files-backend is now initialized with a $GIT_DIR. Converting a submodule
path to where real submodule gitdir is located is done in get_ref_store().
This gives a slight performance improvement for submodules since we
don't convert submodule path to gitdir at every backend call like
before. We pay that once at ref-store creation.
More cleanup in files_downcast() and files_assert_main_repository()
follows shortly. It's separate to keep noises from this patch.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
refs.c: make get_main_ref_store() public and use it
get_ref_store() will soon be renamed to get_submodule_ref_store().
Together with future get_worktree_ref_store(), the three functions
provide an appropriate ref store for different operation modes. New APIs
will be added to operate directly on ref stores.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This is the last function in this code (besides public API) that takes
submodule argument and handles both main/submodule cases. Break it down,
move main store registration in get_main_ref_store() and keep the rest
in register_submodule_ref_store().
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This helps the future changes in this code. And because get_ref_store()
is destined to become get_submodule_ref_store(), the "get main store"
code path will be removed eventually. After this the patch to delete
that code will be cleaner.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
refs: rename lookup_ref_store() to lookup_submodule_ref_store()
With get_main_ref_store() being used inside get_ref_store(),
lookup_ref_store() is only used for submodule code path. Rename to
reflect that and delete dead code.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Given $GIT_DIR and $GIT_COMMON_DIR, files-backend is now in charge of
deciding what goes where (*). The end goal is to pass $GIT_DIR only. A
refs "view" of a linked worktree is a logical ref store that combines
two files backends together.
(*) Not entirely true since strbuf_git_path_submodule() still does path
translation underneath. But that's for another patch.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Keep repo-related path handling in one place. This will make it easier
to add submodule/multiworktree support later.
This automatically adds the "if submodule then use the submodule version
of git_path" to other call sites too. But it does not mean those
operations are submodule-ready. Not yet.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
files-backend: convert git_path() to strbuf_git_path()
git_path() and friends are going to be killed in files-backend.c in near
future. And because there's a risk with overwriting buffer in
git_path(), let's convert them all to strbuf_git_path(). We'll have
easier time killing/converting strbuf_git_path() then because we won't
have to worry about memory management again.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
files-backend: delete dead code in files_init_db()
safe_create_dir() can do adjust_shared_perm() internally, and init-db
has always created 'refs' in shared mode since the beginning, af6e277c5e (git-init-db: initialize shared repositories with --shared -
2005-12-22). So this code looks like extra adjust_shared_perm calls are
unnecessary.
And they are. But let's see why there are here in the first place.
This code was added in 6fb5acfd8f (refs: add methods to init refs db -
2016-09-04). From the diff alone this looks like a faithful refactored
code from init-db.c. But there is a subtle difference:
Between the safe_create_dir() block and adjust_shared_perm() block in
the old init-db.c, we may copy/recreate directories from the repo
template. So it makes sense that adjust_shared_perm() is re-executed
then to fix potential permission screwups.
After 6fb5acfd8f, refs dirs are created after template is copied. Nobody
will change directory permission again. So the extra adjust_shared_perm()
is redudant. Delete them.
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
files-backend.c: delete dead code in files_ref_iterator_begin()
It's not in the diff context, but files_downcast() is called before this
check. If "refs" is NULL, we would have segfaulted before reaching the
check here. And we should never see NULL refs in backend code (frontend
should have caught it).
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule.c: port is_submodule_modified to use porcelain 2
Migrate 'is_submodule_modified' to the new porcelain format of
git-status. This conversion attempts to convert faithfully, i.e.
the behavior ought to be exactly the same.
As the output in the parsing only distinguishes between untracked files
and the rest, this is easy to port to the new format, as we only
need to identify untracked files and the rest is handled in the "else"
case.
untracked files are indicated by only a single question mark instead of
two question marks, so the conversion is easy.
Signed-off-by: Stefan Beller <sbeller@google.com> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
submodule.c: convert is_submodule_modified to use strbuf_getwholeline
Instead of implementing line reading yet again, make use of our beautiful
library function to read one line. By using strbuf_getwholeline instead
of strbuf_read, we avoid having to allocate memory for the entire child
process output at once. That is, we limit maximum memory usage.
Also we can start processing the output as it comes in, no need to
wait for all of it.
Once we know all information that we care about, we can terminate
the child early. In that case we do not care about its exit code as well.
By just closing our side of the pipe the child process will get a SIGPIPE
signal, which it will not report nor do we report it in finish_command, ac78663b0d (run-command: don't warn on SIGPIPE deaths, 2015-12-29).
Helped-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Stefan Beller <sbeller@google.com> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Convert GIT_SHA1_RAWSZ used for allocation to GIT_MAX_RAWSZ
Since we will likely be introducing a new hash function at some point,
and that hash function might be longer than 20 bytes, use the constant
GIT_MAX_RAWSZ, which is designed to be suitable for allocations, instead
of GIT_SHA1_RAWSZ. This will ease the transition down the line by
distinguishing between places where we need to allocate memory suitable
for the largest hash from those where we need to handle the current
hash.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Convert GIT_SHA1_HEXSZ used for allocation to GIT_MAX_HEXSZ
Since we will likely be introducing a new hash function at some point,
and that hash function might be longer than 40 hex characters, use the
constant GIT_MAX_HEXSZ, which is designed to be suitable for
allocations, instead of GIT_SHA1_HEXSZ. This will ease the transition
down the line by distinguishing between places where we need to allocate
memory suitable for the largest hash from those where we need to handle
the current hash.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Define new hash-size constants for allocating memory
Since we will want to transition to a new hash at some point in the
future, and that hash may be larger in size than 160 bits, introduce two
constants that can be used for allocating a sufficient amount of memory.
