-bitmap, we call open_midx_bitmap_1() or
open_pack_bitmap_1() respectively in a loop over the set of MIDXs/packs.
By design, these functions are supposed to be called over every pack and
MIDX, since only one of them should have a valid bitmap.
Ordinarily we return '0' from these two functions in order to indicate
that we successfully loaded a bitmap To signal that we couldn't load a
bitmap corresponding to the MIDX/pack (either because one doesn't exist,
or because there was an error with loading it), we can return '-1'. In
either case, the callers each enumerate all MIDXs/packs to ensure that
at most one bitmap per-kind is present.
But when we fail to load a bitmap that does exist (for example, loading
a MIDX bitmap without finding a corresponding reverse index), we'll
return -1 but leave the 'midx' field non-NULL. So when we fallback to
loading a pack bitmap, we'll complain that the bitmap we're trying to
populate already is "opened", even though it isn't.
Rectify this by setting the '->pack' and '->midx' field back to NULL as
appropriate. Two tests are added: one to ensure that the MIDX-to-pack
bitmap fallback works, and another to ensure we still complain when
there are multiple pack bitmaps in a repository.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a MIDX contains the new `RIDX` chunk, ensure that the reverse index
is read from it instead of the on-disk .rev file. Since we need to
encode the object order in the MIDX itself for correctness reasons,
there is no point in storing the same data again outside of the MIDX.
So, this patch stops writing separate .rev files, and reads it out of
the MIDX itself. This is possible to do with relatively little new code,
since the format of the RIDX chunk is identical to the data in the .rev
file. In other words, we can implement this by pointing the
`revindex_data` field at the reverse index chunk of the MIDX instead of
the .rev file without any other changes.
Note that we have two knobs that are adjusted for the new tests:
GIT_TEST_MIDX_WRITE_REV and GIT_TEST_MIDX_READ_RIDX. The former controls
whether the MIDX .rev is written at all, and the latter controls whether
we read the MIDX's RIDX chunk.
Both are necessary to ensure that the test added at the beginning of
this series continues to work. This is because we always need to write
the RIDX chunk in the MIDX in order to change its checksum, but we want
to make sure reading the existing .rev file still works (since the RIDX
chunk takes precedence by default).
Arguably this isn't a very interesting mode to test, because the
precedence rules mean that we'll always read the RIDX chunk over the
.rev file. But it makes it impossible for a user to induce corruption in
their repository by adjusting the test knobs (since if we had an
either/or knob they could stop writing the RIDX chunk, allowing them to
tweak the MIDX's object order without changing its checksum).
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To prepare for reading the reverse index data out of the MIDX itself,
teach the `test_rev_exists` function to take an expected "source" for
the reverse index data.
When given "rev", it asserts that the MIDX's `.rev` file exists, and is
loaded when verifying the integrity of its bitmaps. Otherwise, it
ensures that trace2 reports the source of the reverse index data as the
same string which was given to test_rev_exists().
The following patch will implement reading the reverse index data from
the MIDX itself.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In t5326, we have a handful of tests that we would like to run twice:
once using the MIDX's new `RIDX` chunk as the source of the
reverse-index cache, and once using the separate `.rev` file.
But because these tests mutate the state of the underlying repository,
and then make assumptions about those mutations occurring in a certain
sequence, simply running the tests twice in the same repository is
awkward.
Instead, extract the core of interesting tests into t/lib-bitmap.sh to
prepare for them to be run twice, each in a separate test script. This
means that they can each operate on a separate repository, removing any
concerns about mutating state.
For now, this patch is a strict cut-and-paste of some tests from t5326.
The tests which did not move are not interesting with respect to the
source of their reverse index data.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To determine which source of data is used for the MIDX's reverse index
cache, introduce a helper which forces loading the reverse index, and
then looks for the special trace2 event introduced in a previous commit.
For now, this helper just looks for when the legacy MIDX .rev file was
loaded, but in a subsequent commit will become parameterized over the
the reverse index's source.
