By doing ls -1 .git/{reftable,refs/heads}, we can capture changes to both
reftable and packed/loose ref storage.
This relies on the fact that git-pack-refs (which we're looking for here)
changes the number (loose/packed storage) and/or names (reftable) files used for
ref storage.
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The previous code tested this by writing $ZERO_OID explicitly in the packed-refs
file. This is a type of corruption that doesn't reflect realistic use-cases. In
addition, even the ref-store test-tool refuses to write invalid OIDs.
(update-ref interprets $ZERO_OID is deleting the ref).
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This test takes a lock on the target of a symref, and then verifies that it is
possible to expire the symref's reflog. In reftable, one can only take a global
lock (which would prevent the symref reflog from being expired altogether.)
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The tests verifies that "pack-refs" causes loose refs to be packed. As both
loose and packed refs are concepts specific to the files backend, mark the test
as REFFILES.
Check the outcome of the pack-refs operation. This was apparently forgotten in
the commit introducing this test: 16feb99d (Mar 26 2017, "t1405: some basic
tests on main ref store").
Signed-off-by: Han-Wen Nienhuys <hanwen@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
With a54e938e5b (strbuf: support long paths w/o read rights in
strbuf_getcwd() on FreeBSD, 2017-03-26) we had t0001 break on systems
like OpenBSD and AIX whose getcwd(3) has standard (but not like glibc
et al) behavior.
This was partially fixed in bed67874e2 (t0001: skip test with
restrictive permissions if getpwd(3) respects them, 2017-08-07).
The problem with that fix is that while its analysis of the problem is
correct, it doesn't actually call getcwd(3), instead it invokes "pwd
-P". There is no guarantee that "pwd -P" is going to call getcwd(3),
as opposed to e.g. being a shell built-in.
On AIX under both bash and ksh this test breaks because "pwd -P" will
happily display the current working directory, but getcwd(3) called by
the "git init" we're testing here will fail to get it.
I checked whether clobbering the $PWD environment variable would
affect it, and it didn't. Presumably these shells keep track of their
working directory internally.
There's possible follow-up work here in teaching strbuf_getcwd() to
get the working directory with whatever method "pwd" uses on these
platforms. See [1] for a discussion of that, but let's take the easy
way out here and just skip these tests by fixing the
GETCWD_IGNORES_PERMS prerequisite to match the limitations of
strbuf_getcwd().
1. https://lore.kernel.org/git/b650bef5-d739-d98d-e9f1-fa292b6ce982@web.de/
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since b243012 (refresh_index(): add flag to ignore SKIP_WORKTREE
entries, 2021-04-08), 'git add --refresh <path>' will output a warning
message when the path is outside the sparse-checkout definition. The
implementation of this warning happened in parallel with the
sparse-index work to add ensure_full_index() calls throughout the
codebase.
Update this loop to have the proper logic that checks to see if the
pathspec is outside the sparse-checkout definition. This avoids the need
to expand the sparse directory entry and determine if the path is
tracked, untracked, or ignored. We simply avoid updating the stat()
information because there isn't even an entry that matches the path!
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The add_pathspec_matches_against_index() focuses on matching a pathspec
to file entries in the index. This already works correctly for its only
use: checking if untracked files exist in the index.
The compatibility checks in t1092 already test that 'git add <dir>'
works for a directory outside of the sparse cone. That provides coverage
for removing this guard.
This finalizes our ability to run 'git add .' without expanding a sparse
index to a full one. This is evidenced by an update to t1092 and by
these performance numbers for p2000-sparse-operations.sh:
Test HEAD~1 HEAD
--------------------------------------------------------------------------------
2000.10: git add . (full-index-v3) 0.37(0.28+0.07) 0.36(0.27+0.06) -2.7%
2000.11: git add . (full-index-v4) 0.33(0.26+0.06) 0.32(0.28+0.05) -3.0%
2000.12: git add . (sparse-index-v3) 0.57(0.53+0.07) 0.06(0.06+0.07) -89.5%
2000.13: git add . (sparse-index-v4) 0.57(0.53+0.07) 0.05(0.03+0.09) -91.2%
While the ~90% improvement is shown by the test results, it is worth
noting that expanding the sparse index was adding overhead in previous
commits. Comparing to the full index case, we see the performance go
from 0.33s to 0.05s, an 85% improvement.