They can be increased to reflect the largest supported hash size.
Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
strbuf: support long paths w/o read rights in strbuf_getcwd() on FreeBSD
FreeBSD implements getcwd(3) as a syscall, but falls back to a version
based on readdir(3) if it fails for some reason. The latter requires
permissions to read and execute path components, while the former does
not. That means that if our buffer is too small and we're missing
rights we could get EACCES, but we may succeed with a bigger buffer.
Keep retrying if getcwd(3) indicates lack of permissions until our
buffer can fit PATH_MAX bytes, as that's the maximum supported by the
syscall on FreeBSD anyway. This way we do what we can to be able to
benefit from the syscall, but we also won't loop forever if there is a
real permission issue.
This fixes a regression introduced with 7333ed17 (setup: convert
setup_git_directory_gently_1 et al. to strbuf, 2014-07-28) for paths
longer than 127 bytes with components that miss read or execute
permissions (e.g. 0711 on /home for privacy reasons); we used a fixed
PATH_MAX-sized buffer before.
Reported-by: Zenobiusz Kunegunda <zenobiusz.kunegunda@interia.pl> Signed-off-by: Rene Scharfe <l.s.r@web.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
doc/SubmittingPatches: show how to get a CLI commit summary
Amend the section which describes how to get a commit summary to show
how do to that with "git show", currently the documentation only shows
how to do that with gitk.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
t/README: clarify the test_have_prereq documentation
Clarify the test_have_prereq documentation so that it's clear in the
reader's mind when the text says "most common use of this directly"
what the answer to "as opposed to what?" is.
Usually this function isn't used in lieu of using the prerequisite
support built into test_expect_*, mention that explicitly.
This changes documentation that I added in commit 9a897893a7 ("t/README: Document the prereq functions, and 3-arg
test_*", 2010-07-02).
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
In an early part of sha1dc/sha1.c, the code checks the endianness of
the target platform by inspecting common CPP macros defined on
big-endian boxes, and sets BIGENDIAN macro to 1. If these common
CPP macros are not defined, the code declares that the target
platform is little endian and does nothing (most notably, it does
not #undef its BIGENDIAN macro).
The code that does so even has this comment
Note that all MSFT platforms are little endian,
so none of these will be defined under the MSC compiler.
and later, the defined-ness of the BIGENDIAN macro is used to switch
the implementation of sha1_load() macro.
One thing the code did not anticipate is that somebody might define
BIGENDIAN macro in some header it includes to 0 on a little-endian
target platform. Because the auto-detection based on common macros
do not touch BIGENDIAN macro when it detects a little-endian target,
such a definition is still valid and then defined-ness test will say
"Ah, BIGENDIAN is defined" and takes the wrong sha1_load().
As this auto-detection logic pretends as if it owns the BIGENDIAN
macro by ignoring the setting that may come from the outside and by
not explicitly unsetting when it decides that it is working for a
little-endian target, solve this problem without breaking that
assumption. Namely, we can rename BIGENDIAN this code uses to
something much less generic, i.e. SHA1DC_BIGENDIAN. For extra
protection, undef the macro on a little-endian target.
It is possible to work it around by instead #undef BIGENDIAN in
the auto-detection code, but a macro (or include) that happens later
in the code can be implemented in terms of BIGENDIAN on Windows and
it is possible that the implementation gets upset when it sees the
CPP macro undef'ed (instead of set to 0). Renaming the private macro
intended to be used only in this file to a less generic name relieves
us from having to worry about that kind of breakage.
Noticed-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The "detect attempt to create collisions" variant of SHA-1
implementation by Marc Stevens (CWI) and Dan Shumow (Microsoft)
has been integrated and made the default.
* jk/sha1dc:
Makefile: make DC_SHA1 the default
t0013: add a basic sha1 collision detection test
Makefile: add DC_SHA1 knob
sha1dc: disable safe_hash feature
sha1dc: adjust header includes for git
sha1dc: add collision-detecting sha1 implementation
The default location "~/.git-credential-cache/socket" for the
socket used to communicate with the credential-cache daemon has
been moved to "~/.cache/git/credential/socket".
* dl/credential-cache-socket-in-xdg-cache:
credential-cache: add tests for XDG functionality
credential-cache: use XDG_CACHE_HOME for socket
path.c: add xdg_cache_home
Merge branch 'jc/config-case-cmdline-take-2' into maint
The code to parse "git -c VAR=VAL cmd" and set configuration
variable for the duration of cmd had two small bugs, which have
been fixed.
This supersedes jc/config-case-cmdline topic that has been discarded.
* jc/config-case-cmdline-take-2:
config: use git_config_parse_key() in git_config_parse_parameter()
config: move a few helper functions up
The code to parse the command line "git grep <patterns>... <rev>
[[--] <pathspec>...]" has been cleaned up, and a handful of bugs
have been fixed (e.g. we used to check "--" if it is a rev).
