git/git.git
2 years agol10n: fr.po v2.20 rnd 1
Jean-Noël Avila [Wed, 28 Nov 2018 21:16:56 +0000 (22:16 +0100)]
l10n: fr.po v2.20 rnd 1

Signed-off-by: Jean-Noël Avila <jn.avila@free.fr>
2 years agoMerge branch 'master' of git://github.com/git-l10n/git-po
Jiang Xin [Tue, 20 Nov 2018 02:07:25 +0000 (10:07 +0800)]
Merge branch 'master' of git://github.com/git-l10n/git-po

2 years agol10n: git.pot: v2.20.0 round 1 (254 new, 27 removed)
Jiang Xin [Tue, 20 Nov 2018 02:06:16 +0000 (10:06 +0800)]
l10n: git.pot: v2.20.0 round 1 (254 new, 27 removed)

Generate po/git.pot from v2.20.0-rc0-23-gbb75be6cb9 for git v2.20.0 l10n
round 1.

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
2 years agoPrepare for 2.20-rc1
Junio C Hamano [Mon, 19 Nov 2018 07:06:54 +0000 (16:06 +0900)]
Prepare for 2.20-rc1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoMerge branch 'sb/cocci-pending'
Junio C Hamano [Mon, 19 Nov 2018 07:24:41 +0000 (16:24 +0900)]
Merge branch 'sb/cocci-pending'

A coding convention around the Coccinelle semantic patches to have
two classes to ease code migration process has been proposed and
its support has been added to the Makefile.

* sb/cocci-pending:
  coccicheck: introduce 'pending' semantic patches

2 years agoMerge branch 'js/test-git-installed'
Junio C Hamano [Mon, 19 Nov 2018 07:24:41 +0000 (16:24 +0900)]
Merge branch 'js/test-git-installed'

Update the "test installed Git" mode of our test suite to work better.

* js/test-git-installed:
  tests: explicitly use `git.exe` on Windows
  tests: do not require Git to be built when testing an installed Git
  t/lib-gettext: test installed git-sh-i18n if GIT_TEST_INSTALLED is set
  tests: respect GIT_TEST_INSTALLED when initializing repositories
  tests: fix GIT_TEST_INSTALLED's PATH to include t/helper/

2 years agoMerge branch 'dd/poll-dot-h'
Junio C Hamano [Mon, 19 Nov 2018 07:24:41 +0000 (16:24 +0900)]
Merge branch 'dd/poll-dot-h'

A build update.

* dd/poll-dot-h:
  git-compat-util: prefer poll.h to sys/poll.h

2 years agoMerge branch 'tb/print-size-t-with-uintmax-format'
Junio C Hamano [Mon, 19 Nov 2018 07:24:41 +0000 (16:24 +0900)]
Merge branch 'tb/print-size-t-with-uintmax-format'

Code preparation to replace ulong vars with size_t vars where
appropriate.

* tb/print-size-t-with-uintmax-format:
  Upcast size_t variables to uintmax_t when printing

2 years agoMerge branch 'tb/xcurl-off-t'
Junio C Hamano [Mon, 19 Nov 2018 07:24:40 +0000 (16:24 +0900)]
Merge branch 'tb/xcurl-off-t'

The xcurl_off_t() helper function is used to cast size_t to
curl_off_t, but some compilers gave warnings against the code to
ensure the casting is done without wraparound, when size_t is
narrower than curl_off_t.  This warning has been squelched.

* tb/xcurl-off-t:
  remote-curl.c: xcurl_off_t is not portable (on 32 bit platfoms)

2 years agoMerge branch 'nd/format-patch-cover-letter-stat-width'
Junio C Hamano [Mon, 19 Nov 2018 07:24:40 +0000 (16:24 +0900)]
Merge branch 'nd/format-patch-cover-letter-stat-width'

"git format-patch --stat=<width>" can be used to specify the width
used by the diffstat (shown in the cover letter).

* nd/format-patch-cover-letter-stat-width:
  format-patch: respect --stat in cover letter's diffstat

2 years agoMerge branch 'ds/push-squelch-ambig-warning'
Junio C Hamano [Mon, 19 Nov 2018 07:24:40 +0000 (16:24 +0900)]
Merge branch 'ds/push-squelch-ambig-warning'

"git push" used to check ambiguities between object-names and
refnames while processing the list of refs' old and new values,
which was unnecessary (as it knew that it is feeding raw object
names).  This has been optimized out.

* ds/push-squelch-ambig-warning:
  pack-objects: ignore ambiguous object warnings

2 years agoMerge branch 'ab/dynamic-gettext-poison'
Junio C Hamano [Mon, 19 Nov 2018 07:24:39 +0000 (16:24 +0900)]
Merge branch 'ab/dynamic-gettext-poison'

Our testing framework uses a special i18n "poisoned localization"
feature to find messages that ought to stay constant but are
incorrectly marked to be translated.  This feature has been made
into a runtime option (it used to be a compile-time option).

* ab/dynamic-gettext-poison:
  Makefile: ease dynamic-gettext-poison transition
  i18n: make GETTEXT_POISON a runtime option

2 years agoRelNotes: name the release properly
Junio C Hamano [Sun, 18 Nov 2018 23:23:09 +0000 (08:23 +0900)]
RelNotes: name the release properly

In the title, we should state for which version this release notes
document is about.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoMerge branch 'master' of https://github.com/Softcatala/git-po
Jiang Xin [Sun, 18 Nov 2018 13:36:13 +0000 (21:36 +0800)]
Merge branch 'master' of https://github.com/Softcatala/git-po

2 years agoGit 2.20-rc0 v2.20.0-rc0
Junio C Hamano [Sun, 18 Nov 2018 09:24:49 +0000 (18:24 +0900)]
Git 2.20-rc0

Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoMerge branch 'jk/close-duped-fd-before-unlock-for-bundle'
Junio C Hamano [Sun, 18 Nov 2018 09:23:59 +0000 (18:23 +0900)]
Merge branch 'jk/close-duped-fd-before-unlock-for-bundle'

When "git bundle" aborts due to an empty commit ranges
(i.e. resulting in an empty pack), it left a file descriptor to an
lockfile open, which resulted in leftover lockfile on Windows where
you cannot remove a file with an open file descriptor.  This has
been corrected.

* jk/close-duped-fd-before-unlock-for-bundle:
  bundle: dup() output descriptor closer to point-of-use

2 years agoMerge branch 'ab/rebase-in-c-escape-hatch'
Junio C Hamano [Sun, 18 Nov 2018 09:23:59 +0000 (18:23 +0900)]
Merge branch 'ab/rebase-in-c-escape-hatch'

The recently merged "rebase in C" has an escape hatch to use the
scripted version when necessary, but it hasn't been documented,
which has been corrected.

