| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
In lock_ref_oid_basic() we'll happily lock a reference that doesn't
exist yet. That's normal, and is how references are initially born,
but we don't need to retain checks here in lock_ref_oid_basic() about
the state of the ref, when what we're checking is either checked
already, or something we're about to discover by trying to lock the
ref with raceproof_create_file().
The one exception is the caller in files_reflog_expire(), who passes
us a "type" to find out if the reference is a symref or not. We can
move the that logic over to that caller, which can now defer its
discovery of whether or not the ref is a symref until it's needed. In
the preceding commit an exhaustive regression test was added for that
case in a new test in "t1417-reflog-updateref.sh".
The improved diagnostics here were added in
5b2d8d6f218 (lock_ref_sha1_basic(): improve diagnostics for ref D/F
conflicts, 2015-05-11), and then much of the surrounding code went
away recently in my 245fbba46d6 (refs/files: remove unused "errno ==
EISDIR" code, 2021-08-23).
The refs_resolve_ref_unsafe() code being removed here looks like it
should be tasked with doing that, but it's actually redundant to other
code.
The reason for that is as noted in 245fbba46d6 this once widely used
function now only has a handful of callers left, which all handle this
case themselves.
To the extent that we're racy between their check and ours removing
this check actually improves the situation, as we'll be doing fewer
things between the not-under-lock initial check and acquiring the
lock.
Why this is OK for all the remaining callers of lock_ref_oid_basic()
is noted below. There are only two of those callers:
* "git branch -[cm] <oldbranch> <newbranch>":
In files_copy_or_rename_ref() we'll call this when we copy or rename
refs via rename_ref() and copy_ref(). but only after we've checked
if the refname exists already via its own call to
refs_resolve_ref_unsafe() and refs_rename_ref_available().
As the updated comment to the latter here notes neither of those are
actually needed. If we delete not only this code but also
refs_rename_ref_available() we'll do just fine, we'll just emit a
less friendly error message if e.g. "git branch -m A B/C" would have
a D/F conflict with a "B" file.
Actually we'd probably die before that in case reflogs for the
branch existed, i.e. when the try to rename() or copy_file() the
relevant reflog, since if we've got a D/F conflict with a branch
name we'll probably also have the same with its reflogs (but not
necessarily, we might have reflogs, but it might not).
As some #leftoverbits that code seems buggy to me, i.e. the reflog
"protocol" should be to get a lock on the main ref, and then perform
ref and/or reflog operations. That code dates back to
c976d415e53 (git-branch: add options and tests for branch renaming,
2006-11-28) and probably pre-dated the solidifying of that
convention. But in any case, that edge case is not our bug or
problem right now.
* "git reflog expire <ref>":
In files_reflog_expire() we'll call this without previous ref
existence checking in files-backend.c, but that code is in turn
called by code that's just finished checking if the refname whose
reflog we're expiring exists.
See ae35e16cd43 (reflog expire: don't lock reflogs using previously
seen OID, 2021-08-23) for the current state of that code, and
5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic
references, 2015-03-03) for the code we'd break if we only did a
"update = !!ref" here, which is covered by the aforementioned
regression test in "t1417-reflog-updateref.sh".
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Add tests that cover blindspots in "git reflog delete --updateref"
behavior. Before this change removing the "type & REF_ISSYMREF" check
added in 5e6f003ca8a (reflog_expire(): ignore --updateref for symbolic
references, 2015-03-03) would not fail any tests.
The "--updateref" option was added in 55f10565371 (git-reflog: add
option --updateref to write the last reflog sha1 into the ref,
2008-02-22) for use in git-stash.sh, see e25d5f9c82e (git-stash: add
new 'drop' subcommand, 2008-02-22).
Even though the regression test I need is just the "C" case here,
let's test all these combinations for good measure. I started out
doing these as a for-loop, but I think this is more readable.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Move the refs_rename_ref_available() function into
"refs/files-backend.c". It is file-backend specific.
This function was added in 5fe7d825da8 (refs.c: pass a list of names
to skip to is_refname_available, 2014-05-01) as rename_ref_available()
and was only ever used in this one file-backend specific codepath. So
let's move it there.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Change the parse_loose_ref_contents() function to stop setting "errno"
and failure, and to instead pass up a "failure_errno" via a
parameter. This requires changing its callers to do the same.
The EINVAL error from parse_loose_ref_contents is used in files-backend
to create a custom error message.
In untangling this we discovered a tricky edge case. The
refs_read_special_head() function was relying on
parse_loose_ref_contents() setting EINVAL.
By converting it to use "saved_errno" we can migrate away from "errno"
in this part of the code entirely, and do away with an existing
"save_errno" pattern, its only purpose was to not clobber the "errno"
we previously needed at the end of files_read_raw_ref().