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Disable command_requires_full_index for 'git add'. This does not require
any additional removals of ensure_full_index(). The main reason is that
'git add' discovers changes based on the pathspec and the worktree
itself. These are then inserted into the index directly, and calls to
index_name_pos() or index_file_exists() already call expand_to_path() at
the appropriate time to support a sparse-index.
Add a test to check that 'git add -A' and 'git add <file>' does not
expand the index at all, as long as <file> is not within a sparse
directory. This does not help the global 'git add .' case.
We can measure the improvement using p2000-sparse-operations.sh with
these results:
Test HEAD~1 HEAD
------------------------------------------------------------------------------
2000.6: git add -A (full-index-v3) 0.35(0.30+0.05) 0.37(0.29+0.06) +5.7%
2000.7: git add -A (full-index-v4) 0.31(0.26+0.06) 0.33(0.27+0.06) +6.5%
2000.8: git add -A (sparse-index-v3) 0.57(0.53+0.07) 0.05(0.04+0.08) -91.2%
2000.9: git add -A (sparse-index-v4) 0.58(0.55+0.06) 0.05(0.05+0.06) -91.4%
While the 91% improvement seems impressive, it's important to recognize
that previously we had significant overhead for expanding the
sparse-index. Comparing to the full index case, 'git add -A' goes from
0.37s to 0.05s, which is "only" an 86% improvement.
This modification to 'git add' creates some behavior change depending on
the use of a sparse index. We modify a test in t1092 to demonstrate
these changes which will be remedied in future changes.
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Conflicts can occur outside of the sparse-checkout definition, and in
that case users might try to resolve the conflicts in several ways.
Document a few of these ways in a test. Make it clear that this behavior
is not necessarily the optimal flow, since users can become confused
when Git deletes these files from the worktree in later commands.
Reviewed-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"TEST_OUTPUT_DIRECTORY=there make test" failed to work, which has
been corrected.
* ps/t0000-output-directory-fix:
t0000: fix test if run with TEST_OUTPUT_DIRECTORY
The code that gives an error message in "git multi-pack-index" when
no subcommand is given tried to print a NULL pointer as a strong,
which has been corrected.
* tb/reverse-midx:
multi-pack-index: fix potential segfault without sub-command
The completion support used to offer alternate spelling of options
that exist only for compatibility, which has been corrected.
* pb/dont-complete-aliased-options:
parse-options: don't complete option aliases by default
"git status" codepath learned to work with sparsely populated index
without hydrating it fully.
* ds/status-with-sparse-index:
t1092: document bad sparse-checkout behavior
fsmonitor: integrate with sparse index
wt-status: expand added sparse directory entries
status: use sparse-index throughout
status: skip sparse-checkout percentage with sparse-index
diff-lib: handle index diffs with sparse dirs
dir.c: accept a directory as part of cone-mode patterns
unpack-trees: unpack sparse directory entries
unpack-trees: rename unpack_nondirectories()
unpack-trees: compare sparse directories correctly
unpack-trees: preserve cache_bottom
t1092: add tests for status/add and sparse files
t1092: expand repository data shape
t1092: replace incorrect 'echo' with 'cat'
sparse-index: include EXTENDED flag when expanding
sparse-index: skip indexes with unmerged entries
Tests that cover protocol bits have been updated and helpers
used there have been consolidated.
* ab/pkt-line-tests:
test-lib-functions: use test-tool for [de]packetize()
Many "printf"-like helper functions we have have been annotated
with __attribute__() to catch placeholder/parameter mismatches.