* ab/rebase-in-c-escape-hatch:
  tests: add a special setup where rebase.useBuiltin is off
  rebase doc: document rebase.useBuiltin

2 years agoMerge branch 'js/rebase-am-options'
Junio C Hamano [Sun, 18 Nov 2018 09:23:59 +0000 (18:23 +0900)]
Merge branch 'js/rebase-am-options'

The way "git rebase" parses and forwards the command line options
meant for underlying "git am" has been revamped, which fixed for
options with parameters that were not passed correctly.

* js/rebase-am-options:
  rebase: validate -C<n> and --whitespace=<mode> parameters early
  rebase: really just passthru the `git am` options

2 years agoMerge branch 'sg/ref-filter-wo-repository'
Junio C Hamano [Sun, 18 Nov 2018 09:23:59 +0000 (18:23 +0900)]
Merge branch 'sg/ref-filter-wo-repository'

"git ls-remote --sort=<thing>" can feed an object that is not yet
available into the comparison machinery and segfault, which has
been corrected to check such a request upfront and reject it.

* sg/ref-filter-wo-repository:
  ref-filter: don't look for objects when outside of a repository

2 years agoMerge branch 'nd/doc-extensions'
Junio C Hamano [Sun, 18 Nov 2018 09:23:58 +0000 (18:23 +0900)]
Merge branch 'nd/doc-extensions'

Doc update.

* nd/doc-extensions:
  doc: move extensions.worktreeConfig to the right place

2 years agoMerge branch 'js/fuzz-cxxflags'
Junio C Hamano [Sun, 18 Nov 2018 09:23:58 +0000 (18:23 +0900)]
Merge branch 'js/fuzz-cxxflags'

The build procedure to link for fuzzing test has been made
customizable with a new Makefile variable.

* js/fuzz-cxxflags:
  Makefile: use FUZZ_CXXFLAGS for linking fuzzers

2 years agoMerge branch 'js/mingw-msdn-url'
Junio C Hamano [Sun, 18 Nov 2018 09:23:58 +0000 (18:23 +0900)]
Merge branch 'js/mingw-msdn-url'

The URL to an MSDN page in a comment has been updated.

* js/mingw-msdn-url:
  mingw: replace an obsolete link with the superseding one

2 years agoMerge branch 'js/mingw-create-hard-link'
Junio C Hamano [Sun, 18 Nov 2018 09:23:58 +0000 (18:23 +0900)]
Merge branch 'js/mingw-create-hard-link'

Windows update.

* js/mingw-create-hard-link:
  mingw: use `CreateHardLink()` directly

2 years agoMerge branch 'js/config-sequence'
Junio C Hamano [Sun, 18 Nov 2018 09:23:57 +0000 (18:23 +0900)]
Merge branch 'js/config-sequence'

A sanity check for start-up sequence has been added in the config
API codepath.

* js/config-sequence:
  config: report a bug if git_dir exists without commondir

2 years agoMerge branch 'lj/mingw-pthread-cond'
Junio C Hamano [Sun, 18 Nov 2018 09:23:57 +0000 (18:23 +0900)]
Merge branch 'lj/mingw-pthread-cond'

Code simplification.

* lj/mingw-pthread-cond:
  win32: replace pthread_cond_*() with much simpler code

2 years agoMerge branch 'nd/command-list-gen-fix'
Junio C Hamano [Sun, 18 Nov 2018 09:23:57 +0000 (18:23 +0900)]
Merge branch 'nd/command-list-gen-fix'

Build tweak.

* nd/command-list-gen-fix:
  build: fix broken command-list.h generation with core.autocrlf

2 years agoMerge branch 'ag/p3400-force-checkout'
Junio C Hamano [Sun, 18 Nov 2018 09:23:57 +0000 (18:23 +0900)]
Merge branch 'ag/p3400-force-checkout'

Perf test tweak.

* ag/p3400-force-checkout:
  p3400: replace calls to `git checkout -b' by `git checkout -B'

2 years agoMerge branch 'cb/notes-freeing-always-null-fix'
Junio C Hamano [Sun, 18 Nov 2018 09:23:57 +0000 (18:23 +0900)]
Merge branch 'cb/notes-freeing-always-null-fix'

Code cleanup.

* cb/notes-freeing-always-null-fix:
  builtin/notes: remove unnecessary free

2 years agoMerge branch 'js/rebase-r-and-merge-head'
Junio C Hamano [Sun, 18 Nov 2018 09:23:56 +0000 (18:23 +0900)]
Merge branch 'js/rebase-r-and-merge-head'

Bugfix for the recently graduated "git rebase --rebase-merges".

* js/rebase-r-and-merge-head:
  status: rebase and merge can be in progress at the same time
  built-in rebase --skip/--abort: clean up stale .git/<name> files
  rebase -i: include MERGE_HEAD into files to clean up
  rebase -r: do not write MERGE_HEAD unless needed
  rebase -r: demonstrate bug with conflicting merges

2 years agoMerge branch 'js/apply-recount-allow-noop'
Junio C Hamano [Sun, 18 Nov 2018 09:23:56 +0000 (18:23 +0900)]
Merge branch 'js/apply-recount-allow-noop'

When editing a patch in a "git add -i" session, a hunk could be
made to no-op.  The "git apply" program used to reject a patch with
such a no-op hunk to catch user mistakes, but it is now updated to
explicitly allow a no-op hunk in an edited patch.

* js/apply-recount-allow-noop:
  apply --recount: allow "no-op hunks"

2 years agoMerge branch 'ra/rev-parse-exclude-glob'
Junio C Hamano [Sun, 18 Nov 2018 09:23:56 +0000 (18:23 +0900)]
Merge branch 'ra/rev-parse-exclude-glob'

"rev-parse --exclude=<pattern> --branches=<pattern>" etc. did not
quite work, which has been corrected.

* ra/rev-parse-exclude-glob:
  refs: fix some exclude patterns being ignored
  refs: show --exclude failure with --branches/tags/remotes=glob

2 years agoMerge branch 'js/builtin-rebase-perf-fix'
Junio C Hamano [Sun, 18 Nov 2018 09:23:55 +0000 (18:23 +0900)]
Merge branch 'js/builtin-rebase-perf-fix'

Code clean-up with correction to make the reimplemented "git
rebase" a more faithful rewrite of the original, which also regains
performance.

* js/builtin-rebase-perf-fix:
  built-in rebase: reinstate `checkout -q` behavior where appropriate
  rebase: prepare reset_head() for more flags
  rebase: consolidate clean-up code before leaving reset_head()

2 years agoMerge branch 'js/mailmap'
Junio C Hamano [Sun, 18 Nov 2018 09:23:55 +0000 (18:23 +0900)]
Merge branch 'js/mailmap'

Update the mailmap to unify multiple entries for the authors with
commits since v2.10.