Let's assert that we can do that by not having files_read_raw_ref()
itself operate on *failure_errno in addition to passing it on. Instead
we'll assert that if we return non-zero we actually do set errno, thus
assuring ourselves and callers that they can trust the resulting
"failure_errno".
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Add a "failure_errno" to refs_read_raw_ref(), his allows
refs_werrres_ref_unsafe() to pass along its "failure_errno", as a
first step before its own callers are migrated to pass it further up
the chain.
We are leaving out out the refs_read_special_head() in
refs_read_raw_ref() for now, as noted in a subsequent commit moving it
to "failure_errno" will require some special consideration.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Add a new refs_werrres_ref_unsafe() function, which is like
refs_resolve_ref_unsafe() except that it explicitly saves away the
"errno" to a passed-in parameter, the refs_resolve_ref_unsafe() then
becomes a wrapper for it.
In subsequent commits we'll migrate code over to it, before finally
making "refs_resolve_ref_unsafe()" with an "errno" parameter the
canonical version, so this this function exists only so that we can
incrementally migrate callers, it will be going away in a subsequent
commit.
As the added comment notes has a rather tortured name to be the same
length as "refs_resolve_ref_unsafe", to avoid churn as we won't need
to re-indent the argument lists, similarly the documentation and
structure of it in refs.h is designed to minimize a diff in a
subsequent commit, where that documentation will be added to the new
refs_resolve_ref_unsafe().
At the end of this migration the "meaningful errno" TODO item left in
76d70dc0c63 (refs.c: make resolve_ref_unsafe set errno to something
meaningful on error, 2014-06-20) will be resolved.
As can be seen from the use of refs_read_raw_ref() we'll also need to
convert some functions that the new refs_werrres_ref_unsafe() itself
calls to take this "failure_errno". That will be done in subsequent
commits.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Add a test for "git branch" to cover the case where .git/refs is
symlinked. To check availability, refs_verify_refname_available() will
run refs_read_raw_ref() on each prefix, leading to a read() from
.git/refs (which is a directory).
It would probably be more robust to re-issue the lstat() as a normal
stat(), in which case, we would fall back to the directory case, but
for now let's just test for the existing behavior as-is. This test
covers a regression in a commit that only ever made it to "next", see
[1].
1. http://lore.kernel.org/git/pull.1068.git.git.1629203489546.gitgitgadget@gmail.com
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
builtin/gc.c has two ways of checking if multi-pack-index is enabled:
- git_config_get_bool() in incremental_repack_auto_condition()
- the_repository->settings.core_multi_pack_index in
maintenance_task_incremental_repack()
The two implementations have existed since the incremental-repack task
was introduced in e841a79a13 (maintenance: add incremental-repack auto
condition, 2020-09-25). These two values can diverge because
prepare_repo_settings() enables the feature in the_repository->settings
by default.
In the case where core.multiPackIndex is not set in the config, the auto
condition would fail, causing the incremental-repack task to not be
run. Because we always want to consider the default values, we should
always use the_repository->settings.
Standardize on using the_repository->settings.core_multi_pack_index to
check if multi-pack-index is enabled.
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Like the previous commit, change fsck to check the
"core_multi_pack_index" variable set in "repo-settings.c" instead of
reading the "core.multiPackIndex" config variable. This fixes a bug
where we wouldn't verify midx if the config key was missing. This bug
was introduced in 18e449f86b (midx: enable core.multiPackIndex by
default, 2020-09-25) where core.multiPackIndex was turned on by default.
Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Change fsck to check the "core_commit_graph" variable set in
"repo-settings.c" instead of reading the "core.commitGraph" variable.
This fixes a bug where we wouldn't verify the commit-graph if the
config key was missing. This bug was introduced in
31b1de6a09 (commit-graph: turn on commit-graph by default, 2019-08-13),
where core.commitGraph was turned on by default.
Add tests to "t5318-commit-graph.sh" to verify that fsck checks the
commit-graph as expected for the 3 values of core.commitGraph. Also,
disable GIT_TEST_COMMIT_GRAPH in t/t0410-partial-clone.sh because some
test cases use fsck in ways that assume that commit-graph checking is
disabled.
Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Glen Choo <chooglen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
gc/use-repo-settings
* ab/ignore-replace-while-working-on-commit-graph:
commit-graph: don't consider "replace" objects with "verify"
commit-graph tests: fix another graph_git_two_modes() helper
commit-graph tests: fix error-hiding graph_git_two_modes() helper
|
|
This function was added in 4981fe750b1 (pkt-line: share
buffer/descriptor reading implementation, 2013-02-23), but in
01f9ec64c8a (Use packet_reader instead of packet_read_line,
2018-12-29) the code that was using it was removed.
