Due to scalability issues, Shawn Pearce has originally proposed a new
"reftable" format more than six years ago [1]. Initially, this new
format was implemented in JGit with promising results. Around two years
ago, we have then added the "reftable" library to the Git codebase via
a4bbd13be3 (Merge branch 'hn/reftable', 2021-12-15). With this we have
landed all the low-level code to read and write reftables. Notably
missing though was the integration of this low-level code into the Git
code base in the form of a new ref backend that ties all of this
together.
This gap is now finally closed by introducing a new "reftable" backend
into the Git codebase. This new backend promises to bring some notable
improvements to Git repositories:
- It becomes possible to do truly atomic writes where either all refs
are committed to disk or none are. This was not possible with the
"files" backend because ref updates were split across multiple loose
files.
- The disk space required to store many refs is reduced, both compared
to loose refs and packed-refs. This is enabled both by the reftable
format being a binary format, which is more compact, and by prefix
compression.
- We can ignore filesystem-specific behaviour as ref names are not
encoded via paths anymore. This means there is no need to handle
case sensitivity on Windows systems or Unicode precomposition on
macOS.
- There is no need to rewrite the complete refdb anymore every time a
ref is being deleted like it was the case for packed-refs. This
means that ref deletions are now constant time instead of scaling
linearly with the number of refs.
- We can ignore file/directory conflicts so that it becomes possible
to store both "refs/heads/foo" and "refs/heads/foo/bar".
- Due to this property we can retain reflogs for deleted refs. We have
previously been deleting reflogs together with their refs to avoid
file/directory conflicts, which is not necessary anymore.
- We can properly enumerate all refs. With the "files" backend it is
not easily possible to distinguish between refs and non-refs because
they may live side by side in the gitdir.
Not all of these improvements are realized with the current "reftable"
backend implementation. At this point, the new backend is supposed to be
a drop-in replacement for the "files" backend that is used by basically
all Git repositories nowadays. It strives for 1:1 compatibility, which
means that a user can expect the same behaviour regardless of whether
they use the "reftable" backend or the "files" backend for most of the
part.
Most notably, this means we artificially limit the capabilities of the
"reftable" backend to match the limits of the "files" backend. It is not
possible to create refs that would end up with file/directory conflicts,
we do not retain reflogs, we perform stricter-than-necessary checks.
This is done intentionally due to two main reasons:
- It makes it significantly easier to land the "reftable" backend as
tests behave the same. It would be tough to argue for each and every
single test that doesn't pass with the "reftable" backend.
- It ensures compatibility between repositories that use the "files"
backend and repositories that use the "reftable" backend. Like this,
hosters can migrate their repositories to use the "reftable" backend
without causing issues for clients that use the "files" backend in
their clones.
It is expected that these artificial limitations may eventually go away
in the long term.
Performance-wise things very much depend on the actual workload. The
following benchmarks compare the "files" and "reftable" backends in the
current version:
- Creating N refs in separate transactions shows that the "files"
backend is ~50% faster. This is not surprising given that creating a
ref only requires us to create a single loose ref. The "reftable"
backend will also perform auto compaction on updates. In real-world
workloads we would likely also want to perform pack loose refs,
which would likely change the picture.
Benchmark 1: update-ref: create refs sequentially (refformat = files, refcount = 1)
Time (mean ± σ): 2.1 ms ± 0.3 ms [User: 0.6 ms, System: 1.7 ms]
Range (min … max): 1.8 ms … 4.3 ms 133 runs
Benchmark 2: update-ref: create refs sequentially (refformat = reftable, refcount = 1)
Time (mean ± σ): 2.7 ms ± 0.1 ms [User: 0.6 ms, System: 2.2 ms]
Range (min … max): 2.4 ms … 2.9 ms 132 runs
Benchmark 3: update-ref: create refs sequentially (refformat = files, refcount = 1000)
Time (mean ± σ): 1.975 s ± 0.006 s [User: 0.437 s, System: 1.535 s]
Range (min … max): 1.969 s … 1.980 s 3 runs
Benchmark 4: update-ref: create refs sequentially (refformat = reftable, refcount = 1000)
Time (mean ± σ): 2.611 s ± 0.013 s [User: 0.782 s, System: 1.825 s]
Range (min … max): 2.597 s … 2.622 s 3 runs
Benchmark 5: update-ref: create refs sequentially (refformat = files, refcount = 100000)
Time (mean ± σ): 198.442 s ± 0.241 s [User: 43.051 s, System: 155.250 s]
Range (min … max): 198.189 s … 198.670 s 3 runs
Benchmark 6: update-ref: create refs sequentially (refformat = reftable, refcount = 100000)
Time (mean ± σ): 294.509 s ± 4.269 s [User: 104.046 s, System: 190.326 s]
Range (min … max): 290.223 s … 298.761 s 3 runs
- Creating N refs in a single transaction shows that the "files"
backend is significantly slower once we start to write many refs.
The "reftable" backend only needs to update two files, whereas the
"files" backend needs to write one file per ref.
Benchmark 1: update-ref: create many refs (refformat = files, refcount = 1)
Time (mean ± σ): 1.9 ms ± 0.1 ms [User: 0.4 ms, System: 1.4 ms]
Range (min … max): 1.8 ms … 2.6 ms 151 runs
Benchmark 2: update-ref: create many refs (refformat = reftable, refcount = 1)
Time (mean ± σ): 2.5 ms ± 0.1 ms [User: 0.7 ms, System: 1.7 ms]
Range (min … max): 2.4 ms … 3.4 ms 148 runs
Benchmark 3: update-ref: create many refs (refformat = files, refcount = 1000)
Time (mean ± σ): 152.5 ms ± 5.2 ms [User: 19.1 ms, System: 133.1 ms]
Range (min … max): 148.5 ms … 167.8 ms 15 runs
Benchmark 4: update-ref: create many refs (refformat = reftable, refcount = 1000)
Time (mean ± σ): 58.0 ms ± 2.5 ms [User: 28.4 ms, System: 29.4 ms]
Range (min … max): 56.3 ms … 72.9 ms 40 runs
Benchmark 5: update-ref: create many refs (refformat = files, refcount = 1000000)
Time (mean ± σ): 152.752 s ± 0.710 s [User: 20.315 s, System: 131.310 s]
Range (min … max): 152.165 s … 153.542 s 3 runs
Benchmark 6: update-ref: create many refs (refformat = reftable, refcount = 1000000)
Time (mean ± σ): 51.912 s ± 0.127 s [User: 26.483 s, System: 25.424 s]
Range (min … max): 51.769 s … 52.012 s 3 runs
- Deleting a ref in a fully-packed repository shows that the "files"
backend scales with the number of refs. The "reftable" backend has
constant-time deletions.