* js/mailmap:
  Update .mailmap

2 years agoMerge branch 'js/rebase-autostash-detach-fix'
Junio C Hamano [Sun, 18 Nov 2018 09:23:55 +0000 (18:23 +0900)]
Merge branch 'js/rebase-autostash-detach-fix'

"git rebase --autostash" did not correctly re-attach the HEAD at times.

* js/rebase-autostash-detach-fix:
  built-in rebase --autostash: leave the current branch alone if possible
  built-in rebase: demonstrate regression with --autostash

2 years agoMerge branch 'ab/range-diff-no-patch'
Junio C Hamano [Sun, 18 Nov 2018 09:23:54 +0000 (18:23 +0900)]
Merge branch 'ab/range-diff-no-patch'

The "--no-patch" option, which can be used to get a high-level
overview without the actual line-by-line patch difference shown, of
the "range-diff" command was earlier broken, which has been
corrected.

* ab/range-diff-no-patch:
  range-diff: make diff option behavior (e.g. --stat) consistent
  range-diff: fix regression in passing along diff options
  range-diff doc: add a section about output stability

2 years agoMerge branch 'jk/verify-sig-merge-into-void'
Junio C Hamano [Sun, 18 Nov 2018 09:23:54 +0000 (18:23 +0900)]
Merge branch 'jk/verify-sig-merge-into-void'

"git merge" and "git pull" that merges into an unborn branch used
to completely ignore "--verify-signatures", which has been
corrected.

* jk/verify-sig-merge-into-void:
  pull: handle --verify-signatures for unborn branch
  merge: handle --verify-signatures for unborn branch
  merge: extract verify_merge_signature() helper

2 years agoMerge branch 'js/mingw-res-rebuild'
Junio C Hamano [Sun, 18 Nov 2018 09:23:53 +0000 (18:23 +0900)]
Merge branch 'js/mingw-res-rebuild'

Windows build update.

* js/mingw-res-rebuild:
  Windows: force-recompile git.res for differing architectures

2 years agoMerge branch 'jk/unused-parameter-fixes'
Junio C Hamano [Sun, 18 Nov 2018 09:23:53 +0000 (18:23 +0900)]
Merge branch 'jk/unused-parameter-fixes'

Various functions have been audited for "-Wunused-parameter" warnings
and bugs in them got fixed.

* jk/unused-parameter-fixes:
  midx: double-check large object write loop
  assert NOARG/NONEG behavior of parse-options callbacks
  parse-options: drop OPT_DATE()
  apply: return -1 from option callback instead of calling exit(1)
  cat-file: report an error on multiple --batch options
  tag: mark "--message" option with NONEG
  show-branch: mark --reflog option as NONEG
  format-patch: mark "--no-numbered" option with NONEG
  status: mark --find-renames option with NONEG
  cat-file: mark batch options with NONEG
  pack-objects: mark index-version option as NONEG
  ls-files: mark exclude options as NONEG
  am: handle --no-patch-format option
  apply: mark include/exclude options as NONEG

2 years agoMerge branch 'jk/curl-ldflags'
Junio C Hamano [Sun, 18 Nov 2018 09:23:53 +0000 (18:23 +0900)]
Merge branch 'jk/curl-ldflags'

The way -lcurl library gets linked has been simplified by taking
advantage of the fact that we can just ask curl-config command how.

* jk/curl-ldflags:
  build: link with curl-defined linker flags

2 years agoMerge branch 'mg/gpg-fingerprint-test'
Junio C Hamano [Sun, 18 Nov 2018 09:23:53 +0000 (18:23 +0900)]
Merge branch 'mg/gpg-fingerprint-test'

Add a few tests for a topic already in 'master'.

* mg/gpg-fingerprint-test:
  t/t7510-signed-commit.sh: add signing subkey to Eris Discordia key
  t/t7510-signed-commit.sh: Add %GP to custom format checks

2 years agoMerge branch 'nd/pthreads'
Junio C Hamano [Sun, 18 Nov 2018 09:23:52 +0000 (18:23 +0900)]
Merge branch 'nd/pthreads'

The codebase has been cleaned up to reduce "#ifndef NO_PTHREADS".

* nd/pthreads:
  Clean up pthread_create() error handling
  read-cache.c: initialize copy_len to shut up gcc 8
  read-cache.c: reduce branching based on HAVE_THREADS
  read-cache.c: remove #ifdef NO_PTHREADS
  pack-objects: remove #ifdef NO_PTHREADS
  preload-index.c: remove #ifdef NO_PTHREADS
  grep: clean up num_threads handling
  grep: remove #ifdef NO_PTHREADS
  attr.c: remove #ifdef NO_PTHREADS
  name-hash.c: remove #ifdef NO_PTHREADS
  index-pack: remove #ifdef NO_PTHREADS
  send-pack.c: move async's #ifdef NO_PTHREADS back to run-command.c
  run-command.h: include thread-utils.h instead of pthread.h
  thread-utils: macros to unconditionally compile pthreads API

2 years agoMerge branch 'ds/reachable-topo-order'
Junio C Hamano [Sun, 18 Nov 2018 09:23:52 +0000 (18:23 +0900)]
Merge branch 'ds/reachable-topo-order'

The revision walker machinery learned to take advantage of the
commit generation numbers stored in the commit-graph file.

* ds/reachable-topo-order:
  t6012: make rev-list tests more interesting
  revision.c: generation-based topo-order algorithm
  commit/revisions: bookkeeping before refactoring
  revision.c: begin refactoring --topo-order logic
  test-reach: add rev-list tests
  test-reach: add run_three_modes method
  prio-queue: add 'peek' operation

2 years agobundle: dup() output descriptor closer to point-of-use
Jeff King [Fri, 16 Nov 2018 09:43:59 +0000 (04:43 -0500)]
bundle: dup() output descriptor closer to point-of-use

When writing a bundle to a file, the bundle code actually creates
"your.bundle.lock" using our lockfile interface. We feed that output
descriptor to a child git-pack-objects via run-command, which has the
quirk that it closes the output descriptor in the parent.

To avoid confusing the lockfile code (which still thinks the descriptor
is valid), we dup() it, and operate on the duplicate.

However, this has a confusing side effect: after the dup() but before we
call pack-objects, we have _two_ descriptors open to the lockfile. If we
call die() during that time, the lockfile code will try to clean up the
partially-written file. It knows to close() the file before unlinking,
since on some platforms (i.e., Windows) the open file would block the
deletion. But it doesn't know about the duplicate descriptor. On
Windows, triggering an error at the right part of the code will result
in the cleanup failing and the lockfile being left in the filesystem.

We can solve this by moving the dup() much closer to start_command(),
shrinking the window in which we have the second descriptor open. It's
easy to place this in such a way that no die() is possible. We could
still die due to a signal in the exact wrong moment, but we already
tolerate races there (e.g., a signal could come before we manage to put
the file on the cleanup list in the first place).