* jk/grep-no-index-fix:
grep: treat revs the same for --untracked as for --no-index
grep: do not diagnose misspelt revs with --no-index
grep: avoid resolving revision names in --no-index case
grep: fix "--" rev/pathspec disambiguation
grep: re-order rev-parsing loop
grep: do not unnecessarily query repo for "--"
grep: move thread initialization a little lower
Merge branch 'jn/remote-helpers-with-git-dir' into maint
"git ls-remote" and "git archive --remote" are designed to work
without being in a directory under Git's control. However, recent
updates revealed that we randomly look into a directory called
.git/ without actually doing necessary set-up when working in a
repository. Stop doing so.
* jn/remote-helpers-with-git-dir:
remote helpers: avoid blind fall-back to ".git" when setting GIT_DIR
remote: avoid reading $GIT_DIR config in non-repo
Merge branch 'jk/push-deadlock-regression-fix' into maint
"git push" had a handful of codepaths that could lead to a deadlock
when unexpected error happened, which has been fixed.
* jk/push-deadlock-regression-fix:
send-pack: report signal death of pack-objects
send-pack: read "unpack" status even on pack-objects failure
send-pack: improve unpack-status error messages
send-pack: use skip_prefix for parsing unpack status
send-pack: extract parsing of "unpack" response
receive-pack: fix deadlock when we cannot create tmpdir
pack.h: define largest possible encoded object size
Several callers use fixed buffers for storing the pack
object header, and they've picked 10 as a magic number. This
is reasonable, since it handles objects up to 2^67. But
let's give them a constant so it's clear that the number
isn't pulled out of thin air.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The encode_in_pack_object_header() writes a variable-length
header to an output buffer, but it doesn't actually know
long the buffer is. At first glance, this looks like it
might be possible to overflow.
In practice, this is probably impossible. The smallest
buffer we use is 10 bytes, which would hold the header for
an object up to 2^67 bytes. Obviously we're not likely to
see such an object, but we might worry that an object could
lie about its size (causing us to overflow before we realize
it does not actually have that many bytes). But the argument
is passed as a uintmax_t. Even on systems that have __int128
available, uintmax_t is typically restricted to 64-bit by
the ABI.
So it's unlikely that a system exists where this could be
exploited. Still, it's easy enough to use a normal out/len
pair and make sure we don't write too far. That protects the
hypothetical 128-bit system, makes it harder for callers to
accidentally specify a too-small buffer, and makes the
resulting code easier to audit.
Note that the one caller in fast-import tried to catch such
a case, but did so _after_ the call (at which point we'd
have already overflowed!). This check can now go away.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The stream_blob() function checks the return value of
snprintf and dies. This is more simply done with
xsnprintf (and matches the similar call in store_object).
The message the user would get is less specific, but since
the point is that this _shouldn't_ ever happen, that's OK.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When we have to write a sha1 with a newline, we do so by
copying both into a single buffer, so that we can issue a
single write() call.
We use snprintf but don't bother to check the output, since
we know it will fit. However, we should use xsnprintf() in
such a case so that we're notified if our assumption turns
out to be wrong (and to make it easier to audit for
unchecked snprintf calls).
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Map both old addresses to the new, hopefully more permanent one.
Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net> Signed-off-by: Michael J Gruber <git@grubix.eu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the test suite to test for these synonyms for --contains and
--no-contains, respectively.
Before this change there were no tests for them at all. This doesn't
exhaustively test for them as well as their --contains and
--no-contains synonyms, but at least it's something.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Reflow the recently changed branch/tag-for-each-ref
documentation. This change shows no changes under --word-diff, except
the innocuous change of moving git-tag.txt's "[--sort=<key>]" around
slightly.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
ref-filter: add --no-contains option to tag/branch/for-each-ref
Change the tag, branch & for-each-ref commands to have a --no-contains
option in addition to their longstanding --contains options.
This allows for finding the last-good rollout tag given a known-bad
<commit>. Given a hypothetically bad commit cf5c7253e0, the git
version to revert to can be found with this hacky two-liner:
As the filtering machinery is shared between the tag, branch &
for-each-ref commands, implement this for those commands too. A
practical use for this with "branch" is e.g. finding branches which
were branched off between v2.8.0 and v2.10.0:
The "describe" command also has a --contains option, but its semantics
are unrelated to what tag/branch/for-each-ref use --contains for. A
--no-contains option for "describe" wouldn't make any sense, other
than being exactly equivalent to not supplying --contains at all,
which would be confusing at best.
Add a --without option to "tag" as an alias for --no-contains, for
consistency with --with and --contains. The --with option is
undocumented, and possibly the only user of it is
Junio (<xmqqefy71iej.fsf@gitster.mtv.corp.google.com>). But it's
trivial to support, so let's do that.
The additions to the the test suite are inverse copies of the
corresponding --contains tests. With this change --no-contains for
tag, branch & for-each-ref is just as well tested as the existing
--contains option.
In addition to those tests, add a test for "tag" which asserts that
--no-contains won't find tree/blob tags, which is slightly
unintuitive, but consistent with how --contains works & is documented.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the --points-at option to default to HEAD for consistency with
its siblings --contains, --merged etc. which default to
HEAD. Previously we'd get:
$ git tag --points-at 2>&1 | head -n 1
error: option `points-at' requires a value
This changes behavior added in commit ae7706b9ac (tag: add --points-at
list option, 2012-02-08).