Benchmark 1: update-ref: delete ref (refformat = files, refcount = 1)
Time (mean ± σ): 1.7 ms ± 0.1 ms [User: 0.4 ms, System: 1.2 ms]
Range (min … max): 1.6 ms … 2.1 ms 316 runs
Benchmark 2: update-ref: delete ref (refformat = reftable, refcount = 1)
Time (mean ± σ): 1.8 ms ± 0.1 ms [User: 0.4 ms, System: 1.3 ms]
Range (min … max): 1.7 ms … 2.1 ms 294 runs
Benchmark 3: update-ref: delete ref (refformat = files, refcount = 1000)
Time (mean ± σ): 2.0 ms ± 0.1 ms [User: 0.5 ms, System: 1.4 ms]
Range (min … max): 1.9 ms … 2.5 ms 287 runs
Benchmark 4: update-ref: delete ref (refformat = reftable, refcount = 1000)
Time (mean ± σ): 1.9 ms ± 0.1 ms [User: 0.5 ms, System: 1.3 ms]
Range (min … max): 1.8 ms … 2.1 ms 217 runs
Benchmark 5: update-ref: delete ref (refformat = files, refcount = 1000000)
Time (mean ± σ): 229.8 ms ± 7.9 ms [User: 182.6 ms, System: 46.8 ms]
Range (min … max): 224.6 ms … 245.2 ms 6 runs
Benchmark 6: update-ref: delete ref (refformat = reftable, refcount = 1000000)
Time (mean ± σ): 2.0 ms ± 0.0 ms [User: 0.6 ms, System: 1.3 ms]
Range (min … max): 2.0 ms … 2.1 ms 3 runs
- Listing all refs shows no significant advantage for either of the
backends. The "files" backend is a bit faster, but not by a
significant margin. When repositories are not packed the "reftable"
backend outperforms the "files" backend because the "reftable"
backend performs auto-compaction.
Benchmark 1: show-ref: print all refs (refformat = files, refcount = 1, packed = true)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.0 ms 1729 runs
Benchmark 2: show-ref: print all refs (refformat = reftable, refcount = 1, packed = true)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 1.8 ms 1816 runs
Benchmark 3: show-ref: print all refs (refformat = files, refcount = 1000, packed = true)
Time (mean ± σ): 4.3 ms ± 0.1 ms [User: 0.9 ms, System: 3.3 ms]
Range (min … max): 4.1 ms … 4.6 ms 645 runs
Benchmark 4: show-ref: print all refs (refformat = reftable, refcount = 1000, packed = true)
Time (mean ± σ): 4.5 ms ± 0.2 ms [User: 1.0 ms, System: 3.3 ms]
Range (min … max): 4.2 ms … 5.9 ms 643 runs
Benchmark 5: show-ref: print all refs (refformat = files, refcount = 1000000, packed = true)
Time (mean ± σ): 2.537 s ± 0.034 s [User: 0.488 s, System: 2.048 s]
Range (min … max): 2.511 s … 2.627 s 10 runs
Benchmark 6: show-ref: print all refs (refformat = reftable, refcount = 1000000, packed = true)
Time (mean ± σ): 2.712 s ± 0.017 s [User: 0.653 s, System: 2.059 s]
Range (min … max): 2.692 s … 2.752 s 10 runs
Benchmark 7: show-ref: print all refs (refformat = files, refcount = 1, packed = false)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 1.9 ms 1834 runs
Benchmark 8: show-ref: print all refs (refformat = reftable, refcount = 1, packed = false)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.4 ms … 2.0 ms 1840 runs
Benchmark 9: show-ref: print all refs (refformat = files, refcount = 1000, packed = false)
Time (mean ± σ): 13.8 ms ± 0.2 ms [User: 2.8 ms, System: 10.8 ms]
Range (min … max): 13.3 ms … 14.5 ms 208 runs
Benchmark 10: show-ref: print all refs (refformat = reftable, refcount = 1000, packed = false)
Time (mean ± σ): 4.5 ms ± 0.2 ms [User: 1.2 ms, System: 3.3 ms]
Range (min … max): 4.3 ms … 6.2 ms 624 runs
Benchmark 11: show-ref: print all refs (refformat = files, refcount = 1000000, packed = false)
Time (mean ± σ): 12.127 s ± 0.129 s [User: 2.675 s, System: 9.451 s]
Range (min … max): 11.965 s … 12.370 s 10 runs
Benchmark 12: show-ref: print all refs (refformat = reftable, refcount = 1000000, packed = false)
Time (mean ± σ): 2.799 s ± 0.022 s [User: 0.735 s, System: 2.063 s]
Range (min … max): 2.769 s … 2.836 s 10 runs
- Printing a single ref shows no real difference between the "files"
and "reftable" backends.
Benchmark 1: show-ref: print single ref (refformat = files, refcount = 1)
Time (mean ± σ): 1.5 ms ± 0.1 ms [User: 0.4 ms, System: 1.0 ms]
Range (min … max): 1.4 ms … 1.8 ms 1779 runs
Benchmark 2: show-ref: print single ref (refformat = reftable, refcount = 1)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.4 ms … 2.5 ms 1753 runs
Benchmark 3: show-ref: print single ref (refformat = files, refcount = 1000)
Time (mean ± σ): 1.5 ms ± 0.1 ms [User: 0.3 ms, System: 1.1 ms]
Range (min … max): 1.4 ms … 1.9 ms 1840 runs
Benchmark 4: show-ref: print single ref (refformat = reftable, refcount = 1000)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.0 ms 1831 runs
Benchmark 5: show-ref: print single ref (refformat = files, refcount = 1000000)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.1 ms 1848 runs
Benchmark 6: show-ref: print single ref (refformat = reftable, refcount = 1000000)
Time (mean ± σ): 1.6 ms ± 0.1 ms [User: 0.4 ms, System: 1.1 ms]
Range (min … max): 1.5 ms … 2.1 ms 1762 runs
So overall, performance depends on the usecases. Except for many
sequential writes the "reftable" backend is roughly on par or
significantly faster than the "files" backend though. Given that the
"files" backend has received 18 years of optimizations by now this can
be seen as a win. Furthermore, we can expect that the "reftable" backend
will grow faster over time when attention turns more towards
optimizations.
The complete test suite passes, except for those tests explicitly marked
to require the REFFILES prerequisite. Some tests in t0610 are marked as
failing because they depend on still-in-flight bug fixes. Tests can be
run with the new backend by setting the GIT_TEST_DEFAULT_REF_FORMAT
environment variable to "reftable".
There is a single known conceptual incompatibility with the dumb HTTP
transport. As "info/refs" SHOULD NOT contain the HEAD reference, and
because the "HEAD" file is not valid anymore, it is impossible for the
remote client to figure out the default branch without changing the
protocol. This shortcoming needs to be handled in a subsequent patch
series.
As the reftable library has already been introduced a while ago, this
commit message will not go into the details of how exactly the on-disk
format works. Please refer to our preexisting technical documentation at
Documentation/technical/reftable for this.