As a bonus, this shields create_bundle() itself from the duplicate-fd
trick, and we can simplify its error handling (note that the lock
rollback now happens unconditionally, but that's OK; it's a noop if we
didn't open the lock in the first place).

The included test uses an empty bundle to cause a failure at the right
spot in the code, because that's easy to trigger (the other likely
errors are write() problems like ENOSPC).  Note that it would already
pass on non-Windows systems (because they are happy to unlink an
already-open file).

Based-on-a-patch-by: Gaël Lhez <gael.lhez@gmail.com>
Signed-off-by: Jeff King <peff@peff.net>
Tested-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agotests: add a special setup where rebase.useBuiltin is off
Ævar Arnfjörð Bjarmason [Wed, 14 Nov 2018 09:15:06 +0000 (09:15 +0000)]
tests: add a special setup where rebase.useBuiltin is off

Add a GIT_TEST_REBASE_USE_BUILTIN=false test mode which is equivalent
to running with rebase.useBuiltin=false. This is needed to spot that
we're not introducing any regressions in the legacy rebase version
while we're carrying both it and the new builtin version.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorebase doc: document rebase.useBuiltin
Ævar Arnfjörð Bjarmason [Wed, 14 Nov 2018 09:15:05 +0000 (09:15 +0000)]
rebase doc: document rebase.useBuiltin

The rebase.useBuiltin variable introduced in 55071ea248 ("rebase:
start implementing it as a builtin", 2018-08-07) was turned on by
default in 5541bd5b8f ("rebase: default to using the builtin rebase",
2018-08-08), but had no documentation.

Let's document it so that users who run into any stability issues with
the C rewrite know there's an escape hatch[1], and make it clear that
needing to turn off builtin rebase means you've found a bug in git.

1. https://public-inbox.org/git/87y39w1wc2.fsf@evledraar.gmail.com/

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agomingw: replace an obsolete link with the superseding one
Johannes Schindelin [Thu, 15 Nov 2018 11:22:40 +0000 (03:22 -0800)]
mingw: replace an obsolete link with the superseding one

The MSDN documentation has been superseded by Microsoft Docs (which is
backed by a repository on GitHub containing many, many files in Markdown
format).

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoMakefile: use FUZZ_CXXFLAGS for linking fuzzers
Josh Steadmon [Wed, 14 Nov 2018 19:41:47 +0000 (11:41 -0800)]
Makefile: use FUZZ_CXXFLAGS for linking fuzzers

OSS-Fuzz requires C++-specific flags to link fuzzers. Passing these in
CFLAGS causes lots of build warnings. Using separate FUZZ_CXXFLAGS
avoids this.

Signed-off-by: Josh Steadmon <steadmon@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agotests: explicitly use `git.exe` on Windows
Johannes Schindelin [Wed, 14 Nov 2018 16:32:11 +0000 (08:32 -0800)]
tests: explicitly use `git.exe` on Windows

On Windows, when we refer to `/an/absolute/path/to/git`, it magically
resolves `git.exe` at that location. Except if something of the name
`git` exists next to that `git.exe`. So if we call `$BUILD_DIR/git`, it
will find `$BUILD_DIR/git.exe` *only* if there is not, say, a directory
called `$BUILD_DIR/git`.

Such a directory, however, exists in Git for Windows when building with
Visual Studio (our Visual Studio project generator defaults to putting
the build files into a directory whose name is the base name of the
corresponding `.exe`).

In the bin-wrappers/* scripts, we already take pains to use `git.exe`
rather than `git`, as this could pick up the wrong thing on Windows
(i.e. if there exists a `git` file or directory in the build directory).

Now we do the same in the tests' start-up code.

This also helps when testing an installed Git, as there might be even
more likely some stray file or directory in the way.

Note: the only way we can record whether the `.exe` suffix is by writing
it to the `GIT-BUILD-OPTIONS` file and sourcing it at the beginning of
`t/test-lib.sh`. This is not a requirement introduced by this patch, but
we move the call to be able to use the `$X` variable that holds the file
extension, if any.

Note also: the many, many calls to `git this` and `git that` are
unaffected, as the regular PATH search will find the `.exe` files on
Windows (and not be confused by a directory of the name `git` that is
in one of the directories listed in the `PATH` variable), while
`/path/to/git` would not, per se, know that it is looking for an
executable and happily prefer such a directory.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agotests: do not require Git to be built when testing an installed Git
Johannes Schindelin [Wed, 14 Nov 2018 16:32:10 +0000 (08:32 -0800)]
tests: do not require Git to be built when testing an installed Git

We really only need the test helpers to be built in the worktree in that
case, but that is not what we test for.

On the other hand it is a perfect opportunity to verify that
`GIT_TEST_INSTALLED` points to a working Git.

So let's test the appropriate Git executable. While at it, also adjust
the error message in the `GIT_TEST_INSTALLED` case.

This patch is best viewed with `-w --patience`.

Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agodoc: move extensions.worktreeConfig to the right place
Nguyễn Thái Ngọc Duy [Wed, 14 Nov 2018 16:02:47 +0000 (17:02 +0100)]
doc: move extensions.worktreeConfig to the right place

All config extensions are described in technical/repository-version.txt.
I made a mistake of adding it in config.txt instead. This patch moves
it back to where it belongs.

Since repository-version.txt is not part of officially generated
documents (it's not even part of DOC_HTML target), it's only visible
to developers who read plain .txt files. Let's include it in
gitrepository-layout.5 for more visibility. Some minor asciidoc fixes
are required in repository-version.txt to make this happen.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoref-filter: don't look for objects when outside of a repository
SZEDER Gábor [Wed, 14 Nov 2018 12:27:25 +0000 (13:27 +0100)]
ref-filter: don't look for objects when outside of a repository

The command 'git ls-remote --sort=authordate <remote>' segfaults when
run outside of a repository, ever since the introduction of its
'--sort' option in 1fb20dfd8e (ls-remote: create '--sort' option,
2018-04-09).

While in general the 'git ls-remote' command can be run outside of a
repository just fine, its '--sort=<key>' option with certain keys does
require access to the referenced objects.  This sorting is implemented
using the generic ref-filter sorting facility, which already handles
missing objects gracefully with the appropriate 'missing object
deadbeef for HEAD' message.  However, being generic means that it
checks replace refs while trying to retrieve an object, and while
doing so it accesses the 'git_replace_ref_base' variable, which has
not been initialized and is still a NULL pointer when outside of a
repository, thus causing the segfault.

Make ref-filter more careful upfront while parsing the format string,
and make it error out when encountering a format atom requiring object
access when we are not in a repository.  Also add a test to ensure
that 'git ls-remote --sort' fails gracefully when executed outside of
a repository.