* ab/attribute-format:
advice.h: add missing __attribute__((format)) & fix usage
*.h: add a few missing __attribute__((format))
*.c static functions: add missing __attribute__((format))
sequencer.c: move static function to avoid forward decl
*.c static functions: don't forward-declare __attribute__
Optimize "git log" for cases where we wasted cycles to load ref
decoration data that may not be needed.
* jk/log-decorate-optim:
load_ref_decorations(): fix decoration with tags
add_ref_decoration(): rename s/type/deco_type/
load_ref_decorations(): avoid parsing non-tag objects
object.h: add lookup_object_by_type() function
object.h: expand docstring for lookup_unknown_object()
log: avoid loading decorations for userformats that don't need it
pretty.h: update and expand docstring for userformat_find_requirements()
"git worktree add --lock" learned to record why the worktree is
locked with a custom message.
* sm/worktree-add-lock:
worktree: teach `add` to accept --reason <string> with --lock
worktree: mark lock strings with `_()` for translation
t2400: clean up '"add" worktree with lock' test
Optimization for repositories with many alternate object store.
* ew/many-alternate-optim:
oidtree: a crit-bit tree for odb_loose_cache
oidcpy_with_padding: constify `src' arg
make object_directory.loose_objects_subdir_seen a bitmap
avoid strlen via strbuf_addstr in link_alt_odb_entry
speed up alt_odb_usable() with many alternates
"git commit --allow-empty-message" won't abort the operation upon
an empty message, but the hint shown in the editor said otherwise.
* hj/commit-allow-empty-message:
commit: remove irrelavent prompt on `--allow-empty-message`
commit: reorganise commit hint strings
Ever since Git learned to detect its install location at runtime, there
was the slightly awkward problem that it was impossible to specify paths
relative to said location.
For example, if a version of Git was shipped with custom SSL
certificates to use, there was no portable way to specify
`http.sslCAInfo`.
In Git for Windows, the problem was "solved" for years by interpreting
paths starting with a slash as relative to the runtime prefix.
However, this is not correct: such paths _are_ legal on Windows, and
they are interpreted as absolute paths in the same drive as the current
directory.
After a lengthy discussion, and an even lengthier time to mull over the
problem and its best solution, and then more discussions, we eventually
decided to introduce support for the magic sequence `%(prefix)/`. If a
path starts with this, the remainder is interpreted as relative to the
detected (runtime) prefix. If built without runtime prefix support, Git
will simply interpolate the compiled-in prefix.
If a user _wants_ to specify a path starting with the magic sequence,
they can prefix the magic sequence with `./` and voilà, the path won't
be expanded.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Originally, we refrained from adding a regression test in 7b6c6496374
(system_path(): Add prefix computation at runtime if RUNTIME_PREFIX set,
2008-08-10), and in 226c0ddd0d (exec_cmd: RUNTIME_PREFIX on some POSIX
systems, 2018-04-10).
The reason was that it was deemed too tricky to test.
Turns out that it is not tricky to test at all: we simply create a
pseudo-root, copy the `git` executable into the `git/` subdirectory of
that pseudo-root, then copy a script into the `libexec/git-core/`
directory and expect that to be picked up.
As long as the trash directory is in a location where binaries can be
executed, this works.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
%(rest) is a atom used for cat-file batch mode, which can split
the input lines at the first whitespace boundary, all characters
before that whitespace are considered to be the object name;
characters after that first run of whitespace (i.e., the "rest"
of the line) are output in place of the %(rest) atom.
In order to let "cat-file --batch=%(rest)" use the ref-filter
interface, add %(rest) atom for ref-filter.
Introduce the reject_atom() to reject the atom %(rest) for
"git for-each-ref", "git branch", "git tag" and "git verify-tag".