[1]: https://public-inbox.org/git/CAJo=hJtyof=HRy=2sLP0ng0uZ4=S-DpZ5dR1aF+VHVETKG20OQ@mail.gmail.com/
Original-idea-by: Shawn Pearce <spearce@spearce.org>
Based-on-patch-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
TEST_PASSES_SANITIZE_LEAK must be set before sourcing test-lib.sh, as we
say in t/README:
GIT_TEST_PASSING_SANITIZE_LEAK=true skips those tests that haven't
declared themselves as leak-free by setting
"TEST_PASSES_SANITIZE_LEAK=true" before sourcing "test-lib.sh". This
test mode is used by the "linux-leaks" CI target.
GIT_TEST_PASSING_SANITIZE_LEAK=check checks that our
"TEST_PASSES_SANITIZE_LEAK=true" markings are current. Rather than
skipping those tests that haven't set "TEST_PASSES_SANITIZE_LEAK=true"
before sourcing "test-lib.sh" this mode runs them with
"--invert-exit-code". This is used to check that there's a one-to-one
mapping between "TEST_PASSES_SANITIZE_LEAK=true" and those tests that
pass under "SANITIZE=leak". This is especially useful when testing a
series that fixes various memory leaks with "git rebase -x".
In a recent commit we fixed a test where it was set after sourcing
test-lib.sh, leading to confusing results.
To prevent future oversights, let's add a simple check to ensure the
value for TEST_PASSES_SANITIZE_LEAK remains unchanged at test_done().
Signed-off-by: Rubén Justo <rjusto@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Introduce a new GIT_TEST_DEFAULT_REF_FORMAT environment variable that
lets developers run the test suite with a different default ref format
without impacting the ref format used by non-test Git invocations. This
is modeled after GIT_TEST_DEFAULT_OBJECT_FORMAT, which does the same
thing for the repository's object format.
Adapt the setup of the `REFFILES` test prerequisite to be conditionally
set based on the default ref format.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Introduce a new "extensions.refStorage" extension that allows us to
specify the ref storage format used by a repository. For now, the only
supported format is the "files" format, but this list will likely soon
be extended to also support the upcoming "reftable" format.
There have been discussions on the Git mailing list in the past around
how exactly this extension should look like. One alternative [1] that
was discussed was whether it would make sense to model the extension in
such a way that backends are arbitrarily stackable. This would allow for
a combined value of e.g. "loose,packed-refs" or "loose,reftable", which
indicates that new refs would be written via "loose" files backend and
compressed into "packed-refs" or "reftable" backends, respectively.
It is arguable though whether this flexibility and the complexity that
it brings with it is really required for now. It is not foreseeable that
there will be a proliferation of backends in the near-term future, and
the current set of existing formats and formats which are on the horizon
can easily be configured with the much simpler proposal where we have a
single value, only.
Furthermore, if we ever see that we indeed want to gain the ability to
arbitrarily stack the ref formats, then we can adapt the current
extension rather easily. Given that Git clients will refuse any unknown
value for the "extensions.refStorage" extension they would also know to
ignore a stacked "loose,packed-refs" in the future.
So let's stick with the easy proposal for the time being and wire up the
extension.
[1]: <pull.1408.git.1667846164.gitgitgadget@gmail.com>
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A limited number of tests require repositories to have the default
repository format or otherwise they would fail to run, e.g. because they
fail to detect the correct hash function. While the hash function is the
only extension right now that creates problems like this, we are about
to add a second extension for the ref format.
Introduce a new DEFAULT_REPO_FORMAT prereq that can easily be amended
whenever we add new format extensions. Next to making any such changes
easier on us, the prerequisite's name should also help to clarify the
intent better.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Stale URLs have been updated to their current counterparts (or
archive.org) and HTTP links are replaced with working HTTPS links.
* js/update-urls-in-doc-and-comment:
doc: refer to internet archive
doc: update links for andre-simon.de
doc: switch links to https
doc: update links to current pages
These sites offer https versions of their content.
Using the https versions provides some protection for users.
Signed-off-by: Josh Soref <jsoref@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
UBSAN options were not propagated through the test framework to git
run via the httpd, unlike ASAN options, which has been corrected.
* jk/test-pass-ubsan-options-to-http-test:
test-lib: set UBSAN_OPTIONS to match ASan
For a long time we have used ASAN_OPTIONS to set abort_on_error. This is
important because we want to notice detected problems even in programs
which are expected to fail. But we never did the same for UBSAN_OPTIONS.
This means that our UBSan test suite runs might silently miss some
cases.
It also causes a more visible effect, which is that t4058 complains
about unexpected "fixes" (and this is how I noticed the issue):
$ make SANITIZE=undefined CC=gcc && (cd t && ./t4058-*)
...
ok 8 - git read-tree does not segfault # TODO known breakage vanished
ok 9 - reset --hard does not segfault # TODO known breakage vanished
ok 10 - git diff HEAD does not segfault # TODO known breakage vanished
The tests themselves aren't that interesting. We have a known bug where
these programs segfault, and they do when compiled without sanitizers.
With UBSan, when the test runs:
test_might_fail git read-tree --reset base
it gets:
cache-tree.c:935:9: runtime error: member access within misaligned address 0x5a5a5a5a5a5a5a5a for type 'struct cache_entry', which requires 8 byte alignment
So that's garbage memory which would _usually_ cause us to segfault, but
UBSan catches it and complains first about the alignment. That makes
sense, but the weird thing is that UBSan then exits instead of aborting,
so our test_might_fail call considers that an acceptable outcome and the
test "passes".
Curiously, this historically seems to have aborted, because I've run
"make test" with UBSan many times (and so did our CI) and we never saw
the problem. Even more curiously, I see an abort if I use clang with
ASan and UBSan together, like:
# this aborts!
make SANITIZE=undefined,address CC=clang
But not with just UBSan, and not with both when used with gcc:
# none of these do
make SANITIZE=undefined CC=gcc
make SANITIZE=undefined CC=clang
make SANITIZE=undefined,address CC=gcc
Likewise moving to older versions of gcc (I tried gcc-11 and gcc-12 on
my Debian system) doesn't abort. Nor does moving around in Git's
history. Neither this test nor the relevant code have been touched in a
while, and going back to v2.41.0 produces the same outcome (even though
many UBSan CI runs have passed in the meantime).
So _something_ changed on my system (and likely will soon on other
people's, since this is stock Debian unstable), but I didn't track
it further. I don't know why it ever aborted in the past, but we
definitely should be explicit here and tell UBSan what we want to
happen.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Tests with LSan from time to time seem to emit harmless message
that makes our tests unnecessarily flakey; we work it around by
filtering the uninteresting output.
* jk/test-lsan-denoise-output:
test-lib: ignore uninteresting LSan output
When I run the tests in leak-checking mode the same way our CI job does,
like:
make SANITIZE=leak \
GIT_TEST_PASSING_SANITIZE_LEAK=true \
GIT_TEST_SANITIZE_LEAK_LOG=true \
test
then LSan can racily produce useless entries in the log files that look
like this:
==git==3034393==Unable to get registers from thread 3034307.