Since it's being removed we can in turn remove the "src" and "src_len"
arguments to packet_read(), all the remaining users just passed a
NULL/NULL pair to it.
That function is only a thin wrapper for packet_read_with_status()
which still needs those arguments, but for the thin packet_read()
convenience wrapper we can do away with it for now.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
This function was added in f1f4d8acf40 (pkt-line: add
packet_buf_write_len function, 2018-03-15) for use in
0f1dc53f45d (remote-curl: implement stateless-connect command,
2018-03-15).
In a97d00799a1 (remote-curl: use post_rpc() for protocol v2 also,
2019-02-21) that only user of it went away, let's remove it.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When writing a MIDX, we atomically move the new MIDX into place via
commit_lock_file(), but do not check to see if that call was successful.
Make sure that we do check in order to prevent us from incorrectly
reporting that we wrote a new MIDX if we actually encountered an error.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Apply similar treatment as in the last commit to the MIDX `repack`
operation.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Before a new MIDX can be written, expire_midx_packs() first loads the
existing MIDX, figures out which packs can be expired, and then writes a
new MIDX based on that information.
In order to load the existing MIDX, it uses load_multi_pack_index(),
which mmaps the multi-pack-index file, but does not store the resulting
`struct multi_pack_index *` in the object store.
write_midx_internal() also needs to open the existing MIDX, and it does
so by iterating the results of get_multi_pack_index(), so that it reuses
the same pointer held by the object store. But before it can move the
new MIDX into place, it close_object_store() to munmap() the
multi-pack-index file to accommodate platforms like Windows which don't
allow overwriting files which are memory mapped.
That's where things get weird. Since expire_midx_packs has its own
*separate* memory mapped copy of the MIDX, the MIDX file is still memory
mapped! Interestingly, this doesn't seem to cause a problem in our
tests. (I believe that this has much more to do with my own lack of
familiarity with Windows than it does a lack of coverage in our tests).
In any case, we can side-step the whole issue by teaching
expire_midx_packs() to use the `struct multi_pack_index` pointer it
found via the object store instead of maintain its own copy. That way,
when write_midx_internal() calls `close_object_store()`, we know that
there are no memory mapped copies of the MIDX laying around.
A couple of other small notes about this patch:
- As far as I can tell, passing `local == 1` to the call to
load_multi_pack_index() was an error, since object_dir could be an
alternate. But it doesn't matter, since even though we write
`m->local = 1`, we never read that field back later on.
- Setting `m = NULL` after write_midx_internal() was likely to prevent
a double-free back from when that function took a `struct
multi_pack_index *` that it called close_midx() on itself. We can
rely on write_midx_internal() to call that for us now.
Finally, this enforces the same "the value of --object-dir must be the
local object store, or an alternate" rule from f57a739691 (midx: avoid
opening multiple MIDXs when writing, 2021-09-01) to the `expire`
sub-command, too.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The first thing that write_midx_internal() does is load the MIDX
corresponding to the given object directory, if one is present.
Prepare for other functions in midx.c to do the same thing by extracting
that operation out to a small helper function.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
* tb/repack-write-midx:
test-read-midx: fix leak of bitmap_index struct
builtin/repack.c: pass `--refs-snapshot` when writing bitmaps
builtin/repack.c: make largest pack preferred
builtin/repack.c: support writing a MIDX while repacking
builtin/repack.c: extract showing progress to a variable
builtin/repack.c: rename variables that deal with non-kept packs
builtin/repack.c: keep track of existing packs unconditionally
midx: preliminary support for `--refs-snapshot`
builtin/multi-pack-index.c: support `--stdin-packs` mode
midx: expose `write_midx_file_only()` publicly
|
|
If we attempt to grep non-ascii log message text with an ascii pattern, we
run into the following issue:
$ git log --color --author='.var.*Bjar' -1 origin/master | grep ^Author
grep: (standard input): binary file matches
So, to fix this teach the grep code to use PCRE2_UTF, as long as the log
output is encoded in UTF-8.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Hamza Mahfooz <someguy@effective-light.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Speed up the "lint-docs" target by making it non-.PHONY. Similar to my
c234e8a0ecf (Makefile: make the "sparse" target non-.PHONY,
2021-09-23). We'll now create empty files corresponding to a
dependency graph for each of these lint scripts.
This speeds things up a bit[1], and makes the output correspond to any
in-tree changes we have:
$ touch git-add.txt; make lint-docs; make lint-docs
GEN cmd-list.made
GEN doc.dep
LINT GITLINK git-add.txt
LINT MAN END git-add.txt
LINT MAN SEC git-add.txt
make: Nothing to be done for 'lint-docs'.