Reviewed-by: Jacob Keller <jacob.keller@gmail.com>
Suggected-by: Jacob Keller <jacob.keller@gmail.com>
Mentored-by: Christian Couder <christian.couder@gmail.com>
Mentored-by: Hariom Verma <hariom18599@gmail.com>
Signed-off-by: ZheNing Hu <adlternative@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Because the perl language can handle binary data correctly,
add the function perl_quote_buf_with_len(), which can specify
the length of the data and prevent the data from being truncated
at '\0' to help `--format="%(raw)"` support `--perl`.
Reviewed-by: Jacob Keller <jacob.keller@gmail.com>
Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: ZheNing Hu <adlternative@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add new formatting option `%(raw)`, which will print the raw
object data without any changes. It will help further to migrate
all cat-file formatting logic from cat-file to ref-filter.
The raw data of blob, tree objects may contain '\0', but most of
the logic in `ref-filter` depends on the output of the atom being
text (specifically, no embedded NULs in it).
E.g. `quote_formatting()` use `strbuf_addstr()` or `*._quote_buf()`
add the data to the buffer. The raw data of a tree object is
`100644 one\0...`, only the `100644 one` will be added to the buffer,
which is incorrect.
Therefore, we need to find a way to record the length of the
atom_value's member `s`. Although strbuf can already record the
string and its length, if we want to replace the type of atom_value's
member `s` with strbuf, many places in ref-filter that are filled
with dynamically allocated mermory in `v->s` are not easy to replace.
At the same time, we need to check if `v->s == NULL` in
populate_value(), and strbuf cannot easily distinguish NULL and empty
strings, but c-style "const char *" can do it. So add a new member in
`struct atom_value`: `s_size`, which can record raw object size, it
can help us add raw object data to the buffer or compare two buffers
which contain raw object data.
Note that `--format=%(raw)` cannot be used with `--python`, `--shell`,
`--tcl`, and `--perl` because if the binary raw data is passed to a
variable in such languages, these may not support arbitrary binary data
in their string variable type.
Reviewed-by: Jacob Keller <jacob.keller@gmail.com>
Mentored-by: Christian Couder <christian.couder@gmail.com>
Mentored-by: Hariom Verma <hariom18599@gmail.com>
Helped-by: Bagas Sanjaya <bagasdotme@gmail.com>
Helped-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Helped-by: Felipe Contreras <felipe.contreras@gmail.com>
Helped-by: Phillip Wood <phillip.wood@dunelm.org.uk>
Helped-by: Junio C Hamano <gitster@pobox.com>
Based-on-patch-by: Olga Telezhnaya <olyatelezhnaya@gmail.com>
Signed-off-by: ZheNing Hu <adlternative@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 'git merge' learned '--autostash' in a03b55530a (merge: teach
--autostash option, 2020-04-07), 'cmd_merge', once it is determined that
we have to create a merge commit, calls 'create_autostash' if
'--autostash' is given.
As explained in a03b55530a, and made more abvious by the tests added in
that commit, the autostash is then applied if the merge succeeds, either
directly or by committing (after conflict resolution or if '--no-commit'
was given), or if the merge is aborted with 'git merge --abort'. In some
other cases, like the user calling 'git reset --merge' or 'git merge
--quit', the autostash is not applied, but saved in the stash list.
However, there exists a scenario that creates an autostash but does not
apply nor save it to the stash list: if the chosen merge strategy
completely fails to handle the merge, i.e. 'try_merge_strategy' returns
2.
Apply the autostash in that case also. An easy way to test that is to
try to merge more than two commits but explicitely ask for the 'recursive'
merge strategy.
Signed-off-by: Philippe Blain <levraiphilippeblain@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since 'git merge' learned '--autostash' in a03b55530a (merge: teach
--autostash option, 2020-04-07), 'cmd_merge', in the fast-forward case,
calls 'create_autostash' before calling 'checkout_fast_forward' if
'--autostash' is given.
However, if 'checkout_fast_forward' fails, the autostash is not applied
to the working tree, nor saved in the stash list, since the code simply
calls 'goto done'.
Be more helpful to the user by applying the autostash in that case.