I think they're mostly harmless based on the source here:
7e0a52e8e9/compiler-rt/lib/lsan/lsan_common.cpp (L414)
which reads:
PtraceRegistersStatus have_registers =
suspended_threads.GetRegistersAndSP(i, ®isters, &sp);
if (have_registers != REGISTERS_AVAILABLE) {
Report("Unable to get registers from thread %llu.\n", os_id);
// If unable to get SP, consider the entire stack to be reachable unless
// GetRegistersAndSP failed with ESRCH.
if (have_registers == REGISTERS_UNAVAILABLE_FATAL)
continue;
sp = stack_begin;
}
The program itself still runs fine and LSan doesn't cause us to abort.
But test-lib.sh looks for any non-empty LSan logs and marks the test as
a failure anyway, under the assumption that we simply missed the failing
exit code somehow.
I don't think I've ever seen this happen in the CI job, but running
locally using clang-14 on an 8-core machine, I can't seem to make it
through a full run of the test suite without having at least one
failure. And it's a different one every time (though they do seem to
often be related to packing tests, which makes sense, since that is one
of our biggest users of threaded code).
We can hack around this by only counting LSan log files that contain a
line that doesn't match our known-uninteresting pattern.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To check for broken &&-chains, we run "fail_117 && $1" as a test
snippet, and check the exit code. We use test_eval_ to do so, because
that's the way we run the actual test.
But we don't need any of its niceties, like "set -x" tracing. In fact,
they hinder us, because we have to explicitly disable them. So let's
skip that and use "eval" more directly, which is simpler. I had hoped it
would also be faster, but it doesn't seem to produce a measurable
improvement (probably because it's just running internal shell commands,
with no subshells or forks).
Note that there is one gotcha: even though we don't intend to run any of
the commands if the &&-chain is intact, an error like this:
test_expect_success 'broken' '
# this next line breaks the &&-chain
true
# and then this one is executed even by the linter
return 1
'
means we'll "return 1" from the eval, and thus from test_run_(). We
actually do notice this in test_expect_success, but only by saying "hey,
this test didn't say it was OK, so it must have failed", which is not
right (it should say "broken &&-chain").
We can handle this by calling test_eval_inner_() instead, which is our
trick for wrapping "return" in a test snippet. But to do that, we have
to push the trace code out of that inner function and into test_eval_().
This is arguably where it belonged in the first place, but it never
mattered because the "inner_" function had only one caller.
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit 99a64e4b73 (tests: lint for run-away here-doc, 2017-03-22)
tweaked the chain-lint test to catch unclosed here-docs. It works by
adding an extra "echo" command after the test snippet, and checking that
it is run (if it gets swallowed by a here-doc, naturally it is not run).
The downside here is that we introduced an extra $() substitution, which
happens in a subshell. This has a measurable performance impact when
run for many tests.
The tradeoff in safety was undoubtedly worth it when 99a64e4b73 was
written. But since the external chainlint.pl learned to find these
recently, we can just rely on it. By switching back to a simpler
chain-lint, hyperfine reports a measurable speedup on t3070 (which has
1800 tests):
'HEAD' ran
1.12 ± 0.01 times faster than 'HEAD~1'
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
To test that we don't break the &&-chain, test-lib.sh does something
like:
(exit 117) && $test_commands
and checks that the result is exit code 117. We don't care what that
initial command is, as long as it exits with a unique code. Using "exit"
works and is simple, but is a bit expensive since it requires a subshell
(to avoid exiting the whole script!). This isn't usually very
noticeable, but it can add up for scripts which have a large number of
tests.
Using "return" naively won't work here, because we'd return from the
function eval-ing the snippet (and it wouldn't find &&-chain breakages).
But if we further push that into its own function, it does exactly what
we want, without extra subshell overhead.
According to hyperfine, this produces a measurable improvement when
running t3070 (which has 1800 tests, all of them quite short):
'HEAD' ran
1.09 ± 0.01 times faster than 'HEAD~1'
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 69b9924b87 (t/Makefile: teach `make test` and `make prove` to run
chainlint.pl, 2022-09-01), we run a single chainlint.pl process for all
scripts, and then instruct each individual script to run with the
equivalent of --no-chain-lint, which tells them not to redundantly run
the chainlint script themselves.
However, this also disables the internal linter run within the shell by
eval-ing "(exit 117) && $1" and confirming we get code 117. In theory
the external linter produces a superset of complaints, and we don't need
the internal one anymore. However, we know there is at least one case
where they differ. A test like:
test_expect_success 'should fail linter' '
false &&
sleep 2 &
pid=$! &&
kill $pid
'
is buggy (it ignores the failure from "false", because it is
backgrounded along with the sleep). The internal linter catches this,
but the external one doesn't (and teaching it to do so is complicated[1]).
So not only does "make test" miss this problem, but it's doubly
confusing because running the script standalone does complain.
Let's teach the suppression in the Makefile to only turn off the
external linter (which we know is redundant, as it was already run) and
leave the internal one intact.
I've used a new environment variable to do this here, and intentionally
did not add a "--no-ext-chain-lint" option. This is an internal
optimization used by the Makefile, and not something that ordinary users
would need to tweak.
[1] For discussion of chainlint.pl and this case, see:
https://lore.kernel.org/git/CAPig+cQtLFX4PgXyyK_AAkCvg4Aw2RAC5MmLbib-aHHgTBcDuw@mail.gmail.com/
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When a comment describing how each test file should start was added in
commit [1], it was the second comment of t/test-lib.sh. The comment
describes how variable "test_description" is supposed to be assigned at
the top of each test file and how "test-lib.sh" should be used by
sourcing it. However, even in [1], the comment was ten lines away from
the usage of the variable by test-lib.sh. Since then, the comment has
drifted away both from the top of the file and from the usage of the
variable. The comment just sits in the middle of the initialization of
the test library, surrounded by unrelated code, almost one hundred lines
away from the usage of "test_description".
Nobody has noticed this drift during evolution of test-lib.sh, which
suggests that this comment has outlived its usefulness. The assignment
of "test_description", sourcing of "test-lib.sh" by tests, and the
process of writing tests in general are described in detail in
"t/README". So drop the obsolete comment.
An alternative solution could be to move the comment either to the top
of the file, or down to the usage of variable "test_description".
[1] e1970ce43a ("[PATCH 1/2] Test framework take two.", 2005-05-13)
Signed-off-by: Andrei Rybak <rybak.a.v@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since [1] first released with Git v2.37.0 the built-in version of "add
-i" has been the default. That built-in implementation was added in
[2], first released with Git v2.25.0.
At this point enough time has passed to allow for finding any
remaining bugs in this new implementation, so let's remove the
fallback code.