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
tag: implicitly supply --list given another list-like option
Change the "tag" command to implicitly turn on its --list mode when
provided with a list-like option such as --contains, --points-at etc.
This is for consistency with how "branch" works. When "branch" is
given a list-like option, such as --contains, it implicitly provides
--list. Before this change "tag" would error out on those sorts of
invocations. I.e. while both of these worked for "branch":
git tag --contains v2.8.0 '*rc*'
git tag --list --contains v2.8.0 '*rc*'
Now "tag", like "branch", will implicitly supply --list when a
list-like option is provided, and no other conflicting non-list
options (such as -d) are present on the command-line.
Reveals that there was no good reason for not allowing this in the
first place. The --contains option added in 32c35cfb1e ("git-tag: Add
--contains option", 2009-01-26) made this an error. All the other
subsequent list-like options that were added copied its pattern of
making this usage an error.
The only tests that break as a result of this change are tests that
were explicitly checking that this "branch-like" usage wasn't
permitted. Change those failing tests to check that this invocation
mode is permitted, add extra tests for the list-like options we
weren't testing, and tests to ensure that e.g. we don't toggle the
list mode in the presence of other conflicting non-list options.
With this change errors messages such as "--contains option is only
allowed with -l" don't make sense anymore, since options like
--contain turn on -l. Instead we error out when list-like options such
as --contain are used in conjunction with conflicting options such as
-d or -v.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the documentation for --list so that it's described as a
toggle, not as an option that takes a <pattern> as an argument.
Junio initially documented this in b867c7c23a ("git-tag: -l to list
tags (usability).", 2006-02-17), but later Jeff King changed "tag" to
accept multiple patterns in 588d0e834b ("tag: accept multiple patterns
for --list", 2011-06-20).
However, documenting this as "-l <pattern>" was never correct, as
these both worked before Jeff's change:
git tag -l 'v*'
git tag 'v*' -l
One would expect an option that was documented like that to only
accept:
git tag --list
git tag --list 'v*rc*'
And after Jeff's change, one that took multiple patterns:
git tag --list 'v*rc*' --list '*2.8*'
But since it's actually a toggle all of these work as well, and
produce identical output to the last example above:
git tag --list 'v*rc*' '*2.8*'
git tag --list 'v*rc*' '*2.8*' --list --list --list
git tag --list 'v*rc*' '*2.8*' --list -l --list -l --list
Now the documentation is more in tune with how the "branch" command
describes its --list option since commit cddd127b9a ("branch:
introduce --list option", 2011-08-28).
Change the test suite to assert that these invocations work for the
cases that weren't already being tested for.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
parse-options: add OPT_NONEG to the "contains" option
Add the OPT_NONEG flag to the "contains" option and its hidden synonym
"with". Since this was added in commit 694a577519 ("git-branch
--contains=commit", 2007-11-07) giving --no-{contains,with} hasn't
been an error, but has emitted the help output since
filter.with_commit wouldn't get set.
Now git will emit "error: unknown option `no-{contains,with}'" at the
top of the help output.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The pager_in_use() function predates git_env_bool(), but
ends up doing the same thing. Let's make use of the latter,
which is shorter and less repetitive.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Created t/perf/p0004-lazy-init-name-hash.sh test
to demonstrate correctness and performance gains
with the multithreaded version of lazy_init_name_hash().
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Add t/helper/test-lazy-init-name-hash.c test code
to demonstrate performance times for lazy_init_name_hash()
using the original single-threaded and the new multi-threaded
code paths.
Includes a --dump option to dump the created hashmaps to
stdout. You can use this to run both code paths and
confirm that they generate the same hashmaps.
Includes a --analyze option to analyze performance of both
code paths over a range of index sizes to help you find a
lower bound for the LAZY_THREAD_COST in name-hash.c.
For example, passing "-a 4000" will set "istate.cache_nr"
to 4000 and then try the multi-threaded code -- probably
giving 2 threads with 2000 entries each. It will then
run both the single-threaded (1x4000) and the multi-threaded
(2x2000) and compare the times. It will then repeat the
test with 8000, 12000, and etc. so that you can see the
cross over.
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
name-hash: perf improvement for lazy_init_name_hash
Improve performance of lazy_init_name_hash() when
ignore_case is set. Teach name-hash to build the
istate.name_hash and istate.dir_hash simultaneously
using a forward-diving technique on the pathname
of the index_entry, rather than adding name_hash
entries and then searching backwards in the pathname
for parent directories.
This borrows algorithm ideas from clear_ce_flags_{1,dir}.
Multiple threads are used with the new algorithm to
speed hashmap construction.
This new code path is only used when threads are present
(a compiler settings) and when the index is large enough
to warrant the pthread complexity.
The code in clear_ce_flags_dir() uses a linear search to
find the adjacent index entries with the same prefix; a
binary search is used here handle_range_dir() to further
speed things up.
The size of LAZY_THREAD_COST was determined from rough
analysis using:
t/helper/test-lazy-init-name-hash --analyze
Signed-off-by: Jeff Hostetler <jeffhost@microsoft.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jan Palus noticed that some here-doc are spelled incorrectly,
resulting the entire remainder of the test snippet being slurped
into the "expect" file as if it were data, e.g. in this sequence
the last command of the test is "cat" that sends everything to
'expect' and succeeds.