Reported-by: H.Merijn Brand <h.m.brand@xs4all.nl>
Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoconfig: report a bug if git_dir exists without commondir
Johannes Schindelin [Wed, 14 Nov 2018 13:59:02 +0000 (05:59 -0800)]
config: report a bug if git_dir exists without commondir

This did happen at some stage, and was fixed relatively quickly. Make
sure that we detect very quickly, too, should that happen again.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Reviewed-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorebase: validate -C<n> and --whitespace=<mode> parameters early
Johannes Schindelin [Wed, 14 Nov 2018 16:25:31 +0000 (08:25 -0800)]
rebase: validate -C<n> and --whitespace=<mode> parameters early

It is a good idea to error out early upon seeing, say, `-Cbad`, rather
than starting the rebase only to have the `--am` backend complain later.

Let's do this.

The only options accepting parameters which we pass through to `git am`
(which may, or may not, forward them to `git apply`) are `-C` and
`--whitespace`. The other options we pass through do not accept
parameters, so we do not have to validate them here.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorebase: really just passthru the `git am` options
Johannes Schindelin [Wed, 14 Nov 2018 16:25:29 +0000 (08:25 -0800)]
rebase: really just passthru the `git am` options

Currently, we parse the options intended for `git am` as if we wanted to
handle them in `git rebase`, and then reconstruct them painstakingly to
define the `git_am_opt` variable.

However, there is a much better way (that I was unaware of, at the time
when I mentored Pratik to implement these options): OPT_PASSTHRU_ARGV.
It is intended for exactly this use case, where command-line options
want to be parsed into a separate `argv_array`.

Let's use this feature.

Incidentally, this also allows us to address a bug discovered by Phillip
Wood, where the built-in rebase failed to understand that the `-C`
option takes an optional argument.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agogit-compat-util: prefer poll.h to sys/poll.h
Đoàn Trần Công Danh [Wed, 14 Nov 2018 01:10:43 +0000 (08:10 +0700)]
git-compat-util: prefer poll.h to sys/poll.h

POSIX specifies that <poll.h> is the correct header for poll(2)
whereas <sys/poll.h> is only needed for some old libc.

Let's follow the POSIX way by default.

This effectively eliminates musl's warning:

    warning redirecting incorrect #include <sys/poll.h> to <poll.h>

Signed-off-by: Đoàn Trần Công Danh <congdanhqx@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorange-diff: make diff option behavior (e.g. --stat) consistent
Ævar Arnfjörð Bjarmason [Tue, 13 Nov 2018 18:55:58 +0000 (18:55 +0000)]
range-diff: make diff option behavior (e.g. --stat) consistent

Make the behavior when diff options (e.g. "--stat") are passed
consistent with how "diff" behaves.

Before 73a834e9e2 ("range-diff: relieve callers of low-level
configuration burden", 2018-07-22) running range-diff with "--stat"
would produce stat output and the diff output, as opposed to how
"diff" behaves where once "--stat" is specified "--patch" also needs
to be provided to emit the patch output.

As noted in a previous change ("range-diff doc: add a section about
output stability", 2018-11-07) the "--stat" output with "range-diff"
is useless at the moment.

But we should behave consistently with "diff" in anticipation of such
output being useful in the future, because it would make for confusing
UI if "diff" and "range-diff" behaved differently when it came to how
they interpret diff options.

The new behavior is also consistent with the existing documentation
added in ba931edd28 ("range-diff: populate the man page",
2018-08-13). See "[...]also accepts the regular diff options[...]" in
git-range-diff(1).

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agowin32: replace pthread_cond_*() with much simpler code
Loo Rong Jie [Tue, 13 Nov 2018 18:52:35 +0000 (10:52 -0800)]
win32: replace pthread_cond_*() with much simpler code

The Win32 CONDITION_VARIABLE has better performance and is easier to
maintain, as the code is a lot shorter now (the semantics of the
CONDITION_VARIABLE matches the pthread_cond_t very well).

Note: CONDITION_VARIABLE is not available in Windows XP and below,
but the declared minimal Windows version required to build and run
Git for Windows is Windows Vista (which is also beyond its
end-of-life, but for less long than Windows XP), so that's okay.

Signed-off-by: Loo Rong Jie <loorongjie@gmail.com>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agomingw: use `CreateHardLink()` directly
Johannes Schindelin [Tue, 13 Nov 2018 14:37:02 +0000 (06:37 -0800)]
mingw: use `CreateHardLink()` directly

The function `CreateHardLink()` is available in all supported Windows
versions (even since Windows XP), so there is no more need to resolve it
at runtime.

Helped-by: Max Kirillov <max@max630.net>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agot/lib-gettext: test installed git-sh-i18n if GIT_TEST_INSTALLED is set
Johannes Schindelin [Mon, 12 Nov 2018 13:48:36 +0000 (05:48 -0800)]
t/lib-gettext: test installed git-sh-i18n if GIT_TEST_INSTALLED is set

It makes very, very little sense to test the built git-sh-i18n when the
user asked specifically to test another one.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agotests: respect GIT_TEST_INSTALLED when initializing repositories
Johannes Schindelin [Mon, 12 Nov 2018 13:48:34 +0000 (05:48 -0800)]
tests: respect GIT_TEST_INSTALLED when initializing repositories

It really makes very, very little sense to use a different git
executable than the one the caller indicated via setting the environment
variable GIT_TEST_INSTALLED.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agotests: fix GIT_TEST_INSTALLED's PATH to include t/helper/
Johannes Schindelin [Mon, 12 Nov 2018 13:48:33 +0000 (05:48 -0800)]
tests: fix GIT_TEST_INSTALLED's PATH to include t/helper/

We really need to be able to find the test helpers... Really. This
change was forgotten when we moved the test helpers into t/helper/

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agococcicheck: introduce 'pending' semantic patches
SZEDER Gábor [Sat, 10 Nov 2018 00:10:52 +0000 (16:10 -0800)]
coccicheck: introduce 'pending' semantic patches

Teach `make coccicheck` to avoid patches named "*.pending.cocci" and
handle them separately in a new `make coccicheck-pending` instead.
This means that we can separate "critical" patches from "FYI" patches.
The former target can continue causing Travis to fail its static
analysis job, while the latter can let us keep an eye on ongoing
(pending) transitions without them causing too much fallout.

Document the intended use-cases around these two targets.
As the process around the pending patches is not yet fully explored,
leave that out.

Signed-off-by: SZEDER Gábor <szeder.dev@gmail.com>
Based-on-work-by: SZEDER Gábor <szeder.dev@gmail.com>
Signed-off-by: Stefan Beller <sbeller@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoTenth batch for 2.20
Junio C Hamano [Tue, 13 Nov 2018 13:13:52 +0000 (22:13 +0900)]
Tenth batch for 2.20

Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agoMerge branch 'nd/complete-format-patch'
Junio C Hamano [Tue, 13 Nov 2018 13:37:28 +0000 (22:37 +0900)]
Merge branch 'nd/complete-format-patch'

The support for format-patch (and send-email) by the command-line
completion script (in contrib/) has been simplified a bit.