This function replaces checking for the existence of the .rev file. We
could write a similar helper to ensure that the .rev file is cleaned up
after repacking, but it will make subsequent tests more difficult to
write, and provides marginal value since we already check that the MIDX
.bitmap file is removed.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The core.multiPackIndex config became true by default back in 18e449f86b
(midx: enable core.multiPackIndex by default, 2020-09-25), so it is no
longer necessary to enable it explicitly.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In a subsequent commit, we'll use the MIDX's new 'RIDX' chunk as a
source for the reverse index's data. But it will be useful for tests to
be able to determine whether the reverse index was loaded from the
separate .rev file, or from a chunk within the MIDX.
To instrument this, add a trace2 event which the tests can look for in
order to determine the reverse index's source.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous patch demonstrates a bug where a MIDX's auxiliary object
order can become out of sync with a MIDX bitmap.
This is because of two confounding factors:
- First, the object order is stored in a file which is named according
to the multi-pack index's checksum, and the MIDX does not store the
object order. This means that the object order can change without
altering the checksum.
- But the .rev file is moved into place with finalize_object_file(),
which link(2)'s the file into place instead of renaming it. For us,
that means that a modified .rev file will not be moved into place if
MIDX's checksum was unchanged.
This fix is to force the MIDX's checksum to change when the preferred
pack changes but the set of packs contained in the MIDX does not. In
other words, when the object order changes, the MIDX's checksum needs to
change with it (regardless of whether the MIDX is tracking the same or
different packs).
This prevents a race whereby changing the object order (but not the
packs themselves) enables a reader to see the new .rev file with the old
MIDX, or similarly seeing the new bitmap with the old object order.
But why can't we just stop hardlinking the .rev into place instead
adding additional data to the MIDX? Suppose that's what we did. Then
when we go to generate the new bitmap, we'll load the old MIDX bitmap,
along with the MIDX that it references. That's fine, since the new MIDX
isn't moved into place until after the new bitmap is generated. But the
new object order *has* been moved into place. So we'll read the old
bitmaps in the new order when generating the new bitmap file, meaning
that without this secondary change, bitmap generation itself would
become a victim of the race described here.
This can all be prevented by forcing the MIDX's checksum to change when
the object order does. By embedding the entire object order into the
MIDX, we do just that. That is, the MIDX's checksum will change in
response to any perturbation of the underlying object order. In t5326,
this will cause the MIDX's checksum to update (even without changing the
set of packs in the MIDX), preventing the stale read problem.
Note that this makes it safe to continue to link(2) the MIDX .rev file
into place, since it is now impossible to have a .rev file that is
out-of-sync with the MIDX whose checksum it references. (But we will do
away with MIDX .rev files later in this series anyway, so this is
somewhat of a moot point).
In theory, it is possible to store a "fingerprint" of the full object
order here, so long as that fingerprint changes at least as often as the
full object order does. Some possibilities here include storing the
identity of the preferred pack, along with the mtimes of the
non-preferred packs in a consistent order. But storing a limited part of
the information makes it difficult to reason about whether or not there
are gaps between the two that would cause us to get bitten by this bug
again.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This patch demonstrates a cause of bitmap corruption that can occur when
the contents of the multi-pack index does not change, but the underlying
object order does.
In this example, we have a MIDX containing two packs, each with a
distinct set of objects (pack A corresponds to the tree, blob, and
commit from the first patch, and pack B corresponds to the second
patch).
First, a MIDX is written where the 'A' pack is preferred. As expected,
the bitmaps generated there are in-tact. But then, we generate an
identical MIDX with a different object order: this time preferring pack
'B'.
Due to a bug which will be explained and fixed in the following commit,
the MIDX is updated, but the .rev file is not, causing the .bitmap file
to be read incorrectly. Specifically, the .bitmap file will contain
correct data, but the auxiliary object order in the .rev file is stale,
causing readers to get confused by reading the new bitmaps using the old
object order.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Reviewed-by: Derrick Stolee <dstolee@microsoft.com>
Reviewed-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix a regression in 2.35 that roke the use of "rebase" and "stash"
in a secondary worktree.
* en/keep-cwd:
sequencer, stash: fix running from worktree subdir
The tree is not open for new development yet, but let's mark the
beginning of the new cycle before we start merging down regression
fix topics.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add an apostrophe to "signatures" to indicate the possessive
relationship in "the signature's creation".