As with the "sparse" target changes this has a hard dependency on the
use of ".DELETE_ON_ERROR" in the Makefile, added here in
db10fc6c09f (doc: simplify Makefile using .DELETE_ON_ERROR,
2021-05-21). This method also depends on the output for us emitting
any errors on STDERR (fixed in a preceding commit), as well us these
scripts exiting with non-zero on any errors (which they were already
doing).
1.
$ git show HEAD~:Documentation/Makefile >Makefile.old
$ hyperfine --warmup 2 -L f ",.old" 'make -j1 -f Makefile{f} lint-docs'
Benchmark #1: make -j1 -f Makefile lint-docs
Time (mean ± σ): 60.8 ms ± 1.4 ms [User: 58.7 ms, System: 2.5 ms]
Range (min … max): 58.9 ms … 64.0 ms 48 runs
Benchmark #2: make -j1 -f Makefile.old lint-docs
Time (mean ± σ): 84.0 ms ± 1.5 ms [User: 78.6 ms, System: 5.7 ms]
Range (min … max): 81.8 ms … 87.8 ms 35 runs
Summary
'make -j1 -f Makefile lint-docs' ran
1.38 ± 0.04 times faster than 'make -j1 -f Makefile.old lint-docs'
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Extend the trick we use to speed up the "clean" target to also extend
to the "lint-docs" target. See 54df87555b1 (Documentation/Makefile:
conditionally include doc.dep, 2020-12-08) for the "clean"
implementation.
The "doc-lint" target only depends on *.txt files, so we don't need to
generate GIT-VERSION-FILE etc. if that's all we're doing. This makes
the "make lint-docs" target more than 2x as fast:
$ git show HEAD~:Documentation/Makefile >Makefile.old
$ hyperfine -L f ",.old" 'make -f Makefile{f} lint-docs'
Benchmark #1: make -f Makefile lint-docs
Time (mean ± σ): 100.2 ms ± 1.3 ms [User: 93.7 ms, System: 6.7 ms]
Range (min … max): 98.4 ms … 103.1 ms 29 runs
Benchmark #2: make -f Makefile.old lint-docs
Time (mean ± σ): 220.0 ms ± 20.0 ms [User: 206.0 ms, System: 18.0 ms]
Range (min … max): 206.6 ms … 267.5 ms 11 runs
Summary
'make -f Makefile lint-docs' ran
2.19 ± 0.20 times faster than 'make -f Makefile.old lint-docs'
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Have all of the scripts invoked by "make check-docs" emit their output
on STDERR. This does not currently matter due to the way we're
invoking them, but will in a subsequent change. It's a good idea to do
this in any case for consistency with other tools we invoke.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Fix the broken "make lint-docs" (or "make check-docs" at the
top-level) target, which has been broken since my cafd9828e89 (doc
lint: lint and fix missing "GIT" end sections, 2021-04-09).
The CI for "seen" is emitting an error about a broken gitlink, but due
to there being 3x scripts chained via ";" instead of "&&" we're not
carrying forward the non-zero exit code.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In 459b8d22e54 (tests: do not borrow from COPYING and README from the
real source, 2015-02-15) tests that used "lib-diff.sh" (called
"diff-lib.sh" then) were made to stop relying on the top-level COPYING
file, but we still had other tests that referenced it.
Let's move them over to use the "COPYING_test_data" utility function
introduced in the preceding commit, and in the case of the one test
that needed the "README" file use a ROT 13 version of that "COPYING"
test data. That test added in afd222967c6 (Extend testing git-mv for
renaming of subdirectories, 2006-07-26) just needs more test data that's not the same as the "COPYING" test data, so a ROT 13 version will do.
This change removes the last references to ../{README,COPYING} in the
test suite.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Follow-up the change in 459b8d22e54 (tests: do not borrow from COPYING
and README from the real source, 2015-02-15) by not shipping a full
copy of older versions of the top-level "COPYING" and "README" files.
The tests that use them just need the small blurb at the top of
"COPYING" as test data, or mock data that's dissimilar. Let's provide
that with a "COPYING_test_data" function instead.
We're not replacing this with some other generic test
data (e.g. "lorum ipsum") because these tests require test file header
to be the old "COPYING" file. See e.g. "t4003-diff-rename-1.sh" which
changes the file, and then does full "test_cmp" comparisons on the
resulting "git diff" output.
This change only changes tests that used the "lib-diff.sh" library,
but splits up what they need into a new "lib-diff-data.sh". A
subsequent commit will change related tests that were missed in
459b8d22e54.