* nd/complete-format-patch:
  completion: use __gitcomp_builtin for format-patch

2 years agoMerge branch 'nd/tree-walk-path-exclusion'
Junio C Hamano [Tue, 13 Nov 2018 13:37:28 +0000 (22:37 +0900)]
Merge branch 'nd/tree-walk-path-exclusion'

Pathspec matching against a tree object were buggy when negative
pathspec elements were involved, which has been fixed.

* nd/tree-walk-path-exclusion:
  tree-walk.c: fix overoptimistic inclusion in :(exclude) matching

2 years agoMerge branch 'sg/travis-install-dependencies'
Junio C Hamano [Tue, 13 Nov 2018 13:37:27 +0000 (22:37 +0900)]
Merge branch 'sg/travis-install-dependencies'

The procedure to install dependencies before testing at Travis CI
is getting revamped for both simplicity and flexibility, taking
advantage of the recent move to the vm-based environment.

* sg/travis-install-dependencies:
  travis-ci: install packages in 'ci/install-dependencies.sh'

2 years agoMerge branch 'bp/add-diff-files-optim'
Junio C Hamano [Tue, 13 Nov 2018 13:37:27 +0000 (22:37 +0900)]
Merge branch 'bp/add-diff-files-optim'

"git add" needs to internally run "diff-files" equivalent, and the
codepath learned the same optimization as "diff-files" has to run
lstat(2) in parallel to find which paths have been updated in the
working tree.

* bp/add-diff-files-optim:
  add: speed up cmd_add() by utilizing read_cache_preload()

2 years agoMerge branch 'jk/xdiff-interface'
Junio C Hamano [Tue, 13 Nov 2018 13:37:27 +0000 (22:37 +0900)]
Merge branch 'jk/xdiff-interface'

The interface into "xdiff" library used to discover the offset and
size of a generated patch hunk by first formatting it into the
textual hunk header "@@ -n,m +k,l @@" and then parsing the numbers
out.  A new interface has been introduced to allow callers a more
direct access to them.

* jk/xdiff-interface:
  xdiff-interface: drop parse_hunk_header()
  range-diff: use a hunk callback
  diff: convert --check to use a hunk callback
  combine-diff: use an xdiff hunk callback
  diff: use hunk callback for word-diff
  diff: discard hunk headers for patch-ids earlier
  diff: avoid generating unused hunk header lines
  xdiff-interface: provide a separate consume callback for hunks
  xdiff: provide a separate emit callback for hunks

2 years agoMerge branch 'jk/misc-unused-fixes'
Junio C Hamano [Tue, 13 Nov 2018 13:37:26 +0000 (22:37 +0900)]
Merge branch 'jk/misc-unused-fixes'

Assorted fixes for bugs found while auditing -Wunused-parameter
warnings.

* jk/misc-unused-fixes:
  approxidate: fix NULL dereference in date_time()
  pathspec: handle non-terminated strings with :(attr)
  approxidate: handle pending number for "specials"
  rev-list: handle flags for --indexed-objects

2 years agoMerge branch 'nd/per-worktree-ref-iteration'
Junio C Hamano [Tue, 13 Nov 2018 13:37:26 +0000 (22:37 +0900)]
Merge branch 'nd/per-worktree-ref-iteration'

The code to traverse objects for reachability, used to decide what
objects are unreferenced and expendable, have been taught to also
consider per-worktree refs of other worktrees as starting points to
prevent data loss.

* nd/per-worktree-ref-iteration:
  git-worktree.txt: correct linkgit command name
  reflog expire: cover reflog from all worktrees
  fsck: check HEAD and reflog from other worktrees
  fsck: move fsck_head_link() to get_default_heads() to avoid some globals
  revision.c: better error reporting on ref from different worktrees
  revision.c: correct a parameter name
  refs: new ref types to make per-worktree refs visible to all worktrees
  Add a place for (not) sharing stuff between worktrees
  refs.c: indent with tabs, not spaces

2 years agoMerge branch 'bp/refresh-index-using-preload'
Junio C Hamano [Tue, 13 Nov 2018 13:37:25 +0000 (22:37 +0900)]
Merge branch 'bp/refresh-index-using-preload'

The helper function to refresh the cached stat information in the
in-core index has learned to perform the lstat() part of the
operation in parallel on multi-core platforms.

* bp/refresh-index-using-preload:
  refresh_index: remove unnecessary calls to preload_index()
  speed up refresh_index() by utilizing preload_index()

2 years agoMerge branch 'ag/rebase-i-in-c'
Junio C Hamano [Tue, 13 Nov 2018 13:37:25 +0000 (22:37 +0900)]
Merge branch 'ag/rebase-i-in-c'

Code clean-up for a topic already in 'master'.

* ag/rebase-i-in-c:
  sequencer.c: remove a stray semicolon

2 years agoMerge branch 'al/send-email-auto-cte-fixup'
Junio C Hamano [Tue, 13 Nov 2018 13:37:25 +0000 (22:37 +0900)]
Merge branch 'al/send-email-auto-cte-fixup'

"git send-email --transfer-encoding=..." in recent versions of Git
sometimes produced an empty "Content-Transfer-Encoding:" header,
which has been corrected.

* al/send-email-auto-cte-fixup:
  send-email: avoid empty transfer encoding header

2 years agoMerge branch 'ds/add-missing-tags'
Junio C Hamano [Tue, 13 Nov 2018 13:37:24 +0000 (22:37 +0900)]
Merge branch 'ds/add-missing-tags'

The history traversal used to implement the tag-following has been
optimized by introducing a new helper.

* ds/add-missing-tags:
  remote: make add_missing_tags() linear
  test-reach: test get_reachable_subset
  commit-reach: implement get_reachable_subset

2 years agoMerge branch 'sh/mingw-safer-compat-poll'
Junio C Hamano [Tue, 13 Nov 2018 13:37:24 +0000 (22:37 +0900)]
Merge branch 'sh/mingw-safer-compat-poll'

Windows fix.

* sh/mingw-safer-compat-poll:
  poll: use GetTickCount64() to avoid wrap-around issues

2 years agoMerge branch 'js/rebase-p-tests'
Junio C Hamano [Tue, 13 Nov 2018 13:37:24 +0000 (22:37 +0900)]
Merge branch 'js/rebase-p-tests'

In preparation to the day when we can deprecate and remove the
"rebase -p", make sure we can skip and later remove tests for
it.

* js/rebase-p-tests:
  tests: optionally skip `git rebase -p` tests
  t3418: decouple test cases from a previous `rebase -p` test case
  t3404: decouple some test cases from outcomes of previous test cases

2 years agoMerge branch 'pw/am-rebase-read-author-script'
Junio C Hamano [Tue, 13 Nov 2018 13:37:23 +0000 (22:37 +0900)]
Merge branch 'pw/am-rebase-read-author-script'

Unify code to read the author-script used in "git am" and the
commands that use the sequencer machinery, e.g. "git rebase -i".