An easy way to test a failing fast-forward is when we are merging a
branch that has a tracked file that conflicts with an untracked file in
the working tree.
Signed-off-by: Philippe Blain <levraiphilippeblain@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git read-tree" checks the existence of the blobs referenced by the
given tree, but does not bulk prefetch them. Add a bulk prefetch.
The lack of prefetch here was noticed at $DAYJOB during a merge
involving some specific commits, but I couldn't find a minimal merge
that didn't also trigger the prefetch in check_updates() in
unpack-trees.c (and in all these cases, the lack of prefetch in
cache-tree.c didn't matter because all the relevant blobs would have
already been prefetched by then). This is why I used read-tree here to
exercise this code path.
Signed-off-by: Jonathan Tan <jonathantanmy@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It can be useful to tell who invoked Git - was it invoked manually by a
user via CLI or script? By an IDE? In some cases - like 'repo' tool -
we can influence the source code and set the GIT_TRACE2_PARENT_SID
environment variable from the caller process. In 'repo''s case, that
parent SID is manipulated to include the string "repo", which means we
can positively identify when Git was invoked by 'repo' tool. However,
identifying parents that way requires both that we know which tools
invoke Git and that we have the ability to modify the source code of
those tools. It cannot scale to keep up with the various IDEs and
wrappers which use Git, most of which we don't know about. Learning
which tools and wrappers invoke Git, and how, would give us insight to
decide where to improve Git's usability and performance.
Unfortunately, there's no cross-platform reliable way to gather the name
of the parent process. If procfs is present, we can use that; otherwise
we will need to discover the name another way. However, the process ID
should be sufficient to look up the process name on most platforms, so
that code may be shareable.
Git for Windows gathers similar information and logs it as a "data_json"
event. However, since "data_json" has a variable format, it is difficult
to parse effectively in some languages; instead, let's pursue a
dedicated "cmd_ancestry" event to record information about the ancestry
of the current process and a consistent, parseable way.
Git for Windows also gathers information about more than one generation
of parent. In Linux further ancestry info can be gathered with procfs,
but it's unwieldy to do so. In the interest of later moving Git for
Windows ancestry logging to the 'cmd_ancestry' event, and in the
interest of later adding more ancestry to the Linux implementation - or
of adding this functionality to other platforms which have an easier
time walking the process tree - let's make 'cmd_ancestry' accept an
array of parentage.
Signed-off-by: Emily Shaffer <emilyshaffer@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the bundle tests to fully compare the expected "git ls-remote"
or "git bundle list-heads" output, instead of merely grepping it.
This avoids subtle regressions in the tests. In
f62e0a39b6 (t5704 (bundle): add tests for bundle --stdin, 2010-04-19)
the "bundle --stdin <rev-list options>" test was added to make sure we
didn't include the tag.
But since the --stdin mode didn't work until 5bb0fd2cab (bundle:
arguments can be read from stdin, 2021-01-11) our grepping of
"master" (later "main") missed the important part of the test.
Namely that we should not include the "refs/tags/tag" tag in that
case. Since the test only grepped for "main" in the output we'd miss a
regression in that code.
So let's use test_cmp instead, and also in the other nearby tests
where it's easy.
This does make things a bit more verbose in the case of the test
that's checking the bundle header, since it's different under SHA1 and
SHA256. I think this makes test easier to follow.
I've got some WIP changes to extend the "git bundle" command to dump
parts of the header out, which are easier to understand if we test the
output explicitly like this.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change uses of ":" on the LHS of a ">" to the more commonly used
">file" pattern in t/t5607-clone-bundle.sh.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"git rev-list" learns to omit the "commit <object-name>" header
lines from the output with the `--no-commit-header` option.
* bc/rev-list-without-commit-line:
rev-list: add option for --pretty=format without header
"git send-email" optimization.