Signed-off-by: Greg Hurrell <greg@hurrell.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Specifically, replace the tab between "the" and "first" with a space.
Signed-off-by: Greg Hurrell <greg@hurrell.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Remove the now-unused "failure_errno" parameter from the
refs_resolve_ref_unsafe() signature. In my recent 96f6623ada (Merge
branch 'ab/refs-errno-cleanup', 2021-11-29) series we made all of its
callers explicitly request the errno via an output parameter.
As that series shows all but one caller ended up passing in a
boilerplate "ignore_errno", since they only cared about whether the
return value was NULL or not, i.e. if the ref could be resolved.
There was one small issue with that series fixed with a follow-up in
31e3912369 (Merge branch 'ab/refs-errno-cleanup', 2022-01-14) a small
bug in that series was fixed.
After those two there was one caller left in sequencer.c that used the
"failure_errno', but as of the preceding commit it uses a boilerplate
"ignore_errno" instead.
This leaves the public refs API without any use of "failure_errno" at
all. We could still do with a bit of cleanup and generalization
between refs.c and refs/files-backend.c before the "reftable"
integration lands, but that's all internal to the reference code
itself.
So let's remove this output parameter. Not only isn't it used now, but
it's unlikely that we'll want it again in the future. We'd like to
slowly move the refs API to a more file-backend independent way of
communicating error codes, having it use a "failure_errno" was only
the first step in that direction. If this or any other function needs
to communicate what specifically is wrong with the requested "refname"
it'll be better to have the function set some output enum of
well-defined error states than piggy-backend on "errno".
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change code that was faithfully migrated to the new "resolve_errno"
API in ed90f04155 (refs API: make resolve_ref_unsafe() not set errno,
2021-10-16) to stop caring about the errno at all.
When we fail to resolve "HEAD" after the sequencer runs it doesn't
really help to say what the "errno" value is, since the fake backend
errno may or may not reflect anything real about the state of the
".git/HEAD". With the upcoming reftable backend this fakery will
become even more pronounced.
So let's just die() instead of die_errno() here. This will also help
simplify the refs_resolve_ref_unsafe() API. This was the only user of
it that wasn't ignoring the "failure_errno" output parameter.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Now that reset_head() can handle the initial checkout of onto
correctly use it in the "merge" backend instead of forking "git
checkout". This opens the way for us to stop calling the
post-checkout hook in the future. Not running "git checkout" means
that "rebase -i/m" no longer recurse submodules when checking out
"onto" (thanks to Philippe Blain for pointing this out). As the rest
of rebase does not know what to do with submodules this is probably a
good thing. When using merge-ort rebase ought be able to handle
submodules correctly if it parsed the submodule config, such a change
is left for a future patch series.
The "apply" based rebase has avoided forking git checkout
since ac7f467fef ("builtin/rebase: support running "git rebase
<upstream>"", 2018-08-07). The code that handles the checkout was
moved into libgit by b309a97108 ("reset: extract reset_head() from
rebase", 2020-04-07).
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
At the start of a rebase, ORIG_HEAD is updated to the tip of the
branch being rebased. Unfortunately reset_head() always uses the
current value of HEAD for this which is incorrect if the rebase is
started with "git rebase <upstream> <branch>" as in that case
ORIG_HEAD should be updated to <branch>. This only affects the "apply"
backend as the "merge" backend does not yet use reset_head() for the
initial checkout. Fix this by passing in orig_head when calling
reset_head() and add some regression tests.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
move_to_original_branch() passes the message intended for the branch
reflog as `orig_head_msg`. Fix this by adding a `branch_msg` member to
struct reset_head_opts and add a regression test. Note that these
reflog messages do not respect GIT_REFLOG_ACTION. They are not alone
in that and will be fixed in a future series.