* pw/am-rebase-read-author-script:
  sequencer: use read_author_script()
  add read_author_script() to libgit
  am: rename read_author_script()
  am: improve author-script error reporting
  am: don't die in read_author_script()

2 years agoMerge branch 'jc/war-on-string-list'
Junio C Hamano [Tue, 13 Nov 2018 13:37:23 +0000 (22:37 +0900)]
Merge branch 'jc/war-on-string-list'

Replace three string-list instances used as look-up tables in "git
fetch" with hashmaps.

* jc/war-on-string-list:
  fetch: replace string-list used as a look-up table with a hashmap

2 years agoMerge branch 'ag/rev-parse-all-exclude-fix'
Junio C Hamano [Tue, 13 Nov 2018 13:37:23 +0000 (22:37 +0900)]
Merge branch 'ag/rev-parse-all-exclude-fix'

"git rev-parse --exclude=* --branches --branches"  (i.e. first
saying "add only things that do not match '*' out of all branches"
and then adding all branches, without any exclusion this time")
worked as expected, but "--exclude=* --all --all" did not work the
same way, which has been fixed.

* ag/rev-parse-all-exclude-fix:
  rev-parse: clear --exclude list after 'git rev-parse --all'

2 years agoMerge branch 'jt/tighten-fetch-proto-v2-response'
Junio C Hamano [Tue, 13 Nov 2018 13:37:22 +0000 (22:37 +0900)]
Merge branch 'jt/tighten-fetch-proto-v2-response'

"git fetch" was a bit loose in parsing resposes from the other side
when talking over the protocol v2.

* jt/tighten-fetch-proto-v2-response:
  fetch-pack: be more precise in parsing v2 response

2 years agoMerge branch 'ao/submodule-wo-gitmodules-checked-out'
Junio C Hamano [Tue, 13 Nov 2018 13:37:22 +0000 (22:37 +0900)]
Merge branch 'ao/submodule-wo-gitmodules-checked-out'

The submodule support has been updated to read from the blob at
HEAD:.gitmodules when the .gitmodules file is missing from the
working tree.

* ao/submodule-wo-gitmodules-checked-out:
  t/helper: add test-submodule-nested-repo-config
  submodule: support reading .gitmodules when it's not in the working tree
  submodule: add a helper to check if it is safe to write to .gitmodules
  t7506: clean up .gitmodules properly before setting up new scenario
  submodule: use the 'submodule--helper config' command
  submodule--helper: add a new 'config' subcommand
  t7411: be nicer to future tests and really clean things up
  t7411: merge tests 5 and 6
  submodule: factor out a config_set_in_gitmodules_file_gently function
  submodule: add a print_config_from_gitmodules() helper

2 years agoMerge branch 'nb/worktree-api-doc'
Junio C Hamano [Tue, 13 Nov 2018 13:37:21 +0000 (22:37 +0900)]
Merge branch 'nb/worktree-api-doc'

Code readability fix.

* nb/worktree-api-doc:
  worktree: rename is_worktree_locked to worktree_lock_reason
  worktree: update documentation for lock_reason and lock_reason_valid

2 years agoMerge branch 'ma/sequencer-do-reset-saner-loop-termination'
Junio C Hamano [Tue, 13 Nov 2018 13:37:21 +0000 (22:37 +0900)]
Merge branch 'ma/sequencer-do-reset-saner-loop-termination'

Code readability fix.

* ma/sequencer-do-reset-saner-loop-termination:
  sequencer: break out of loop explicitly

2 years agoMerge branch 'js/mingw-utf8-env'
Junio C Hamano [Tue, 13 Nov 2018 13:37:21 +0000 (22:37 +0900)]
Merge branch 'js/mingw-utf8-env'

Windows fix.

* js/mingw-utf8-env:
  mingw: reencode environment variables on the fly (UTF-16 <-> UTF-8)
  t7800: fix quoting

2 years agoMerge branch 'js/mingw-perl5lib'
Junio C Hamano [Tue, 13 Nov 2018 13:37:20 +0000 (22:37 +0900)]
Merge branch 'js/mingw-perl5lib'

Windows fix.

* js/mingw-perl5lib:
  mingw: unset PERL5LIB by default
  config: move Windows-specific config settings into compat/mingw.c
  config: allow for platform-specific core.* config settings
  config: rename `dummy` parameter to `cb` in git_default_config()

2 years agoMerge branch 'js/mingw-isatty-and-dup2'
Junio C Hamano [Tue, 13 Nov 2018 13:37:20 +0000 (22:37 +0900)]
Merge branch 'js/mingw-isatty-and-dup2'

Windows fix.

* js/mingw-isatty-and-dup2:
  mingw: fix isatty() after dup2()

2 years agoMerge branch 'ab/pack-tests-cleanup'
Junio C Hamano [Tue, 13 Nov 2018 13:37:20 +0000 (22:37 +0900)]
Merge branch 'ab/pack-tests-cleanup'

A couple of tests used to leave the repository in a state that is
deliberately corrupt, which have been corrected.

* ab/pack-tests-cleanup:
  index-pack tests: don't leave test repo dirty at end
  pack-objects tests: don't leave test .git corrupt at end
  pack-objects test: modernize style

2 years agoMerge branch 'ds/test-multi-pack-index'
Junio C Hamano [Tue, 13 Nov 2018 13:37:19 +0000 (22:37 +0900)]
Merge branch 'ds/test-multi-pack-index'

Tests for the recently introduced multi-pack index machinery.

* ds/test-multi-pack-index:
  packfile: close multi-pack-index in close_all_packs
  multi-pack-index: define GIT_TEST_MULTI_PACK_INDEX
  midx: close multi-pack-index on repack
  midx: fix broken free() in close_midx()

2 years agoMerge branch 'nd/wildmatch-double-asterisk'
Junio C Hamano [Tue, 13 Nov 2018 13:37:19 +0000 (22:37 +0900)]
Merge branch 'nd/wildmatch-double-asterisk'

A pattern with '**' that does not have a slash on either side used
to be an invalid one, but the code now treats such double-asterisks
the same way as two normal asterisks that happen to be adjacent to
each other.

* nd/wildmatch-double-asterisk:
  wildmatch: change behavior of "foo**bar" in WM_PATHNAME mode

2 years agoMerge branch 'nd/per-worktree-config'
Junio C Hamano [Tue, 13 Nov 2018 13:37:18 +0000 (22:37 +0900)]
Merge branch 'nd/per-worktree-config'

A fourth class of configuration files (in addition to the
traditional "system wide", "per user in the $HOME directory" and
"per repository in the $GIT_DIR/config") has been introduced so
that different worktrees that share the same repository (hence the
same $GIT_DIR/config file) can use different customization.