Fixing these issues in t7004 and t7030 reveals that "git tag -v"
and "git verify-tag" with their --format option do not work as the
test was expecting originally. Instead of showing both valid tags
and tags with incorrect signatures on their output, tags that do not
pass verification are omitted from the output. Another breakage that
is uncovered is that these tests must be restricted to environment
where gpg is available.
Arguably, that is a safer behaviour, and because the format
specifiers like %(tag) do not have a way to show if the signature
verifies correctly, the command with the --format option cannot be
used to get a list of tags annotated with their signature validity
anyway.
For now, let's fix the here-doc syntax, update the expectation to
match the reality, and update the test prerequisite.
Maybe later when we extend the --format language available to "git
tag -v" and "git verify-tag" to include things like "%(gpg:status)",
we may want to change the behaviour so that piping a list of tag
names into
becomes a good way to produce such a list, but that is a separate
topic.
Noticed-by: Jan Palus <jan.palus@gmail.com> Helped-by: Jeff King <peff@peff.net> Signed-off-by: Santiago Torres <santiago@nyu.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The paragraph begins with a sample command line `git branch <name>`
that has nothing to do with the option being described. Remove it,
but use the space to instead show that multiple patterns can be
given.
Also mention the unfortunate `-l` that can be easily confused with
it.
Signed-off-by: Junio C Hamano <gitster@pobox.com> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Reviewed-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
branch doc: change `git branch <pattern>` to use `<branchname>`
Change an example for `git branch <pattern>` to say `git branch
<branchname>` to be consistent with the synopsis. This changes
documentation added in d8d33736b5 ("branch: allow pattern arguments",
2011-08-28).
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We found a few run-away here documents that are started with an
end-of-here-doc marker that is incorrectly spelled, e.g.
git some command >actual &&
cat <<EOF >expect
...
EOF &&
test_cmp expect actual
which ends up slurping the entire remainder of the script as if it
were the data. Often the command that gets misused like this exits
without failure (e.g. "cat" in the above example), which makes the
command appear to work, without ever executing the remainder of the
test.
Piggy-back on the test that catches &&-chain breakage to detect this
case as well.
completion: offer ctags symbol names for 'git log -S', '-G' and '-L:'
Just like in the case of search patterns for 'git grep', see 29eec71f2
(completion: match ctags symbol names in grep patterns, 2011-10-21)),
a common thing to look for using 'git log -S', '-G' and '-L:' is the
name of a symbol.
Teach the completion for 'git log' to offer ctags symbol names after
these options, both in stuck and in unstuck forms.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: extract completing ctags symbol names into helper function
The previous commit doubled the number of __git_match_ctag()'s
positional parameters, and, to keep the position of existing
parameters for the sake of backwards compatibility, the prefix,
current word and suffix parameters ended up in different order than in
other functions accepting the same parameters. Then there is a
condition checking the existence of the tag file before invoking this
function.
We could still live with this if there were only a single callsite,
but the next commit will add a few more, so it's worth providing a
cleaner interface.
Add the wrapper function __git_complete_symbol(), which encompasses
the condition for checking the presence of the tag file and filling
COMPREPLY, and accepts '--opt=val'-style options with default values
that keep callsites simpler.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: put matching ctags symbol names directly into COMPREPLY
The one-liner awk script in __git_match_ctag() listing ctags symbol
names for 'git grep <TAB>' is already smart enough to list only symbol
names matching the current word to be completed.
Extend this helper function to accept prefix and suffix parameters to
be prepended and appended, respectively, to each listed symbol name in
the awk script, so its output won't require any additional processing
or filtering in the completion script before being handed over to
Bash. Use the faster __gitcomp_direct() helper instead of
__gitcomp_nl() to fill the fully processed matching symbol names into
Bash's COMPREPLY array.
Right after 'git grep <TAB>' in current git.git with 14k+ symbol names
in the tag file, best of five:
Before:
$ time __gitcomp_nl "$(__git_match_ctag "" tags)"
real 0m0.178s
user 0m0.176s
sys 0m0.000s
After:
$ time __gitcomp_direct "$(__git_match_ctag "" tags "" " ")"
real 0m0.058s
user 0m0.048s
sys 0m0.008s
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When using rebase --interactive where one of the lines is marked as
'edit' this is the resulting output:
Stopped at ec3b9c4... stuffYou can amend the commit now, with
git commit --amend
Once you are satisfied with your changes, run
git rebase --continue
A newline character is missing at the end of the "Stopped at ..." line and
before the "You can amend ..." line. This patch fixes the malformed output by
adding the missing newline character to the end of the "Stopped at ..." line.
Signed-off-by: Brandon Williams <bmwill@google.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Modify __git_heads() and __git_tags() and the few callsites they have,
so we can let 'git for-each-ref' do all the hard work and these
functions' output won't need any further processing or filtering
before being handed over to Bash, resulting in faster branch and tag
completion. These are some of the same tricks used in the previous
commits to speed up refs completion, namely:
- Extend both functions to accept prefix, current word and suffix
positional parameters, all optional and all empty by default to
keep the parameterless behavior unaltered.
- Specify appropriate globbing patterns to 'git for-each-ref' to
list only branches or tags matching the given current word
parameter.
- Modify the 'git for-each-ref --format=<...>' to include the given
prefix and suffix.