The "merge" backend already has tests that check both the branch and
HEAD reflogs.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This function takes a confusingly large number of parameters which
makes it difficult to remember which order to pass them in. The
following commits will add a couple more parameters which makes the
problem worse. To address this change the function to take a struct of
options. Using a struct means that it is no longer necessary to
remember which order to pass the parameters in and anyone reading the
code can easily see which value is passed to each parameter.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If ORIG_HEAD is not set by passing RESET_ORIG_HEAD then there is no
need to pass anything for reflog_orig_head. In addition to the callers
fixed in this commit move_to_original_branch() also passes
reflog_orig_head without setting ORIG_HEAD. That caller is mistakenly
passing the message it wants to put in the branch reflog which is not
currently possible so we delay fixing that caller until we can pass
the message as the branch reflog.
A later commit will make it a BUG() to pass reflog_orig_head without
RESET_ORIG_HEAD, that changes cannot be done here as it needs to wait
for move_to_original_branch() to be fixed first.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The default_reflog parameter of create_autostash() is passed to
reset_head(). However as creating a stash does not involve updating
any refs the parameter is not used by reset_head(). Removing the
parameter from create_autostash() simplifies the callers.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This parameter is only needed when a ref is going to be updated and
the caller does not pass an explicit reflog message. Callers that are
only discarding uncommitted changes in the working tree such as such
as "rebase --skip" or create_autostash() do not update any refs so
should not have to worry about passing this parameter.
This change is not intended to have any user visible changes. The
pointer comparison between `oid` and `&head_oid` checks that the
caller did not pass an oid to be checked out. As no callers pass
RESET_HEAD_RUN_POST_CHECKOUT_HOOK without passing an oid there are
no changes to when the post-checkout hook is run. As update_ref() only
updates the ref if the oid passed to it differs from the current ref
there are no changes to when HEAD is updated.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In the next commit we will stop trying to update HEAD when we are
removing uncommitted changes from the working tree. Move the code that
updates the refs to its own function in preparation for that.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The only use of the action parameter is to setup the error messages
for unpack_trees(). All but two cases pass either "checkout" or
"reset". The case that passes "reset --hard" would be better passing
"reset" so that the error messages match the builtin reset command
like all the other callers that are doing a reset. The case that
passes "Fast-forwarded" is only updating HEAD and so the parameter is
unused in that case as it does not call unpack_trees(). The value to
pass to setup_unpack_trees_porcelain() can be determined by checking
whether flags contains RESET_HEAD_HARD without the caller having to
specify it.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The hook should only be run if the worktree and refs were successfully
updated. This primarily affects "rebase --apply" but also "rebase
--merge" when it fast-forwards.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If "git rebase [--apply|--merge] <upstream> <branch>" detects that
<upstream> is an ancestor of <branch> then it will fast-forward and
checkout <branch>. Normally a checkout or picking a commit during a
rebase will refuse to overwrite untracked files, however rebase does
overwrite untracked files when checking out <branch>.
The fix is to only set reset in `unpack_tree_opts` if flags contains
`RESET_HEAD_HARD`. t5403 may seem like an odd home for the new test
but it will be extended in the next commit to check that the
post-checkout hook is not run when the checkout fails.
The test for `!detach_head` dates back to the
original implementation of reset_head() in
ac7f467fef ("builtin/rebase: support running "git rebase <upstream>"",
2018-08-07) and was correct until e65123a71d
("builtin rebase: support `git rebase <upstream> <switch-to>`",
2018-09-04) started using reset_head() to checkout <switch-to> when
fast-forwarding.
Note that 480d3d6bf9 ("Change unpack_trees' 'reset' flag into an
enum", 2021-09-27) also fixes this bug as it changes reset_head() to
never remove untracked files. I think this fix is still worthwhile as
it makes it clear that the same settings are used for detached and
non-detached checkouts.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a rebase started with "rebase [--apply|--merge] <upstream> <branch>"
detects that <upstream> is an ancestor of <branch> then it fast-forwards
and checks out <branch>. Unfortunately in that case it passed the null
oid as the first argument to the post-checkout hook rather than the oid
of HEAD.
A side effect of this change is that the call to update_ref() which
updates HEAD now always receives the old value of HEAD. This provides
protection against another process updating HEAD during the checkout.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
These tests only test the default backend and do not check that the
arguments passed to the hook are correct. Fix this by running the
tests with both backends and adding checks for the hook arguments.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This code is heavily indented and it will be convenient later in the
series to have it in its own function.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In commits bc3ae46b42 ("rebase: do not attempt to remove
startup_info->original_cwd", 2021-12-09) and 0fce211ccc ("stash: do not
attempt to remove startup_info->original_cwd", 2021-12-09), we wanted to
allow the subprocess to know which directory the parent process was
running from, so that the subprocess could protect it. However...