As with similar migrations for "stash"[3] and "rebase"[4] we're
keeping a mention of "add.interactive.useBuiltin" in the
documentation, but adding a warning() to notify any outstanding users
that the built-in is now the default. As with [5] and [6] we should
follow-up in the future and eventually remove that warning.
1. 0527ccb1b5 (add -i: default to the built-in implementation,
2021-11-30)
2. f83dff60a7 (Start to implement a built-in version of `git add
--interactive`, 2019-11-13)
3. 8a2cd3f512 (stash: remove the stash.useBuiltin setting,
2020-03-03)
4. d03ebd411c (rebase: remove the rebase.useBuiltin setting,
2019-03-18)
5. deeaf5ee07 (stash: remove documentation for `stash.useBuiltin`,
2022-01-27)
6. 9bcde4d531 (rebase: remove transitory rebase.useBuiltin setting &
env, 2021-03-23)
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since [1] there has been no reason for keeping "git env--helper" a
built-in. The reason it was a built-in to begin with was to support
the GIT_TEST_GETTEXT_POISON mode removed in that commit. I.e. unlike
the rest of "test-tool" it would potentially be called by the
installed git via "git-sh-i18n.sh".
As none of that applies since [1] we should stop carrying this
technical debt, and move it to t/helper/*. As this mostly move-only
change shows this has the nice bonus that we'll stop wasting time
translating the internal-only strings it emits.
Even though this was a built-in, it was intentionally never
documented, see its introduction in [2]. It never saw use outside of
the test suite, except for the "GIT_TEST_GETTEXT_POISON" use-case
noted above.
1. d162b25f95 (tests: remove support for GIT_TEST_GETTEXT_POISON,
2021-01-20)
2. b4f207f339 (env--helper: new undocumented builtin wrapping
git_env_*(), 2019-06-21)
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Acked-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Use `git diff --no-index` as a test_cmp on Windows.
We'd probably need to revisit "do we really want to, and have to,
lose CRLF vs LF?" later, at which time we may be able to further
clean this up by replacing "git diff --no-index" with "diff -u".
* js/drop-mingw-test-cmp:
tests(mingw): avoid very slow `mingw_test_cmp`
When Git's test suite uses `test_cmp`, it is not actually trying to
compare binary files as the name `cmp` would suggest to users familiar
with Unix' tools, but the tests instead verify that actual output
matches the expected text.
On Unix, `cmp` works well enough for Git's purposes because only Line
Feed characters are used as line endings. However, on Windows, while
most tools accept Line Feeds as line endings, many tools produce
Carriage Return + Line Feed line endings, including some of the tools
used by the test suite (which are therefore provided via Git for Windows
SDK). Therefore, `cmp` would frequently fail merely due to different
line endings.
To accommodate for that, the `mingw_test_cmp` function was introduced
into Git's test suite to perform a line-by-line comparison that ignores
line endings. This function is a Bash function that is only used on
Windows, everywhere else `cmp` is used.
This is a double whammy because `cmp` is fast, and `mingw_test_cmp` is
slow, even more so on Windows because it is a Bash script function, and
Bash scripts are known to run particularly slowly on Windows due to
Bash's need for the POSIX emulation layer provided by the MSYS2 runtime.
The commit message of 32ed3314c1 (t5351: avoid using `test_cmp` for
binary data, 2022-07-29) provides an illuminating account of the
consequences: On Windows, the platform on which Git could really use all
the help it can get to improve its performance, the time spent on one
entire test script was reduced from half an hour to less than half a
minute merely by avoiding a single call to `mingw_test_cmp` in but a
single test case.
Learning the lesson to avoid shell scripting wherever possible, the Git
for Windows project implemented a minimal replacement for
`mingw_test_cmp` in the form of a `test-tool` subcommand that parses the
input files line by line, ignoring line endings, and compares them.
Essentially the same thing as `mingw_test_cmp`, but implemented in
C instead of Bash. This solution served the Git for Windows project
well, over years.
However, when this solution was finally upstreamed, the conclusion was
reached that a change to use `git diff --no-index` instead of
`mingw_test_cmp` was more easily reviewed and hence should be used
instead.
The reason why this approach was not even considered in Git for Windows
is that in 2007, there was already a motion on the table to use Git's
own diff machinery to perform comparisons in Git's test suite, but it
was dismissed in https://lore.kernel.org/git/xmqqbkrpo9or.fsf@gitster.g/
as undesirable because tests might potentially succeed due to bugs in
the diff machinery when they should not succeed, and those bugs could
therefore hide regressions that the tests try to prevent.
By the time Git for Windows' `mingw-test-cmp` in C was finally
contributed to the Git mailing list, reviewers agreed that the diff
machinery had matured enough and should be used instead.
When the concern was raised that the diff machinery, due to its
complexity, would perform substantially worse than the test helper
originally implemented in the Git for Windows project, a test
demonstrated that these performance differences are well lost within the
100+ minutes it takes to run Git's test suite on Windows.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update to build procedure with VS using CMake/CTest.
* js/cmake-updates:
cmake: increase time-out for a long-running test
cmake: avoid editing t/test-lib.sh
add -p: avoid ambiguous signed/unsigned comparison
cmake: copy the merge tools for testing
cmake: make it easier to diagnose regressions in CTest runs
In 7f5397a07c (cmake: support for testing git when building out of the
source tree, 2020-06-26), we implemented support for running Git's test
scripts even after building Git in a different directory than the source
directory.
The way we did this was to edit the file `t/test-lib.sh` to override
`GIT_BUILD_DIR` to point somewhere else than the parent of the `t/`
directory.
This is unideal because it always leaves a tracked file marked as
modified, and it is all too easy to commit that change by mistake.
Let's change the strategy by teaching `t/test-lib.sh` to detect the
presence of a file called `GIT-BUILD-DIR` in the source directory. If it
exists, the contents are interpreted as the location to the _actual_
build directory. We then write this file as part of the CTest
definition.
To support building Git via a regular `make` invocation after building
it using CMake, we ensure that the `GIT-BUILD-DIR` file is deleted (for
convenience, this is done as part of the Makefile rule that is already
run with every `make` invocation to ensure that `GIT-BUILD-OPTIONS` is
up to date).
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 131b94a10a (test-lib.sh: Use GLIBC_TUNABLES instead of
MALLOC_CHECK_ on glibc >= 2.34, 2022-03-04) compiling with
SANITIZE=leak has missed reporting some leaks. The old MALLOC_CHECK
method used before glibc 2.34 seems to have been (mostly?) compatible
with it, but after 131b94a10a e.g. running:
TEST_NO_MALLOC_CHECK=1 make SANITIZE=leak test T=t6437-submodule-merge.sh
Would report a leak in builtin/commit.c, but this would not:
TEST_NO_MALLOC_CHECK= make SANITIZE=leak test T=t6437-submodule-merge.sh
Since the interaction is clearly breaking the SANITIZE=leak mode,
let's mark them as explicitly incompatible.