* nd/per-worktree-config:
  worktree: add per-worktree config files
  t1300: extract and use test_cmp_config()

2 years agoMerge branch 'jk/stream-pack-non-delta-clarification'
Junio C Hamano [Tue, 13 Nov 2018 13:37:18 +0000 (22:37 +0900)]
Merge branch 'jk/stream-pack-non-delta-clarification'

Additional comment on a tricky piece of code to help developers.

* jk/stream-pack-non-delta-clarification:
  read_istream_pack_non_delta(): document input handling

2 years agoMerge branch 'jk/proto-v2-ref-prefix-fix'
Junio C Hamano [Tue, 13 Nov 2018 13:37:17 +0000 (22:37 +0900)]
Merge branch 'jk/proto-v2-ref-prefix-fix'

"git ls-remote $there foo" was broken by recent update for the
protocol v2 and stopped showing refs that match 'foo' that are not
refs/{heads,tags}/foo, which has been fixed.

* jk/proto-v2-ref-prefix-fix:
  ls-remote: pass heads/tags prefixes to transport
  ls-remote: do not send ref prefixes for patterns

2 years agoMerge branch 'jk/detect-truncated-zlib-input'
Junio C Hamano [Tue, 13 Nov 2018 13:37:17 +0000 (22:37 +0900)]
Merge branch 'jk/detect-truncated-zlib-input'

A regression in Git 2.12 era made "git fsck" fall into an infinite
loop while processing truncated loose objects.

* jk/detect-truncated-zlib-input:
  cat-file: handle streaming failures consistently
  check_stream_sha1(): handle input underflow
  t1450: check large blob in trailing-garbage test

2 years agoMerge branch 'nd/config-split'
Junio C Hamano [Tue, 13 Nov 2018 13:37:16 +0000 (22:37 +0900)]
Merge branch 'nd/config-split'

Split the overly large Documentation/config.txt file into million
little pieces.  This potentially allows each individual piece
included into the manual page of the command it affects more easily.

* nd/config-split: (81 commits)
  config.txt: remove config/dummy.txt
  config.txt: move worktree.* to a separate file
  config.txt: move web.* to a separate file
  config.txt: move versionsort.* to a separate file
  config.txt: move user.* to a separate file
  config.txt: move url.* to a separate file
  config.txt: move uploadpack.* to a separate file
  config.txt: move uploadarchive.* to a separate file
  config.txt: move transfer.* to a separate file
  config.txt: move tag.* to a separate file
  config.txt: move submodule.* to a separate file
  config.txt: move stash.* to a separate file
  config.txt: move status.* to a separate file
  config.txt: move splitIndex.* to a separate file
  config.txt: move showBranch.* to a separate file
  config.txt: move sequencer.* to a separate file
  config.txt: move sendemail-config.txt to config/
  config.txt: move reset.* to a separate file
  config.txt: move rerere.* to a separate file
  config.txt: move repack.* to a separate file
  ...

2 years agoformat-patch: respect --stat in cover letter's diffstat
Nguyễn Thái Ngọc Duy [Sat, 10 Nov 2018 05:46:24 +0000 (06:46 +0100)]
format-patch: respect --stat in cover letter's diffstat

Commit 43662b23ab (format-patch: keep cover-letter diffstat wrapped in
72 columns - 2018-01-24) uncondtionally sets stat width to 72 when
generating diffstat for the cover letter, ignoring --stat from command
line. But it should only do so when stat width is still default
(i.e. stat_width == 0).

In order to fix this, we should only set stat_width if stat_width is
zero. But it will never be. Commit 071dd0ba43 (format-patch: reduce
patch diffstat width to 72 - 2018-02-01) makes sure that default stat
width will be 72 (ignoring $COLUMNS, but could still be overriden by
--stat). So all we need to do here is drop the assignment.

Reported-by: Laszlo Ersek <lersek@redhat.com>
Helped-by: Leif Lindholm <leif.lindholm@linaro.org>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agobuilt-in rebase: reinstate `checkout -q` behavior where appropriate
Johannes Schindelin [Mon, 12 Nov 2018 11:44:32 +0000 (03:44 -0800)]
built-in rebase: reinstate `checkout -q` behavior where appropriate

When we converted a `git checkout -q $onto^0` call to use
`reset_head()`, we inadvertently incurred a change from a twoway_merge
to a oneway_merge, as if we wanted a `git reset --hard` instead.

This has performance ramifications under certain, though, as the
oneway_merge needs to lstat() every single index entry whereas
twoway_merge does not.

So let's go back to the old behavior.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorebase: prepare reset_head() for more flags
Johannes Schindelin [Mon, 12 Nov 2018 11:44:31 +0000 (03:44 -0800)]
rebase: prepare reset_head() for more flags

Currently, we only accept the flag indicating whether the HEAD should be
detached not. In the next commit, we want to introduce another flag: to
toggle between emulating `reset --hard` vs `checkout -q`.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorebase: consolidate clean-up code before leaving reset_head()
Johannes Schindelin [Mon, 12 Nov 2018 11:44:30 +0000 (03:44 -0800)]
rebase: consolidate clean-up code before leaving reset_head()

The same clean-up code is repeated quite a few times; Let's DRY up the
code some.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorefs: fix some exclude patterns being ignored
Rafael Ascensão [Mon, 12 Nov 2018 13:25:44 +0000 (13:25 +0000)]
refs: fix some exclude patterns being ignored

`--exclude` from rev-list and rev-parse fails to exclude references if
the next `--branches`, `--tags` or `--remotes` use the optional
inclusive glob because those options are implemented as particular cases
of `--glob=`, which itself requires that exclude patterns begin with
'refs/'.

But it makes sense for `--branches=glob` and friends to be aware that
exclusions patterns for them shouldn't be 'refs/<type>/' prefixed, the
same way exclude patterns for `--branches` and friends (without the
optional glob) already are.

Let's record in 'refs.c:struct ref_filter' which context the exclude
pattern is tied to, so refs.c:filter_refs() can decide if it should
ignore the prefix when trying to match.

Signed-off-by: Rafael Ascensão <rafa.almas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
2 years agorefs: show --exclude failure with --branches/tags/remotes=glob
Rafael Ascensão [Mon, 12 Nov 2018 13:25:43 +0000 (13:25 +0000)]
refs: show --exclude failure with --branches/tags/remotes=glob

The documentation of `--exclude=` option from rev-list and rev-parse
explicitly states that exclude patterns *should not* start with 'refs/'
when used with `--branches`, `--tags` or `--remotes`.

However, following this advice results in refereces not being excluded
if the next `--branches`, `--tags`, `--remotes` use the optional
inclusive glob.

Demonstrate this failure.

Signed-off-by: Rafael Ascensão <rafa.almas@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>