For the test in "t4008-diff-break-rewrite.sh" the "README" file can go
away in favor of echoing the line "some dissimilar content" to a file
in the one test that needed it.
The point of that test is to start with files "A" and "B", and then
have A be more similar to the state of "B" than to its old version (by
copying over the content from the "COPYING" file). Just comparing the
pre-image of "some dissimilar content" and later a munged version of
the "COPYING" output serves that purpose.
While we're at it get rid of a stray "echo $tree" debugging line added
in 15d061b435a ([PATCH] Fix the way diffcore-rename records unremoved
source., 2005-05-27), and stop calling "hash-object" to get the hash
of an object we've just added to the index. We can instead extract
that information from the index itself with "rev-parse".
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Extend the code added in d6538246d3d (commit-graph: not compatible
with replace objects, 2018-08-20) which ignored replace objects in the
"write" command to ignore it in the "verify" command too.
We can just move this assignment to the cmd_commit_graph(), it
dispatches to "write" and "verify", and we're unlikely to ever get a
sub-command that would like to consider replace refs.
This will make tests added in eddc1f556cd (mktag tests: test
update-ref and reachable fsck, 2021-06-17) pass in combination with
the "GIT_TEST_COMMIT_GRAPH" mode added in 859fdc0c3cf (commit-graph:
define GIT_TEST_COMMIT_GRAPH, 2018-08-29), except that mode is
currently broken (but is being fixed concurrently). See the discussion
starting at [1].
1. https://lore.kernel.org/git/87wnmihswp.fsf@evledraar.gmail.com/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
In 135a7123755 (commit-graph: add --split option to builtin,
2019-06-18) this function was copy/pasted to the split commit-graph
tests, as in the preceding commit we need to fix this to use
&&-chaining, so it won't be hiding errors.
Unlike its sister function in "t5318-commit-graph.sh", which we got
lucky with, this one was hiding a real test failure. A tests added in
c523035cbd8 (commit-graph: allow cross-alternate chains, 2019-06-18)
has never worked as intended. Unlike most other graph_git_behavior
uses in this file it clones the repository into a sub-directory, so
we'll need to refer to "commits/6" as "origin/commits/6".
It's not easy to simply move the "graph_git_behavior" to the test
above it, since it itself spawns a "test_expect_success". Let's
instead add support to "graph_git_behavior()" and
"graph_git_two_modes()" to pass a "-C" argument to git.
We also need to add a "test -d fork" here, because otherwise we'll
fail on e.g.:
GIT_SKIP_TESTS=t5324.13 ./t5324-split-commit-graph.sh
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
The graph_git_two_modes() helper added in 177722b3442 (commit:
integrate commit graph with commit parsing, 2018-04-10) didn't
&&-chain its "git commit-graph" invocations, which as can be seen with
SANITIZE=leak will happily mark tests as passing if both of these
commands die, since test_cmp() will be comparing two empty files.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Try to re-chmod the trash directory on startup if we fail to "rm -rf"
it. This fixes problems where the test leaves the trash directory
behind in a bad permission state for whatever reason.
This fixes an interaction between [1] where t0004-unwritable.sh was
made to use "test_when_finished" for cleanup, and [2] which added the
"--immediate" mode. If a test in this file failed when running with
"--immediate" we wouldn't run the "test_when_finished" block, which
re-chmods the ".git/objects" directory (see [1]).
This can be demonstrated as e.g. (output snipped for less verbosity):
$ ./t0004-unwritable.sh --run=3 --immediate
ok 1 # skip setup (--run)
ok 2 # skip write-tree should notice unwritable repository (--run)
not ok 3 - commit should notice unwritable repository
[...]
$ ./t0004-unwritable.sh --run=3 --immediate
rm: cannot remove '[...]/trash directory.t0004-unwritable/.git/objects/info': Permission denied
FATAL: Cannot prepare test area
[...]
Instead of some version of reverting [1] let's make the test-lib.sh
resilient to this edge-case, it will happen due to [1], but also
e.g. if the relevant "test-lib.sh" process is kill -9'd during the
test run. We should try harder to recover in this case. If we fail to
remove the test directory let's retry after (re-)chmod-ing it.
This doesn't need to be guarded by something that's equivalent to
"POSIXPERM" since if we don't support "chmod" we were about to fail
anyway.
Let's also discard any error output from (a possibly nonexisting)
"chmod", we'll fail on the subsequent "rm -rf" anyway, likewise for
the first "rm -rf" invocation, we don't want to get the "cannot
remove" output if we can get around it with the "chmod", but we do
want any error output from the second "rm -rf", in case that doesn't
fix the issue.