A related regression for SANITIZE=address was fixed in
067109a5e7 (tests: make SANITIZE=address imply TEST_NO_MALLOC_CHECK,
2022-04-09).
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Acked-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
By automatically invoking chainlint.sed upon each test it runs,
`test_run_` in test-lib.sh ensures that broken &&-chains will be
detected early as tests are modified or new are tests created since it
is typical to run a test script manually (i.e. `./t1234-test-script.sh`)
during test development. Now that the implementation of chainlint.pl is
complete, modify test-lib.sh to invoke it automatically instead of
chainlint.sed each time a test script is run.
This change reduces the number of "linter" invocations from 26800+ (once
per test run) down to 1050+ (once per test script), however, a
subsequent change will drop the number of invocations to 1 per `make
test`, thus fully realizing the benefit of the new linter.
Note that the "magic exit code 117" &&-chain checker added by bb79af9d09
(t/test-lib: introduce --chain-lint option, 2015-03-20) which is built
into t/test-lib.sh is retained since it has near zero-cost and
(theoretically) may catch a broken &&-chain not caught by chainlint.pl.
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
`test_run_` in test-lib.sh "lints" the body of a test by sending it down
a `sed chainlint.sed | grep` pipeline; this happens once for each test
run by a test script. Although this pipeline may seem relatively cheap
in isolation, it can become expensive when invoked 26800+ times by `make
test`, once for each test run, despite the existence of only 16500+ test
definitions across all tests scripts.
This difference in the number of tests defined in the scripts (16500+)
and the number of tests actually run by `make test` (26800+) is
explained by the fact that some test scripts run a very large number of
small tests, all driven by a series of functions/loops which fill in the
test bodies. This means that certain test definitions are being linted
repeatedly (tens or hundreds of times) unnecessarily. To avoid such
unnecessary work, 2d86a96220 (t: avoid sed-based chain-linting in some
expensive cases, 2021-05-13) added an optimization hack which allows
individual scripts to manually suppress the unnecessary repeated linting
of the same test definition.
However, unlike chainlint.sed which checks a test body as the test is
run, chainlint.pl checks each test definition just once, no matter how
many times the test is run, thus the sort of optimization hack
introduced by 2d86a96220 is no longer needed and can be retired.
Therefore, revert 2d86a96220.
Signed-off-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Avoid repeatedly running getconf to ask libc version in the test
suite, and instead just as it once per script.
* pw/use-glibc-tunable-for-malloc-optim:
tests: cache glibc version check
131b94a10a ("test-lib.sh: Use GLIBC_TUNABLES instead of MALLOC_CHECK_
on glibc >= 2.34", 2022-03-04) introduced a check for the version of
glibc that is in use. This check is performed as part of
setup_malloc_check() which is called at least once for each test. As
the test involves forking `getconf` and `expr` cache the result and
use that within setup_malloc_check() to avoid forking these extra
processes for each test.
Signed-off-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
As noted in previous on-list discussions[1] we have various tests that
will falsely report being leak-free because we're missing the relevant
exit code from LSAN as summarized below.
We should fix those issues, but in the meantime and as an additional
sanity check we can and should consider our own ASAN logs before
reporting that a test is leak-free.
Before this compiling with SANITIZE=leak and running:
./t6407-merge-binary.sh
Will exit successfully, now we'll get an error and an informative
message on:
GIT_TEST_SANITIZE_LEAK_LOG=true ./t6407-merge-binary.sh
Even better, as noted in the updated t/README we'll now error out when
combined with the "check" mode:
GIT_TEST_PASSING_SANITIZE_LEAK=check \
GIT_TEST_SANITIZE_LEAK_LOG=true \
./t4058-diff-duplicates.sh
Why do we miss these leaks? Because:
* We have leaks inside "test_expect_failure" blocks, which by design
will not distinguish a "normal" failure from an abort() or
segfault. See [1] for a discussion of it shortcomings.
* We have "git" invocations outside of "test_expect_success",
e.g. setup code in the main body of the test, or in test helper
functions that don't use &&-chaining.
* Our tests will otherwise catch segfaults and abort(), but if we
invoke a command that invokes another command it needs to ferry the
exit code up to us.
Notably a command that e.g. might invoke "git pack-objects" might
itself exit with status 128 if that "pack-objects" segfaults or
abort()'s. If the test invoking the parent command(s) is using
"test_must_fail" we'll consider it an expected "ok" failure.
* run-command.c doesn't (but probably should) ferry up such exit
codes, so for e.g. "git push" tests where we expect a failure and an
underlying "git" command fails we won't ferry up the segfault or
abort exit code.
* We have gitweb.perl and some other perl code ignoring return values
from close(), i.e. ignoring exit codes from "git rev-parse" et al.
* We have in-tree shellscripts like "git-merge-one-file.sh" invoking
git commands, they'll usually return their own exit codes on "git"
failure, rather then ferrying up segfault or abort() exit code.
E.g. these invocations in git-merge-one-file.sh leak, but aren't
reflected in the "git merge" exit code:
src1=$(git unpack-file $2)
src2=$(git unpack-file $3)
That case would be easily "fixed" by adding a line like this after
each assignment:
test $? -ne 0 && exit $?
But we'd then in e.g. "t6407-merge-binary.sh" run into
write_tree_trivial() in "builtin/merge.c" calling die() instead of
ferrying up the relevant exit code.
Let's remove "TEST_PASSES_SANITIZE_LEAK=true" from tests we
were falsely marking as leak-free.
In the case of t6407-merge-binary.sh it was marked as leak-free in
9081a421a6 (checkout: fix "branch info" memory leaks,
2021-11-16). I'd previously removed other bad
"TEST_PASSES_SANITIZE_LEAK=true" opt-ins in the series merged in
ea05fd5fbf (Merge branch 'ab/keep-git-exit-codes-in-tests',
2022-03-16). The case of t1060-object-corruption.sh is more subtle,
and will be discussed in a subsequent commit.
1. https://lore.kernel.org/git/cover-0.7-00000000000-20220318T002951Z-avarab@gmail.com/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add a new "GIT_TEST_PASSING_SANITIZE_LEAK=check" mode to the
test-lib.sh.
As noted in the updated "t/README" this compliments the existing
"GIT_TEST_PASSING_SANITIZE_LEAK=true" mode added in
956d2e4639 (tests: add a test mode for SANITIZE=leak, run it in CI,
2021-09-23).
Rather than document this all in one (even more) dense paragraph split
up the discussion of how it combines with --immediate into its own
paragraph following the discussion of
"GIT_TEST_SANITIZE_LEAK_LOG=true".
Before the removal of "test_external" in a preceding commit we would
have had to special-case t9700-perl-git.sh and t0202-gettext-perl.sh.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Remove the "test_external" function added in [1]. This arguably makes
the output of t9700-perl-git.sh and friends worse. But as we'll argue
below the trade-off is worth it, since "chaining" to another TAP
emitter in test-lib.sh is more trouble than it's worth.