- Adjust all callsites to specify the proper prefix, current word
and suffix parameters, and to fill COMPREPLY using
__gitcomp_direct().
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: fill COMPREPLY directly when completing fetch refspecs
The __git_complete_fetch_refspecs() has to iterate over __git_refs()'s
output anyway to turn the listed matching refs into refspecs, and it
knows about the prefix and suffix that has to be added to each
refspec.
Modify this function to add the prefix and suffix to each refspec
while iterating and feed the result, since it doesn't need further
processing, to the new __gitcomp_direct() helper added in the previous
commit, because it should be faster when there are a lot of refspecs
to list.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: fill COMPREPLY directly when completing refs
__gitcomp_nl() iterates over all the possible completion words it gets
as argument
- filtering matching words,
- appending a trailing space to each matching word (in all but two
cases),
- prepending a prefix to each matching word (when completing words
after e.g. '--option=<TAB>' or 'master..<TAB>'), and
- adding each matching word to the COMPREPLY array.
This takes a while when a lot of refs are passed to __gitcomp_nl().
The previous changes in this series ensure that __git_refs() lists
only refs matching the current word to be completed, making a second
filtering in __gitcomp_nl() redundant.
Adding the necessary prefix and suffix could be done in __git_refs()
as well:
- When refs come from 'git for-each-ref', then that prefix and
suffix could be added much more efficiently using a 'git
for-each-ref' format containing said prefix and suffix. Care
should be taken, though, because that prefix might contain
'for-each-ref' format specifiers as part of the left hand side of
a '..' range or '...' symmetric difference notation or
fetch/push/etc. refspec, e.g. 'git log "evil-%(refname)..br<TAB>'.
Doubling every '%' in the prefix will prevent 'git for-each-ref'
from interpolating any of those contained specifiers.
- When refs come from 'git ls-remote', then that prefix and suffix
can be added in the shell loop that has to process 'git
ls-remote's output anyway.
- Finally, the prefix and suffix can be added to that handful of
potentially matching symbolic and pseudo refs right away in the
shell loop listing them.
And then all what is still left to do is to assign a bunch of
newline-separated words to a shell array, which can be done without a
shell loop iterating over each word, basically making all of
__gitcomp_nl() unnecessary for refs completion.
Add the helper function __gitcomp_direct() to fill the COMPREPLY array
with prefiltered and preprocessed words without any additional
processing, without a shell loop, with just one single compound
assignment. Modify __git_refs() to accept prefix and suffix
parameters and add them to each and every listed ref as described
above. Modify __git_complete_refs() to pass the prefix and suffix
parameters to __git_refs() and to feed __git_refs()'s output to
__gitcomp_direct() instead of __gitcomp_nl().
This speeds up refs completion when there are a lot of refs matching
the current word to be completed. Listing all branches for completion
in a repo with 100k local branches, all packed, best of five:
On Linux, near the beginning of this series, for reference:
$ time __git_complete_refs
real 0m2.028s
user 0m1.692s
sys 0m0.344s
Before this patch:
real 0m1.135s
user 0m1.112s
sys 0m0.024s
After:
real 0m0.367s
user 0m0.352s
sys 0m0.020s
On Windows, near the beginning:
real 0m13.078s
user 0m1.609s
sys 0m0.060s
Before this patch:
real 0m2.093s
user 0m1.641s
sys 0m0.060s
After:
real 0m0.683s
user 0m0.203s
sys 0m0.076s
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: let 'for-each-ref' sort remote branches for 'checkout' DWIMery
When listing unique remote branches for 'git checkout's tracking
DWIMery, __git_refs() runs the classic '... |sort |uniq -u' pattern to
filter out duplicate remote branches.
Let 'git for-each-ref' do the sorting, sparing the overhead of
fork()+exec()ing 'sort' and a stage in the pipeline where potentially
relatively large amount of data can be passed between two subsequent
pipeline stages.
This speeds up refs completion for 'git checkout' a bit when a lot of
remote branches match the current word to be completed. Listing a
single local and 100k remote branches, all packed, best of five:
On Linux, before:
$ time __git_complete_refs --track
real 0m1.856s
user 0m1.816s
sys 0m0.060s
After:
real 0m1.550s
user 0m1.512s
sys 0m0.060s
On Windows, before:
real 0m3.128s
user 0m2.155s
sys 0m0.183s
After:
real 0m2.781s
user 0m1.826s
sys 0m0.136s
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: let 'for-each-ref' filter remote branches for 'checkout' DWIMery
The code listing unique remote branches for 'git checkout's tracking
DWIMery outputs only remote branches that match the current word to be
completed, but the filtering is done in a shell loop iterating over
all remote refs.
Let 'git for-each-ref' do the filtering, as it can do so much more
efficiently and we can remove that shell loop entirely.
This speeds up refs completion for 'git checkout' considerably when
there are a lot of non-matching remote refs to be filtered out.
Uniquely completing a branch in a repository with 100k remote
branches, all packed, best of five:
On Linux, before:
$ time __git_complete_refs --cur=maste --track
real 0m1.993s
user 0m1.740s
sys 0m0.304s
After:
real 0m0.266s
user 0m0.248s
sys 0m0.012s
On Windows, before:
real 0m6.187s
user 0m3.358s
sys 0m2.121s
After:
real 0m0.750s
user 0m0.015s
sys 0m0.090s
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: let 'for-each-ref' strip the remote name from remote branches
The code listing unique remote branches for 'git checkout's tracking
DWIMery uses a shell parameter expansion in a loop iterating over each
listed ref to remove the remote's name from the remote branches, i.e.
the leading path component from the short ref. When listing refs from
a configured remote repository, '| sed s///' is used for the same
purpose.