* ab/send-email-optim:
perl: nano-optimize by replacing Cwd::cwd() with Cwd::getcwd()
send-email: move trivial config handling to Perl
perl: lazily load some common Git.pm setup code
send-email: lazily load modules for a big speedup
send-email: get rid of indirect object syntax
send-email: use function syntax instead of barewords
send-email: lazily shell out to "git var"
send-email: lazily load config for a big speedup
send-email: copy "config_regxp" into git-send-email.perl
send-email: refactor sendemail.smtpencryption config parsing
send-email: remove non-working support for "sendemail.smtpssl"
send-email tests: test for boolean variables without a value
send-email tests: support GIT_TEST_PERL_FATAL_WARNINGS=true
With multiple heads, we should not allow rebasing or fast-forwarding.
Make sure any fast-forward request calls out specifically the fact that
multiple branches are in play. Also, since we cannot fast-forward to
multiple branches, fix our computation of can_ff.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We have for some time shown a long warning when the user does not
specify how to reconcile divergent branches with git pull. Make it an
error now.
Initial-patch-by: Alex Henrie <alexhenrie24@gmail.com>
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix the last few precedence tests failing in t7601 by now implementing
the logic to have --[no-]rebase override a pull.ff=only config setting.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The warning about pulling without specifying how to reconcile divergent
branches says that after setting pull.rebase to true, --ff-only can
still be passed on the command line to require a fast-forward. Make that
actually work.
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
[en: updated tests; note 3 fixes and 1 new failure]
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
There were already code checking that --rebase was incompatible with
a merge of multiple heads. However, we were sometimes throwing warnings
about lack of specification of rebase vs. merge when given multiple
heads. Since rebasing is disallowed with multiple merge heads, that
seems like a poor warning to print; we should instead just assume
merging is wanted.
Add a few tests checking multiple merge head behavior.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The interaction of rebase and merge flags and options was not well
tested. Add several tests to check for correct behavior from the
following rules:
* --ff-only vs. --[no-]rebase
(and the related pull.ff=only vs. pull.rebase)
* --rebase[=!false] vs. --no-ff and --ff
(and the related pull.rebase=!false overrides pull.ff=!only)
* command line flags take precedence over config, except:
* --no-rebase heeds pull.ff=!only
* pull.rebase=!false vs --no-ff and --ff
For more details behind these rules and a larger table of individual
cases, refer to https://lore.kernel.org/git/xmqqwnpqot4m.fsf@gitster.g/
and the links found therein.
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When running unpack_trees() with a sparse index, we attempt to operate
on the index without expanding the sparse directory entries. Thus, we
operate by manipulating entire directories and passing them to the
unpack function. In the case of the 'git checkout' command, this is the
twoway_merge() function.
There are several cases in twoway_merge() that handle different
situations. One new one to add is the case of a directory/file conflict
where the directory is sparse. Before the sparse index, such a conflict
would appear as a list of file additions and deletions. Now,
twoway_merge() initializes 'current', 'oldtree', and 'newtree' from
src[0], src[1], and src[2], then sets 'oldtree' to NULL because it is
equal to the df_conflict_entry. The way to determine that we have a
directory/file conflict is to test that 'current' and 'newtree' disagree
on being sparse directory entries.
When we are in this case, we want to resolve the situation by calling
merged_entry(). This allows replacing the 'current' entry with the
'newtree' entry. This is important for cases where we want to run 'git
checkout' across the conflict and have the new HEAD represent the new
file type at that path. The first NEEDSWORK comment dropped in t1092
demonstrates this necessary behavior.
However, we still are in a confusing state when 'current' corresponds to
a staged change within a sparse directory that is not present at HEAD.
This should be atypical, because it requires adding a change outside of
the sparse-checkout cone, but it is possible. Since we are unable to
determine that this is a staged change within twoway_merge(), we cannot
add a case to reject the merge at this point. I believe this is due to
the use of df_conflict_entry in the place of 'oldtree' instead of using
the valud at HEAD, which would provide some perspective to this
decision. Any change that would allow this differentiation for staged
entries would need to involve information further up in unpack_trees().