The new output of t9700-perl-git.sh is now:
$ ./t9700-perl-git.sh
ok 1 - set up test repository
ok 2 - use t9700/test.pl to test Git.pm
# passed all 2 test(s)
1..2
Whereas before this change it would be:
$ ./t9700-perl-git.sh
ok 1 - set up test repository
# run 1: Perl API (perl /home/avar/g/git/t/t9700/test.pl)
ok 2 - use Git;
[... omitting tests 3..46 from t/t9700/test.pl ...]
ok 47 - unquote escape sequences
1..47
# test_external test Perl API was ok
# test_external_without_stderr test no stderr: Perl API was ok
At the time of its addition supporting "test_external" was easy, but
when test-lib.sh itself started to emit TAP in [2] we needed to make
everything surrounding the emission of the plan consider
"test_external". I added that support in [2] so that we could run:
prove ./t9700-perl-git.sh :: -v
But since then in [3] the door has been closed on combining
$HARNESS_ACTIVE and -v, we'll now just die:
$ prove ./t9700-perl-git.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
So the only use of this has been that *if* we had failure in one of
these tests we could e.g. in CI see which test failed based on the
test number. Now we'll need to look at the full verbose logs to get
that same information.
I think this trade-off is acceptable given the reduction in
complexity, and it brings these tests in line with other similar
tests, e.g. the reftable tests added in [4] will be condensed down to
just one test, which invokes the C helper:
$ ./t0032-reftable-unittest.sh
ok 1 - unittests
# passed all 1 test(s)
1..1
It would still be nice to have that ":: -v" form work again, it
never *really* worked, but even though we've had edge cases test
output screwing up the TAP it mostly worked between d998bd4ab6 and
[3], so we may have been overzealous in forbidding it outright.
I have local patches which I'm planning to submit sooner than later
that get us to that goal, and in a way that isn't buggy. In the
meantime getting rid of this special case makes hacking on this area
of test-lib.sh easier, as we'll do in subsequent commits.
The switch from "perl" to "$PERL_PATH" here is because "perl" is
defined as a shell function in the test suite, see a5bf824f3b (t:
prevent '-x' tracing from interfering with test helpers' stderr,
2018-02-25). On e.g. the OSX CI the "command perl"... will be part of
the emitted stderr.
1. fb32c41008 (t/test-lib.sh: add test_external and
test_external_without_stderr, 2008-06-19)
2. d998bd4ab6 (test-lib: Make the test_external_* functions
TAP-aware, 2010-06-24)
3. 614fe01521 (test-lib: bail out when "-v" used under
"prove", 2016-10-22)
4. ef8a6c6268 (reftable: utility functions, 2021-10-07)
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add the ability to run the test suite under a new
"GIT_TEST_SANITIZE_LEAK_LOG=true" mode, when true we'll log the leaks
we find an a new "test-results/<test-name>.leak" directory.
That new path is consistent with the existing
"test-results/<test-name>.<type>" results, except that those are all
files, not directories.
We also set "log_exe_name=1" to include the name of the executable in
the filename. This gives us files like "trace.git.<pid>" instead of
the default of "trace.<pid>". I.e. we'll be able to distinguish "git"
leaks from "test-tool", "git-daemon" etc.
We then set "dedup_token_length" to non-zero ("0" is the default) to
succinctly log a token we can de-duplicate these stacktraces on. The
string is simply a one-line stack-trace with only function names up to
N frames, which we limit at "9999" as a shorthand for
"infinite" (there appears to be no way to say "no limit").
With these combined we can now easily get e.g. the top 10 leaks in the
test suite grouped by full stacktrace:
grep -o -P -h '(?<=DEDUP_TOKEN: ).*' test-results/*.leak/trace.git.* | sort | uniq -c | sort -nr | head -n 10
Or add "grep -E -o '[^-]+'" to that to group by functions instead of
stack traces:
grep -o -P -h '(?<=DEDUP_TOKEN: ).*' test-results/*.leak/trace.git.* | grep -E -o '[^-]+' | sort | uniq -c | sort -nr | head -n 20
This new mode requires git to be compiled with SANITIZE=leak, rather
than explaining that in the documentation let's make it
self-documenting by bailing out if the user asks for this without git
having been compiled with SANITIZE=leak, as we do with
GIT_TEST_PASSING_SANITIZE_LEAK=true.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Reword the documentation added in 956d2e4639 (tests: add a test mode
for SANITIZE=leak, run it in CI, 2021-09-23) for brevity.
The comment added in the same commit was also misleading: We skip
certain tests if SANITIZE=leak and GIT_TEST_PASSING_SANITIZE_LEAK=true,
not if we're compiled with SANITIZE=leak. Let's just remove the
comment, the control flow here is obvious enough that the code can
speak for itself.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add the ability to have those tests that fail return 0, and those
tests that succeed return 1. This is useful e.g. to run "--stress"
tests on tests that fail 99% of the time on some setup, i.e. to smoke
out the flaky run which yielded success.
In a subsequent commit a new SANITIZE=leak mode will make use of this.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change various "exit 1" checks that happened after our "die" handler
had been set up to use BAIL_OUT instead. See 234383cd40 (test-lib.sh:
use "Bail out!" syntax on bad SANITIZE=leak use, 2021-10-14) for the
benefits of the BAIL_OUT function.
The previous use of "error" here was not a logic error, but the "exit"
without "GIT_EXIT_OK" would emit the "FATAL: Unexpected exit with code
$code" message on top of the error we wanted to emit.
Since we'd also like to stop "prove" in its tracks here, the right
thing to do is to emit a "Bail out!" message.
Let's also move the "GIT_EXIT_OK=t" assignments to just above the
"exit [01]" in "test_done". It's not OK if we exit in
e.g. finalize_test_output.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the control flow in test_done so that we'll set GIT_EXIT_OK=t
after we call test_atexit_handler(). This seems to have been a mistake
in 900721e15c (test-lib: introduce 'test_atexit', 2019-03-13). It
doesn't make sense to allow our "atexit" handling to call "exit"
without us emitting the errors we'll emit without GIT_EXIT_OK=t being
set.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Clarify that these two functions never take N arguments, they'll only
ever receive one. They've needlessly used $@ over $1 since
41ac414ea2 (Sane use of test_expect_failure, 2008-02-01).
In the future we might want to pass the test source to these, but now
that's not the case. This preparatory change helps to clarify a
follow-up change.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fixes for tests when the source directory has unusual characters in
its path, e.g. whitespaces, double-quotes, etc.
source: <cover-v2-0.3-00000000000-20220630T101646Z-avarab@gmail.com>
* ab/test-quoting-fix:
config tests: fix harmless but broken "rm -r" cleanup
test-lib.sh: fix prepend_var() quoting issue
tests: add missing double quotes to included library paths
Tweak tests so that they still work when the "git init" template
did not create .git/info directory.