The lack of &&-chaining between the "chmod" and "rm -rf" is
intentional, if we fail the first "rm -rf", can't chmod, but then
succeed the second time around that's what we were hoping for. We just
want to nuke the directory, not carry forward every possible error
code or error message.
1. dbda967684d (t0004 (unwritable files): simplify error handling,
2010-09-06)
2. b586744a864 (test: skip clean-up when running under --immediate
mode, 2011-06-27)
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Improve the "GIT_TEST_PASSING_SANITIZE_LEAK=true" test mode added in
956d2e4639b (tests: add a test mode for SANITIZE=leak, run it in CI,
2021-09-23) to use a TAP "Bail out!" message when exiting. This will
cause the test run to exit immediately under a TAP consumer like
"prove(1)".
See 614fe015212 (test-lib: bail out when "-v" used under "prove",
2016-10-22) for the initial introduction of "Bail out!" to the
--verbose being amended here.
Before this compiling with "SANITIZE=" and running the tests with
"prove(1)" would cause all the tests to be run to the end (output
trimmed for fewer columns):
$ GIT_TEST_PASSING_SANITIZE_LEAK=true make
rm -f -r 'test-results'
*** prove ***
t0000-basic.sh ......... Dubious, test returned 1 (wstat 256, 0x100)
No subtests run
t0001-init.sh .......... Dubious, test returned 1 (wstat 256, 0x100)
No subtests run
[...output where we list every single t[0-9]*.sh file as failing snipped]
Whereas now we'll fail early, like this ("->" line wrapping added):
$ GIT_TEST_PASSING_SANITIZE_LEAK=true make
[...]
t0000-basic.sh ..................................... Bailout called. Further testing stopped:
-> GIT_TEST_PASSING_SANITIZE_LEAK=true has no effect except when compiled with SANITIZE=leak
FAILED--Further testing stopped: GIT_TEST_PASSING_SANITIZE_LEAK=true has no effect except
-> when compiled with SANITIZE=leak
make: *** [Makefile:53: prove] Error 1
This change also adds a red color to the "Bailout called" line, as
we're now using "say_color error". That improves existing output in
the case of e.g.:
$ prove -j8 t[0-9]*.sh :: -v
Bailout called. Further testing stopped: verbose mode forbidden under TAP harness; try --verbose-log
FAILED--Further testing stopped: verbose mode forbidden under TAP harness; try --verbose-log
We don't need to have a "Bail out! " prefix when we're not running
under a TAP consumer (i.e. if test -n "$HARNESS_ACTIVE"), but let's
not make the output conditional on that. Showing it under e.g.:
$ GIT_TEST_PASSING_SANITIZE_LEAK=true ./t0095-bloom.sh
Bail out! GIT_TEST_PASSING_SANITIZE_LEAK=true has no effect except when compiled with SANITIZE=leak
Doesn't harm anything, and I don't think the (small) complexity of
only adding this if we're under "$HARNESS_ACTIVE" is worth it.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
De-duplicate the "finalize_junit_xml; GIT_EXIT_OK=t; exit 1" code
shared between the "error()" and "--immediate on failure" code paths,
in preparation for adding a third user in a subsequent commit.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
A few kinds of changes "git status" can show were not documented.
* ja/doc-status-types-and-copies:
Documentation/git-status: mention how to detect copies
Documentation/git-status: document porcelain status T (typechange)
Documentation/diff-format: state in which cases porcelain status is T
Documentation/git-status: remove impossible porcelain status DR and DC
|
|
Prevent "make sparse" from running for the source files that
haven't been modified.
* ab/make-sparse-for-real:
Makefile: make the "sparse" target non-.PHONY
|
|
When "git cmd -h" shows more than one line of usage text (e.g.
the cmd subcommand may take sub-sub-command), parse-options API
learned to align these lines, even across i18n/l10n.
* ab/align-parse-options-help:
parse-options: properly align continued usage output
git rev-parse --parseopt tests: add more usagestr tests
send-pack: properly use parse_options() API for usage string
parse-options API users: align usage output in C-strings
|
|
Teach "git help -c" into helping the command line completion of
configuration variables.
* ab/help-config-vars:
help: move column config discovery to help.c library
help / completion: make "git help" do the hard work
help tests: test --config-for-completion option & output
help: simplify by moving to OPT_CMDMODE()
help: correct logic error in combining --all and --guides
help: correct logic error in combining --all and --config
help tests: add test for --config output
help: correct usage & behavior of "git help --guides"
help: correct the usage string in -h and documentation
|
|
Built-in fsmonitor (part 1).