When run from a non-main worktree, setup_git_directory() will note
that the discovered git directory
(/PATH/TO/.git/worktree/non-main-worktree) does not match
DEFAULT_GIT_DIR_ENVIRONMENT (see setup_discovered_git_dir()), and
decide to set GIT_DIR in the environment. This matters because...
Whenever git is run with the GIT_DIR environment variable set, and
GIT_WORK_TREE not set, it presumes that '.' is the working tree. So...
This combination results in the subcommand being very confused about
the working tree. Fix it by also setting the GIT_WORK_TREE environment
variable along with setting cmd.dir.
A possibly more involved fix we could consider for later would be to
make setup.c set GIT_WORK_TREE whenever (a) it discovers both the git
directory and the working tree and (b) it decides to set GIT_DIR in the
environment. I did not attempt that here as such would be too big of a
change for a 2.35.1 release.
Test-case-by: Glen Choo <chooglen@google.com>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When cloning a branchless and tagless but not refless remote using
protocol v0 or v1, Git calls transport_fetch_refs() with an empty ref
list. This makes the clone fail with the message "remote transport
reported error".
Git should have refrained from calling transport_fetch_refs(), just like
it does in the case that the remote is refless. Therefore, teach Git to
do this.
In protocol v2, this does not happen because the client passes
ref-prefix arguments that filter out non-branches and non-tags in the
ref advertisement, making the remote appear empty.
Note that this bug concerns logic in builtin/clone.c and only affects
cloning, not fetching.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have a copy of uncompress2() implementation in compat/ so that we
can build with an older version of zlib that lack the function, and
the build procedure selects if it is used via the NO_UNCOMPRESS2
$(MAKE) variable. This is yet another "annoying" knob the porters
need to tweak on platforms that are not common enough to have the
default set in the config.mak.uname file.
Attempt to instead ask the system header <zlib.h> to decide if we
need the compatibility implementation. This is a deviation from the
way we have been handling the "compatiblity" features so far, and if
it can be done cleanly enough, it could work as a model for features
that need compatibility definition we discover in the future. With
that goal in mind, avoid expedient but ugly hacks, like shoving the
code that is conditionally compiled into an unrelated .c file, which
may not work in future cases---instead, take an approach that uses a
file that is independently compiled and stands on its own.
Compile and link compat/zlib-uncompress2.c file unconditionally, but
conditionally hide the implementation behind #if/#endif when zlib
version is 1.2.9 or newer, and unconditionally archive the resulting
object file in the libgit.a to be picked up by the linker.
There are a few things to note in the shape of the code base after
this change:
- We no longer use NO_UNCOMPRESS2 knob; if the system header
<zlib.h> claims a version that is more cent than the library
actually is, this would break, but it is easy to add it back when
we find such a system.
- The object file compat/zlib-uncompress2.o is always compiled and
archived in libgit.a, just like a few other compat/ object files
already are.
- The inclusion of <zlib.h> is done in <git-compat-util.h>; we used
to do so from <cache.h> which includes <git-compat-util.h> as the
first thing it does, so from the *.c codes, there is no practical
change.
- Until objects in libgit.a that is already used gains a reference
to the function, the reftable code will be the only one that
wants it, so libgit.a on the linker command line needs to appear
once more at the end to satisify the mutual dependency.
- Beat found a trick used by OpenSSL to avoid making the
conditionally-compiled object truly empty (apparently because
they had to deal with compilers that do not want to see an
effectively empty input file). Our compat/zlib-uncompress2.c
file borrows the same trick for portabilty.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Helped-by: Beat Bolli <dev+git@drbeat.li>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
mem_pool_alloc uses sizeof(uintmax_t) as a proxy for what should be
_Alignof(max_align_t) in C11. On most architectures this is sufficient
(though on m68k it is in fact overly strict, since the de-facto ABI,
which differs from the specified System V ABI, has the maximum alignment
of all types as 2 bytes), but on CHERI, and thus Arm's Morello
prototype, it is insufficient for any type that stores a pointer, which
must be aligned to 128 bits (on 64-bit architectures extended with
CHERI), whilst uintmax_t is a 64-bit integer.