Let 'git for-each-ref' strip one more leading path component from the
refs, i.e. use the format 'refname:strip=3' instead of '=2', making
that parameter expansion and 'sed' execution unnecessary.
This speeds up refs completion for 'git checkout'. Uniquely
completing a branch for 'git checkout maste<TAB>' in a repo with 100k
remote branches, all packed, best of five:
On Linux, near the beginning of this series, for reference:
$ time __git_complete_refs --cur=maste --track
real 0m8.185s
user 0m6.896s
sys 0m1.616s
Before this patch:
real 0m2.714s
user 0m2.344s
sys 0m0.436s
After:
real 0m1.993s
user 0m1.740s
sys 0m0.304s
On Windows, near the beginning:
real 1m8.421s
user 0m7.591s
sys 0m3.557s
Before this patch:
real 0m8.191s
user 0m4.638s
sys 0m2.918s
After:
real 0m6.187s
user 0m3.358s
sys 0m2.121s
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: let 'for-each-ref' and 'ls-remote' filter matching refs
When completing refs, several __git_refs() code paths list all the
refs from the refs/{heads,tags,remotes}/ hierarchy and then
__gitcomp_nl() iterates over those refs in a shell loop to filter out
refs not matching the current ref to be completed. This comes with a
considerable performance penalty when a repository contains a lot of
refs but the current ref can be uniquely completed or when only a
handful of refs match the current ref.
Reduce the number of iterations in __gitcomp_nl() from the number of
refs to the number of matching refs by specifying appropriate globbing
patterns to 'git for-each-ref' and 'git ls-remote' to list only those
refs that match the current ref to be completed. However, do so only
when the ref to match is explicitly given as parameter, because the
current word on the command line might contain a prefix like
'--option=' or 'branch..'. The __git_complete_refs() and
__git_complete_fetch_refspecs() helpers introduced previously in this
patch series already call __git_refs() specifying this current ref
parameter, so all their callsites, i.e. all places in the completion
script doing refs completion, can benefit from this optimization.
Furthermore, list only those symbolic and pseudo refs that match the
current ref to be completed. Though it doesn't matter at all in
itself performance-wise, it will allow us further significant
optimizations later in this series.
This speeds up refs completion considerably when there are a lot of
non-matching refs to be filtered out. Uniquely completing a branch in
a repository with 100k local branches, all packed, best of five:
On Linux, before:
$ time __git_complete_refs --cur=maste
real 0m0.831s
user 0m0.808s
sys 0m0.028s
After:
real 0m0.119s
user 0m0.104s
sys 0m0.008s
On Windows, before:
real 0m1.480s
user 0m1.031s
sys 0m0.060s
After:
real 0m0.377s
user 0m0.015s
sys 0m0.030s
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When the completion script lists short refs it does so using the 'git
for-each-ref' format 'refname:short', which makes sure that all listed
refs are unambiguous. While disambiguating refs is technically
correct in this case, as opposed to the cases discussed in the
previous patch, this disambiguation involves several stat() syscalls
for each ref, thus, unfortunately, comes at a steep cost especially on
Windows and/or when there are a lot of refs to be listed. A user of
Git for Windows reported[1] 'git checkout <TAB>' taking ~11 seconds in
a repository with just about 4000 refs.
However, it's questionable whether ambiguous refs are really that bad
to justify that much extra cost:
- Ambiguous refs are not that common,
- even if a repository contains ambiguous refs, they only hurt when
the user actually happens to want to do something with one of the
ambiguous refs, and
- the issue can be easily circumvented by renaming those ambiguous
refs.
- On the other hand, apparently not that many refs are needed to
make refs completion unacceptably slow on Windows,
- and this slowness bites each and every time the user attempts refs
completion, even when the repository doesn't contain any ambiguous
refs.
- Furthermore, circumventing the issue might not be possible or
might be considerably more difficult and requires various
trade-offs (e.g. working in a repository with only a few selected
important refs while keeping a separate repository with all refs
for reference).
Arguably, in this case the benefits of technical correctness are
rather minor compared to the price we pay for it, and we are better
off opting for performance over correctness.
Use the 'git for-each-ref' format 'refname:strip=2' to list short refs
to spare the substantial cost of disambiguating.
This speeds up refs completion considerably. Uniquely completing a
branch in a repository with 100k local branches, all packed, best of
five:
When the completion script has to list only tags or only branches, it
uses the 'git for-each-ref' format 'refname:short', which makes sure
that all listed tags and branches are unambiguous. However,
disambiguating tags and branches in these cases is wrong, because:
- __git_tags(), the helper function listing possible tagname
arguments for 'git tag', lists an ambiguous tag
'refs/tags/ambiguous' as 'tags/ambiguous'. Its only consumer,
'git tag' expects its tagname argument to be under 'refs/tags/',
thus it interprets that abgiguous tag as
'refs/tags/tags/ambiguous'. Clearly wrong.
- __git_heads() lists possible branchname arguments for 'git branch'
and possible 'branch.<branchname>' configuration subsections.