* jh/builtin-fsmonitor-part1:
t/helper/simple-ipc: convert test-simple-ipc to use start_bg_command
run-command: create start_bg_command
simple-ipc/ipc-win32: add Windows ACL to named pipe
simple-ipc/ipc-win32: add trace2 debugging
simple-ipc: move definition of ipc_active_state outside of ifdef
simple-ipc: preparations for supporting binary messages.
trace2: add trace2_child_ready() to report on background children
|
|
Mostly preliminary clean-up in the hook API.
* ab/config-based-hooks-1:
hook-list.h: add a generated list of hooks, like config-list.h
hook.c users: use "hook_exists()" instead of "find_hook()"
hook.c: add a hook_exists() wrapper and use it in bugreport.c
hook.[ch]: move find_hook() from run-command.c to hook.c
Makefile: remove an out-of-date comment
Makefile: don't perform "mv $@+ $@" dance for $(GENERATED_H)
Makefile: stop hardcoding {command,config}-list.h
Makefile: mark "check" target as .PHONY
|
|
Updates to the tests in t0000 to test the test framework.
* ab/lib-subtest:
test-lib tests: get rid of copy/pasted mock test code
test-lib tests: assert 1 exit code, not non-zero
test-lib tests: refactor common part of check_sub_test_lib_test*()
test-lib tests: avoid subshell for "test_cmp" for readability
test-lib tests: don't provide a description for the sub-tests
test-lib tests: split up "write and run" into two functions
test-lib tests: move "run_sub_test" to a new lib-subtest.sh
|
|
Various fixes in code paths that move untracked files away to make room.
* en/removing-untracked-fixes:
Documentation: call out commands that nuke untracked files/directories
Comment important codepaths regarding nuking untracked files/dirs
unpack-trees: avoid nuking untracked dir in way of locally deleted file
unpack-trees: avoid nuking untracked dir in way of unmerged file
Change unpack_trees' 'reset' flag into an enum
Remove ignored files by default when they are in the way
unpack-trees: make dir an internal-only struct
unpack-trees: introduce preserve_ignored to unpack_trees_options
read-tree, merge-recursive: overwrite ignored files by default
checkout, read-tree: fix leak of unpack_trees_options.dir
t2500: add various tests for nuking untracked files
|
|
"git grep --recurse-submodules" takes trees and blobs from the
submodule repository, but the textconv settings when processing a
blob from the submodule is not taken from the submodule repository.
A test is added to demonstrate the issue, without fixing it.
* mt/grep-submodule-textconv:
grep: demonstrate bug with textconv attributes and submodules
|
|
"git add", "git mv", and "git rm" have been adjusted to avoid
updating paths outside of the sparse-checkout definition unless
the user specifies a "--sparse" option.
* ds/add-rm-with-sparse-index:
advice: update message to suggest '--sparse'
mv: refuse to move sparse paths
rm: skip sparse paths with missing SKIP_WORKTREE
rm: add --sparse option
add: update --renormalize to skip sparse paths
add: update --chmod to skip sparse paths
add: implement the --sparse option
add: skip tracked paths outside sparse-checkout cone
add: fail when adding an untracked sparse file
dir: fix pattern matching on dirs
dir: select directories correctly
t1092: behavior for adding sparse files
t3705: test that 'sparse_entry' is unstaged
|
|
A link to the bundle-format was added in 5c8273d57c (bundle doc: rewrite
the "DESCRIPTION" section, 2021-07-31).
Ensure `technical/bundle-format.html` is created to avoid a broken link
in `git-bundle.html`.
Signed-off-by: Todd Zullinger <tmz@pobox.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Users often use "git difftool HEAD^" to review their work, and have
"mergetool.prompt" set to false so that difftool does not prompt them
before diffing each file.
This is very convenient because users can see all their diffs by
reviewing the xxdiff windows one at a time.
A problem occurs when xxdiff encounters some binary files.
It can segfault and return exit code 128, which is special-cased
by git-difftool-helper as being an extraordinary situation that
aborts the process.
Suppress the exit code from xxdiff in its diff_cmd() implementation
when we see exit code 128 so that the GIT_EXTERNAL_DIFF loop continues
on uninterrupted to the next file rather than aborting when it
encounters the first binary file.
Signed-off-by: David Aguilar <davvid@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Fix a memory leak introduced in 9055e401dd6 (sequencer: introduce new
commands to reset the revision, 2018-04-25), which called
setup_unpack_trees_porcelain() without a corresponding call to
clear_unpack_trees_porcelain().
This introduces a change in behavior in that we now start calling
clear_unpack_trees_porcelain() even without having called the
setup_unpack_trees_porcelain(). That's OK, that clear function, like
most others, will accept a zero'd out struct.