Fix this by introducing our own approximation for max_align_t and a
means to compute _Alignof it without relying on C11. Currently this
union only contains uintmax_t and void *, but more types can be added as
needed.
Signed-off-by: Jessica Clarke <jrtc27@jrtc27.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We added an unrelated sanity checking that leads to a BUG() while
plugging a leak, which triggered in a repository with symrefs in
the local branch namespace that point at a ref outside. Partially
revert the change to avoid triggering the BUG().
* ab/checkout-branch-info-leakfix:
checkout: avoid BUG() when hitting a broken repository
... at least for now. So let's error out if we are even trying to
initialize the split index when the index is sparse, or when trying to
write the split index extension for a sparse index.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 61feddcdf2 (tests: disable GIT_TEST_SPLIT_INDEX for sparse index
tests, 2021-08-26), it was already called out that the split index
feature is incompatible with the sparse index feature, and its commit
message wondered aloud whether more checks would be required to ensure
that the split index and sparse index features aren't enabled at the
same time.
We are about to introduce such additional checks, and indeed, t1091
would utterly fail with them. Therefore, let's preemptively disable the
split index for the entirety of t1091.
This partially reverts above-mentioned patch because it covered only one
test case whereas we want to cover the entire test script.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 6e773527b6 (sparse-index: convert from full to sparse, 2021-03-30),
we introduced initial support for a sparse index, and were careful to
avoid converting to a sparse index in the presence of a split index.
However, when we _just_ read a freshly-initialized index, it might not
contain a split index even if _writing_ it will add one by virtue of
being asked for via the `GIT_TEST_SPLIT_INDEX` variable.
We did not notice any problems with checking _only_ for `split_index`
(and not `GIT_TEST_SPLIT_INDEX`) right until both
`vd/sparse-sparsity-fix-on-read` _and_ `vd/sparse-reset` were merged.
Those two topics' interplay triggers a bug in conjunction with running
t1091.15 when `GIT_TEST_SPLIT_INDEX=true` in the following way:
`vd/sparse-sparsity-fix-on-read` ensures that the index is made sparse
right after reading, and `vd/sparse-reset` ensures that the index is
made non-sparse again unless running in the `--soft` mode. Since the
split index feature is incompatible with the sparse index feature, we
see a symptom like this:
fatal: position for replacement 4 exceeds base index size 4
Let's fix this by avoiding the conversion to a sparse index when
`GIT_TEST_SPLIT_INDEX=true`.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When 9081a421 (checkout: fix "branch info" memory leaks, 2021-11-16)
cleaned up existing memory leaks, we added an unrelated sanity check
to ensure that a local branch is truly local and not a symref to
elsewhere that dies with BUG() otherwise. This was misguided in two
ways. First of all, such a tightening did not belong to a leak-fix
patch. And the condition it detected was *not* a bug in our program
but a problem in user data, where warning() or die() would have been
more appropriate.
As the condition is not fatal (the result of computing the local
branch name in the code that is involved in the faulty check is only
used as a textual label for the commit), let's revert the code to
the original state, i.e. strip "refs/heads/" to compute the local
branch name if possible, and otherwise leave it NULL. The consumer
of the information in merge_working_tree() is prepared to see NULL
in there and act accordingly.
cf. https://bugzilla.redhat.com/show_bug.cgi?id=2042920
Reported-by: Petr Šplíchal <psplicha@redhat.com>
Reported-by: Todd Zullinger <tmz@pobox.com>
Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There were two commit_lists created in cmd_merge() that were only
conditionally free()'d. Add a quick conditional call to
free_commit_list() for each of them at the end of the function.