* ab/test-without-templates:
tests: don't assume a .git/info for .git/info/sparse-checkout
tests: don't assume a .git/info for .git/info/exclude
tests: don't assume a .git/info for .git/info/refs
tests: don't assume a .git/info for .git/info/attributes
tests: don't assume a .git/info for .git/info/grafts
tests: don't depend on template-created .git/branches
t0008: don't rely on default ".git/info/exclude"
Fixes for tests when the source directory has unusual characters in
its path, e.g. whitespaces, double-quotes, etc.
* ab/test-quoting-fix:
config tests: fix harmless but broken "rm -r" cleanup
test-lib.sh: fix prepend_var() quoting issue
tests: add missing double quotes to included library paths
Fix a quoting issue in the function introduced in
b9638d7286 (test-lib: make $GIT_BUILD_DIR an absolute path,
2022-02-27), running the test suite where the git checkout was on a
path with e.g. a space in it would fail.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add and use a LIBCURL prerequisite for tests added in
6dcbdc0d66 (remote: create fetch.credentialsInUrl config,
2022-06-06).
These tests would get as far as emitting a couple of the warnings we
were testing for, but would then die as we had no "git-remote-https"
program compiled.
It would be more consistent with other prerequisites (e.g. PERL for
NO_PERL) to name this "CURL", but since e9184b0789 (t5561: skip tests
if curl is not available, 2018-04-03) we've had that prerequisite
defined for checking of we have the curl(1) program.
The existing "CURL" prerequisite is only used in one place, and we
should probably name it "CURL_PROGRAM", then rename "LIBCURL" to
"CURL" as a follow-up, but for now (pre-v2.37.0) let's aim for the
most minimal fix possible.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix an issue that existed before 0527ccb1b5 (add -i: default to the
built-in implementation, 2021-11-30), but which became the default
with that change, we should not be marking tests that are known to
pass as "TODO" tests.
When GIT_TEST_ADD_I_USE_BUILTIN=1 was made the default we started
passing the tests added in 0f0fba2cc8 (t3701: add a test for advanced
split-hunk editing, 2019-12-06) and 1bf01040f0 (add -p: demonstrate
failure when running 'edit' after a split, 2015-04-16).
Thus we've been emitting this sort of output:
$ prove ./t3701-add-interactive.sh
./t3701-add-interactive.sh .. ok
All tests successful.
Test Summary Report
-------------------
./t3701-add-interactive.sh (Wstat: 0 Tests: 70 Failed: 0)
TODO passed: 45, 47
Files=1, Tests=70, 2 wallclock secs ( 0.03 usr 0.00 sys + 0.86 cusr 0.33 csys = 1.22 CPU)
Result: PASS
Which isn't just cosmetic, but due to issues with
test_expect_failure (see [1]) we could e.g. be hiding something as bad
as a segfault in the new implementation. It makes sense catch that,
especially before we put out a release with the built-in "add -i", so
let's generalize the check we were already doing in 0527ccb1b5 with a
new "ADD_I_USE_BUILTIN" prerequisite.
1. https://lore.kernel.org/git/patch-1.7-4624abc2591-20220318T002951Z-avarab@gmail.com/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Update the GitHub workflow support to make it quicker to get to the
failing test.
* js/ci-github-workflow-markup:
ci: call `finalize_test_case_output` a little later
ci(github): mention where the full logs can be found
ci: use `--github-workflow-markup` in the GitHub workflow
ci(github): avoid printing test case preamble twice
ci(github): skip the logs of the successful test cases
ci: optionally mark up output in the GitHub workflow
ci/run-build-and-tests: add some structure to the GitHub workflow output
ci: make it easier to find failed tests' logs in the GitHub workflow
ci/run-build-and-tests: take a more high-level view
test(junit): avoid line feeds in XML attributes
tests: refactor --write-junit-xml code
ci: fix code style
As noted in c8a58ac5a5 (Revert "Don't create the $GIT_DIR/branches
directory on init", 2009-10-31) there was an attempt long ago in
0cc5691a8b (Don't create the $GIT_DIR/branches directory on init,
2009-10-30) to get rid of the legacy "branches" directory.
We should probably get rid of its creation by removing the
"templates/branches--" file. But whatever our default behavior, our
tests should be tightened up to explicitly create the .git/branches
directory if they rely on our default templates, to make the
dependency on those templates clear.
So let's amend the two tests that would fail if .git/branches wasn't
created. To do this introduce a new "TEST_CREATE_REPO_NO_TEMPLATE"
variable, which we'll set before sourcing test-lib.sh, and change the
"git clone" and "git init" commands in the tests themselves to
explicitly pass "--template=".
This way they won't get a .git/branches in either their top-level
.git, or in the ones they create. We can then amend the tests that
rely on the ".git/branches" directory existing to create it
explicitly, and to remove it after its creation.
This new "TEST_CREATE_REPO_NO_TEMPLATE" variable is a less
heavy-handed version of the "NO_SET_GIT_TEMPLATE_DIR" variable. See
a94d305bf8 (t/t0001-init.sh: add test for 'init with init.templatedir
set', 2010-02-26) for its implementation.
Unlike "TEST_CREATE_REPO_NO_TEMPLATE", this new
"TEST_CREATE_REPO_NO_TEMPLATE" variable is narrowly scoped to what the
"git init" in test-lib.sh does, as opposed to the global effect of
"NO_SET_GIT_TEMPLATE_DIR" and the setting of "GIT_TEMPLATE_DIR" in
wrap-for-bin.sh.
I experimented with adding a new "GIT_WRAP_FOR_BIN_VIA_TEST_LIB"
variable set in test-lib.sh, which would cause wrap-for-bin.sh to not
set GIT_TEMPLATE_DIR, GITPERLLIB etc, as we set those in
test-lib.sh. I think that's a viable approach, but it would interact
e.g. with the appending feature of GITPERLLIB added in
8bade1e12e (wrap-for-bin: make bin-wrappers chainable, 2013-07-04).
Doing so would allow us to convert the tests in t0001-init.sh that now
use "NO_SET_GIT_TEMPLATE_DIR" to simply unset "GIT_TEMPLATE_DIR" in a
sub-shell before invoking "git init" or "git clone". I think that
approach is worth pursuing, but let's table it for now. Some future
wrap-for-bin.sh refactoring can try to address it.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A bit of test framework fixes with a few fixes to issues found by
valgrind.
* ab/valgrind-fixes:
commit-graph.c: don't assume that stat() succeeds
object-file: fix a unpack_loose_header() regression in 3b6a8db3b0
log test: skip a failing mkstemp() test under valgrind
tests: using custom GIT_EXEC_PATH breaks --valgrind tests
We used to call that function already before printing the final verdict.
However, now that we added grouping to the GitHub workflow output, we
will want to include even that part in the collapsible group for that
test case.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