This inches us closer to passing various tests in
"t34*.sh" (e.g. "t3434-rebase-i18n.sh"), but because they have so many
other memory leaks in revisions.c this doesn't make any test file or
even a single test pass.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
Restructure code that's mostly added in 9055e401dd6 (sequencer:
introduce new commands to reset the revision, 2018-04-25) to avoid
code duplication, and to make freeing other resources easier in a
subsequent commit.
It's safe to initialize "tree_desc" to be zero'd out in order to
unconditionally free desc.buffer, it won't be initialized on the first
couple of "goto"'s.
There are three earlier "return"'s in this function which should
probably be made to use this new "cleanup" too, per [1] it looks like
they're leaving behind stale locks. But let's not try to fix every
potential bug here now, I'm just trying to narrowly plug a memory
leak.
1. https://lore.kernel.org/git/CABPp-BH=3DP-dXRCphY53-3eZd1TU8h5GY_M12nnbEGm-UYB9Q@mail.gmail.com/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
On the Windows build agents, a lot of programs are installed, and added
to the PATH automatically.
One such program is Git for Windows, and due to the way it is set up,
unfortunately its copy of `gpg.exe` is also reachable via the PATH.
This usually does not pose any problems. To the contrary, it even allows
us to test the GPG parts of Git's test suite even if `gpg.exe` is not
delivered as part of `git-sdk-64-minimal`, the minimal subset of Git for
Windows' SDK that we use in the CI builds to compile Git.
However, every once in a while we build a new MSYS2 runtime, which means
that there is a mismatch between the copy in `git-sdk-64-minimal` and
the copy in C:\Program Files\Git\usr\bin. When that happens we hit the
dreaded problem where only one `msys-2.0.dll` is expected to be in the
PATH, and things start to fail.
Let's avoid all of this by restricting the PATH to the minimal set. This
is actually done by `git-sdk-64-minimal`'s `/etc/profile`, and we just
have to source this file manually (one would expect that it is sourced
automatically, but the Bash steps in Azure Pipelines/GitHub workflows
are explicitly run using `--noprofile`, hence the need for doing this
explicitly).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
facca53ac added verification for ssh signatures but incorrectly
described the usage of gpg.minTrustLevel. While the verifications
trustlevel is stil set to fully or undefined depending on if the key is
known or not it has no effect on the verification result. Unknown keys
will always fail verification. This commit updates the docs to match
this behaviour.
Signed-off-by: Fabian Stelzer <fs@gigacodes.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
A signature attached to a signed commit, and the contents of the
commit that merged a signed tag, are both recorded as a value of an
object header field as a multi-line value, and are subject to the
formatting convention for multi-line values in the headers, with a
leading SP signaling that the rest of the line is a continuation of
the previous line. Most notably, an empty line in such a multi-line
value would result in a line with a sole SP on it.
Examples in the signature-format technical documentation include a
few of these cases but we did not show these otherwise invisible SPs
in the example. These trailing spaces cannot be seen on display or
on paper, and forces the readers to look for them in their editors
or pagers, even if we added them to the document.
Extend the overview section to explain the multi-line value
formatting and highlight these otherwise invisible SPs by inventing
the "a dollar-sign at the end of line that appears after SP merely
signals that there is a SP there, and the dollar-sign itself does
not appear in the real file" notation, inspired by "cat -e" output,
to help readers to learn exactly where such "a single SP that is
originally an empty line" appears in the examples.
Reported-by: Rob Browning <rlb@defaultvalue.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
|
When parsing a URL to normalize it, we allow hostnames to contain only
dot (".") or dash ("-"), plus brackets and colons for IPv6 literals.
This matches the old URL standard in RFC 1738, which says:
host = hostname | hostnumber
hostname = *[ domainlabel "." ] toplabel
domainlabel = alphadigit | alphadigit *[ alphadigit | "-" ] alphadigit
But this was later updated by RFC 3986, which is more liberal:
host = IP-literal / IPv4address / reg-name
reg-name = *( unreserved / pct-encoded / sub-delims )
unreserved = ALPHA / DIGIT / "-" / "." / "_" / "~"
While names with underscore in them are not common and possibly violate
some DNS rules, they do work in practice, and we will happily contact
them over http://, git://, or ssh://. It seems odd to ignore them for
purposes of URL matching, especially when the URL RFC seems to allow
them.
There shouldn't be any downside here. It's not a syntactically
significant character in a URL, so we won't be confused about parsing;
we'd have simply rejected such a URL previously (the test here checks
the url code directly, but the obvious user-visible effect would be
failing to match credential.http://foo_bar.example.com.helper, or
similar config in http.<url>.*).
Arguably we'd want to allow tilde ("~") here, too. There's likewise
probably no downside, but I didn't add it simply because it seems like
an even less likely character to appear in a hostname.
Reported-by: Alex Waite <alex@waite.eu>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
|