That work should be done, sometime, because we are further confusing the
behavior of a directory/file conflict when staging a change in the
directory. The two cases 'checkout behaves oddly with df-conflict-?' in
t1092 demonstrate that even without a sparse-checkout, Git is not
consistent in its behavior. Neither of the two options seems correct,
either. This change makes the sparse-index behave differently than the
typcial sparse-checkout case, but it does match the full checkout
behavior in the df-conflict-2 case.
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add new branches to the test repo that demonstrate directory/file
conflicts in different ways. Since the directory 'folder1/' has
adjacent files 'folder1-', 'folder1.txt', and 'folder10' it causes
searches for 'folder1/' to land in a different place in the index than a
search for 'folder1'. This causes a change in behavior when working with
the df-conflict-1 and df-conflict-2 branches, whose only difference is
that the first uses 'folder1' as the conflict and the other uses
'folder2' which does not have these adjacent files.
We can extend two tests that compare the behavior across different 'git
checkout' commands, and we see already that the behavior will be
different in some cases and not in others. The difference between the
two test loops is that one uses 'git reset --hard' between iterations.
Further, we isolate the behavior of creating a staged change within a
directory and then checking out a branch where that directory is
replaced with a file. A full checkout behaves differently across these
two cases, while a sparse-checkout cone behaves consistently. In both
cases, the behavior is wrong. In one case, the staged change is dropped
entirely. The other case the staged change is kept, replacing the file
at that location, but none of the other files in the directory are kept.
Likely, the correct behavior in this case is to reject the checkout and
report the conflict, leaving HEAD in its previous location. None of the
cases behave this way currently. Use comments to demonstrate that the
tested behavior is only a documentation of the current, incorrect
behavior to ensure we do not _accidentally_ change it. Instead, we would
prefer to change it on purpose with a future change.
At this point, the sparse-index does not handle these 'git checkout'
commands correctly. Or rather, it _does_ reject the 'git checkout' when
we have the staged change, but for the wrong reason. It also rejects the
'git checkout' commands when there is no staged change and we want to
replace a directory with a file. A fix for that unstaged case will
follow in the next change, but that will make the sparse-index agree
with the full checkout case in these documented incorrect behaviors.
Helped-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Derrick Stolee <dstolee@microsoft.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The merge algorithm mostly consists of the following three functions:
collect_merge_info()
detect_and_process_renames()
process_entries()
Prior to the trivial directory resolution optimization of the last half
dozen commits, process_entries() was consistently the slowest, followed
by collect_merge_info(), then detect_and_process_renames(). When the
trivial directory resolution applies, it often dramatically decreases
the amount of time spent in the two slower functions.
Looking at the performance results in the previous commit, the trivial
directory resolution optimization helps amazingly well when there are no
relevant renames. It also helps really well when reapplying a long
series of linear commits (such as in a rebase or cherry-pick), since the
relevant renames may well be cached from the first reapplied commit.
But when there are any relevant renames that are not cached (represented
by the just-one-mega testcase), then the optimization does not help at
all.
Often, I noticed that when the optimization does not apply, it is
because there are a handful of relevant sources -- maybe even only one.
It felt frustrating to need to recurse into potentially hundreds or even
thousands of directories just for a single rename, but it was needed for
correctness.
However, staring at this list of functions and noticing that
process_entries() is the most expensive and knowing I could avoid it if
I had cached renames suggested a simple idea: change
collect_merge_info()
detect_and_process_renames()
process_entries()
into
collect_merge_info()
detect_and_process_renames()
<cache all the renames, and restart>
collect_merge_info()
detect_and_process_renames()
process_entries()
This may seem odd and look like more work. However, note that although
we run collect_merge_info() twice, the second time we get to employ
trivial directory resolves, which makes it much faster, so the increased
time in collect_merge_info() is small. While we run
detect_and_process_renames() again, all renames are cached so it's
nearly a no-op (we don't call into diffcore_rename_extended() but we do
have a little bit of data structure checking and fixing up). And the
big payoff comes from the fact that process_entries(), will be much
faster due to having far fewer entries to process.