Testing this commit against t6404 under valgrind shows that this patch
fixes the following two leaks:
16 bytes in 1 blocks are definitely lost in loss record 16 of 126
at 0x484086F: malloc (vg_replace_malloc.c:380)
by 0x69FFEB: do_xmalloc (wrapper.c:41)
by 0x6A0073: xmalloc (wrapper.c:62)
by 0x52A72D: commit_list_insert (commit.c:556)
by 0x47FC93: reduce_parents (merge.c:1114)
by 0x4801EE: collect_parents (merge.c:1214)
by 0x480B56: cmd_merge (merge.c:1465)
by 0x40686E: run_builtin (git.c:464)
by 0x406C51: handle_builtin (git.c:716)
by 0x406E96: run_argv (git.c:783)
by 0x40730A: cmd_main (git.c:914)
by 0x4E7DFA: main (common-main.c:56)
8 (16 direct, 32 indirect) bytes in 1 blocks are definitely lost in \
loss record 61 of 126
at 0x484086F: malloc (vg_replace_malloc.c:380)
by 0x69FFEB: do_xmalloc (wrapper.c:41)
by 0x6A0073: xmalloc (wrapper.c:62)
by 0x52A72D: commit_list_insert (commit.c:556)
by 0x52A8F2: commit_list_insert_by_date (commit.c:620)
by 0x5270AC: get_merge_bases_many_0 (commit-reach.c:413)
by 0x52716C: repo_get_merge_bases (commit-reach.c:438)
by 0x480E5A: cmd_merge (merge.c:1520)
by 0x40686E: run_builtin (git.c:464)
by 0x406C51: handle_builtin (git.c:716)
by 0x406E96: run_argv (git.c:783)
by 0x40730A: cmd_main (git.c:914)
There are still 3 leaks in chdir_notify_register() after this, but
chdir_notify_register() has been brought up on the list before and folks
were not a fan of fixing those, so I'm not touching them.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The documentation for merge_incore_recursive(), modelled after
merge_recursive(), notes that
merge_bases will be consumed (emptied) so make a copy if you need it
However, in merge_ort_internal() (which merge_incore_recursive() calls),
it runs
merged_merge_bases = pop_commit(&merge_bases);
...
for (iter = merge_bases; iter; iter = iter->next) {
...
}
In other words, it only consumes the *first* entry of merge_bases, and
the rest it iterates through. If it iterated through all of them, the
caller could be responsible for free'ing the memory. If it consumed all
of them, the current documentation would be correct and the callers
would need to do nothing. The current middle ground makes it impossible
for callers to avoid memory leaks, since any attempt to use the
merge_bases it passes in would result in a use-after-free.
It turns out this part of the code was copied from merge-recursive.c,
which has had the same bug for 15.5 years. However, since we are trying
to keep merge-recursive.c stable as we sunset it, let's just fix the
leak in in merge_ort_internal() by having it actually consume all the
elements of the merge_bases commit_list.
Testing this commit against t6404 (the first testcase specifically
about recursive merges) under valgrind shows that this patch fixes
the following leak:
32 (16 direct, 16 indirect) bytes in 1 blocks are definitely lost \
in loss record 49 of 126
at 0x484086F: malloc (vg_replace_malloc.c:380)
by 0x69FFEB: do_xmalloc (wrapper.c:41)
by 0x6A0073: xmalloc (wrapper.c:62)
by 0x52A72D: commit_list_insert (commit.c:556)
by 0x47EC86: try_merge_strategy (merge.c:751)
by 0x48143B: cmd_merge (merge.c:1679)
by 0x40686E: run_builtin (git.c:464)
by 0x406C51: handle_builtin (git.c:716)
by 0x406E96: run_argv (git.c:783)
by 0x40730A: cmd_main (git.c:914)
by 0x4E7DFA: main (common-main.c:56)
Reported-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When creating the sparse-checkout file, Git does not create the leading
directory, "$GIT_DIR/info", if it does not exist. This causes problems
if the repository does not have that directory. Therefore, ensure that
the leading directory is created.
This is the only "open" in builtin/sparse-checkout.c that does not have
a leading directory check. (The other one in write_patterns_and_update()
does.)
Note that the test needs to explicitly specify a template when running
"git init" because the default template used in the tests has the
"info/" directory included.
Helped-by: Jose Lopes <jabolopes@google.com>
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