Both of these expect branchnames to be under 'refs/heads/' and
misinterpret a disambiguated branchname like 'heads/ambiguous'.
Furthermore, disambiguation involves several stat() syscalls for each
tag or branch, thus comes at a steep cost especially on Windows and/or
when there are a lot of tags or branches to be listed.
Use the 'git for-each-ref' format 'refname:strip=2' instead of
'refname:short' to avoid harmful disambiguation of tags and branches
in __git_tags() and __git_heads().
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: support completing fully qualified non-fast-forward refspecs
After 'git fetch <remote> <TAB>' our completion script offers refspecs
that will fetch to a local branch with the same name as in the remote
repository, e.g. 'master:master'. This also completes
non-fast-forward refspecs, i.e. after a '+' prefix like
'+master:master', and fully qualified refspecs, e.g.
'refs/heads/master:refs/heads/master'. However, it does not complete
non-fast-forward fully qualified refspecs (or fully qualified refspecs
following any other prefix, e.g. '--option=', though currently no git
command supports such an option, but third party git commands might).
These refspecs are listed by the __git_refs2() function, which is just
a thin wrapper iterating over __git_refs()'s output, turning each
listed ref into a refspec. Now, it's certainly possible to modify
__git_refs2() and its callsite to pass an extra parameter containing
only the ref part of the current word to be completed (to follow suit
of the previous commit) to deal with prefixed fully qualified refspecs
as well. Unfortunately, keeping the current behavior unchanged in the
"no extra parameter" case brings in a bit of subtlety, which makes the
resulting code ugly and compelled me to write a 8-line long comment in
the proof of concept. Not good. However, since the callsite has to
be modified for proper functioning anyway, we might as well leave
__git_refs2() as is and introduce a new helper function without
backwards compatibility concerns.
Add the new function __git_complete_fetch_refspecs() that has all the
necessary parameters to do the right thing in all cases mentioned
above, including non-fast-forward fully qualified refspecs. This new
function can also easier benefit from optimizations coming later in
this patch series.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: support completing full refs after '--option=refs/<TAB>'
Completing full refs currently only works when the full ref stands on
in its own on the command line, but doesn't work when the current word
to be completed contains a prefix before the full ref, e.g.
'--option=refs/<TAB>' or 'master..refs/bis<TAB>'.
The reason is that __git_refs() looks at the current word to be
completed ($cur) as a whole to decide whether it has to list full (if
it starts with 'refs/') or short refs (otherwise). However, $cur also
holds said '--option=' or 'master..' prefixes, which of course throw
off this decision. Luckily, the default action is to list short refs,
that's why completing short refs happens to work even after a
'master..<TAB>' prefix and similar cases.
Pass only the ref part of the current word to be completed to
__git_refs() as a new positional parameter, so it can make the right
decision even if the whole current word contains some kind of a
prefix.
Make this new parameter the 4. positional parameter and leave the 3.
as an ignored placeholder for now (it will be used later in this patch
series).
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
completion: wrap __git_refs() for better option parsing
__git_refs() currently accepts two optional positional parameters: a
remote and a flag for 'git checkout's tracking DWIMery. To fix a
minor bug, and, more importantly, for faster refs completion, this
series will add three more parameters: a prefix, the current word to
be completed and a suffix, i.e. the options accepted by __gitcomp() &
friends, and will change __git_refs() to list only refs matching that
given current word and to add that given prefix and suffix to the
listed refs.
However, __git_refs() is the helper function that is most likely used
in users' custom completion scriptlets for their own git commands, and
we don't want to break those, so
- we can't change __git_refs()'s default output format, i.e. we
can't by default append a trailing space to every listed ref,
meaning that the suffix parameter containing the default trailing
space would have to be specified on every invocation, and
- we can't change the position of existing positional parameters
either, so there would have to be plenty of set-but-empty
placeholder positional parameters all over the completion script.
Furthermore, with five positional parameters it would be really hard
to remember which position means what.
To keep callsites simple, add the new wrapper function
__git_complete_refs() around __git_refs(), which:
- instead of positional parameters accepts real '--opt=val'-style
options and with minimalistic option parsing translates them to
__git_refs()'s and __gitcomp_nl()'s positional parameters, and
- includes the '__gitcomp_nl "$(__git_refs ...)" ...' command
substitution to make its behavior match its name and the behavior
of other __git_complete_* functions, and to limit future changes
in this series to __git_refs() and this new wrapper function.
Call this wrapper function instead of __git_refs() wherever possible
throughout the completion script, i.e. when __git_refs()'s output is
fed to __gitcomp_nl() right away without further processing, which
means all callsites except a single one in the __git_refs2() helper.
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change "suceed" to "succeed" in a test description. The typo has been
here since the code was originally added in commit ef5a6fb597 ("Add
test-script for git-tag", 2007-06-28).
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Change the test for "git tag -l" to not have an associated TODO
comment saying that it should return non-zero if there's no tags.
This was added in commit ef5a6fb597 ("Add test-script for git-tag",
2007-06-28) when the tests for "tag" were initially added, but at this
point changing this would be inconsistent with how "git tag" is a
synonym for "git tag -l", and would needlessly break external code
that relies on this porcelain command.
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Teach remote-curl to understand push options and to be able to convey
them across HTTP.
Signed-off-by: Brandon Williams <bmwill@google.com> Reviewed-by: Jonathan Nieder <jrnieder@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>