This restarting only makes sense if we can save recursing into enough
directories to make it worth our while. Introduce a simple heuristic to
guide this. Note that this heuristic uses a "wanted_factor" that I have
virtually no actual real world data for, just some back-of-the-envelope
quasi-scientific calculations that I included in some comments and then
plucked a simple round number out of thin air. It could be that
tweaking this number to make it either higher or lower improves the
optimization. (There's slightly more here; when I first introduced this
optimization, I used a factor of 10, because I was completely confident
it was big enough to not cause slowdowns in special cases. I was
certain it was higher than needed. Several months later, I added the
rough calculations which make me think the optimal number is close to 2;
but instead of pushing to the limit, I just bumped it to 3 to reduce the
risk that there are special cases where this optimization can result in
slowing down the code a little. If the ratio of path counts is below 3,
we probably will only see minor performance improvements at best
anyway.)
Also, note that while the diffstat looks kind of long (nearly 100
lines), more than half of it is in two comments explaining how things
work.
For the testcases mentioned in commit 557ac0350d ("merge-ort: begin
performance work; instrument with trace2_region_* calls", 2020-10-28),
this change improves the performance as follows:
Before After
no-renames: 205.1 ms ± 3.8 ms 204.2 ms ± 3.0 ms
mega-renames: 1.564 s ± 0.010 s 1.076 s ± 0.015 s
just-one-mega: 479.5 ms ± 3.9 ms 364.1 ms ± 7.0 ms
Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Testcases in t0000 are quite special given that they many of them run
nested testcases to verify that testing functionality itself works as
expected. These nested testcases are realized by writing a new ad-hoc
test script which again sources test-lib.sh, where the new script is
created in a nested subdirectory located beneath the current trash
directory. We then execute the new test script with the nested
subdirectory as current working directory and explicitly re-export
TEST_OUTPUT_DIRECTORY to point to that directory.
While this works as expected in the general case, it falls apart when
the developer has TEST_OUTPUT_DIRECTORY explicitly defined either via
the environment or via config.mak and runs "make test". In that case,
test-lib.sh will clobber the value that we've just carefully set up to
instead contain what the developer has defined. As a result, the
TEST_OUTPUT_DIRECTORY continues to point at the root output directory,
not at the nested one.
This issue causes breakage in the 'test_atexit is run' test case: the
nested test case writes files into "../../", which is assumed to be the
parent's trash directory. But because TEST_OUTPUT_DIRECTORY already
points to to the root output directory, we instead end up writing those
files outside of the output directory. The parent test case will then
try to check whether those files still exist in its own trash directory,
which thus must fail now.
Fix the issue by adding a new TEST_OUTPUT_DIRECTORY_OVERRIDE variable.
If set, then we'll always override the TEST_OUTPUT_DIRECTORY with its
value after sourcing GIT-BUILD-OPTIONS.
Signed-off-by: Patrick Steinhardt <ps@pks.im>
Acked-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Since cd57bc41bb (builtin/multi-pack-index.c: display usage on
unrecognized command, 2021-03-30) we have used a "usage" label to avoid
having two separate callers of usage_with_options (one when no arguments
are given, and another for unrecognized sub-commands).
But the first caller has been broken since cd57bc41bb, since it will
happily jump to usage without arguments, and then pass argv[0] to the
"unrecognized subcommand" error.
Many compilers will save us from a segfault here, but the end result is
ugly, since it mentions an unrecognized subcommand when we didn't even
pass one, and (on GCC) includes "(null)" in its output.
Move the "usage" label down past the error about unrecognized
subcommands so that it is only triggered when it should be. While we're
at it, bulk up our test coverage in this area, too.
Signed-off-by: Taylor Blau <me@ttaylorr.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
