OSDN Git Service

git-core/git.git
9 years agot2026 needs procondition SANITY
Torsten Bögershausen [Tue, 27 Jan 2015 15:38:49 +0000 (16:38 +0100)]
t2026 needs procondition SANITY

When running t0026 as root 'prune directories with unreadable gitdir' fails.
Skip this test if SANITY is not set (the use of POSIXPERM is wrong here)

Signed-off-by: Torsten Bögershausen <tboegi@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-checkout.txt: a note about multiple checkout support for submodules
Nguyễn Thái Ngọc Duy [Sat, 3 Jan 2015 09:41:27 +0000 (16:41 +0700)]
git-checkout.txt: a note about multiple checkout support for submodules

The goal seems to be using multiple checkouts to reduce disk space.
But we have not reached an agreement how things should be. There are a
couple options.

 - You may want to keep $SUB repos elsewhere (perhaps in a central
   place) outside $SUPER. This is also true for nested submodules
   where a superproject may be a submodule of another superproject.

 - You may want to keep all $SUB repos in $SUPER/modules (or some
   other place in $SUPER)

 - We could even push it further and merge all $SUB repos into $SUPER
   instead of storing them separately. But that would at least require
   ref namespace enabled.

On top of that, git-submodule.sh expects $GIT_DIR/config to be
per-worktree, at least for the submodule.* part. Here I think we have
two options, either update config.c to also read
$GIT_DIR/config.worktree (which is per worktree) in addition to
$GIT_DIR/config (shared) and store worktree-specific vars in the new
place, or update git-submodule.sh to read/write submodule.* directly
from $GIT_DIR/config.submodule (per worktree).

These take time to address properly. Meanwhile, make a note to the
user that they should not use multiple worktrees in submodule context.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: add --ignore-other-wortrees
Nguyễn Thái Ngọc Duy [Sat, 3 Jan 2015 09:41:26 +0000 (16:41 +0700)]
checkout: add --ignore-other-wortrees

Noticed-by: Mark Levedahl <mlevedahl@gmail.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: pass whole struct to parse_branchname_arg instead of individual flags
Nguyễn Thái Ngọc Duy [Sat, 3 Jan 2015 09:41:25 +0000 (16:41 +0700)]
checkout: pass whole struct to parse_branchname_arg instead of individual flags

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-common-dir: make "modules/" per-working-directory directory
Max Kirillov [Sun, 30 Nov 2014 08:24:59 +0000 (15:24 +0700)]
git-common-dir: make "modules/" per-working-directory directory

Each working directory of main repository has its own working directory
of submodule, and in most cases they should be checked out to different
revisions. So they should be separated.

It looks logical to make submodule instances in different working
directories to reuse the submodule directory in the common dir of
the main repository, and probably this is how "checkout --to" should
initialize them called on the main repository, but they also should work
fine being completely separated clones.

Testfile t7410-submodule-checkout-to.sh demostrates the behavior.

Signed-off-by: Max Kirillov <max@max630.net>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: do not fail if target is an empty directory
Max Kirillov [Sun, 30 Nov 2014 08:24:58 +0000 (15:24 +0700)]
checkout: do not fail if target is an empty directory

Non-recursive checkout creates empty directpries in place of submodules.
If then I try to "checkout --to" submodules there, it refuses to do so,
because directory already exists.

Fix by allowing checking out to empty directory. Add test and modify the
existing one so that it uses non-empty directory.

Signed-off-by: Max Kirillov <max@max630.net>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agot2025: add a test to make sure grafts is working from a linked checkout
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:57 +0000 (15:24 +0700)]
t2025: add a test to make sure grafts is working from a linked checkout

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: don't require a work tree when checking out into a new one
Dennis Kaarsemaker [Sun, 30 Nov 2014 08:24:56 +0000 (15:24 +0700)]
checkout: don't require a work tree when checking out into a new one

For normal use cases, it does not make sense for 'checkout' to work on
a bare repository, without a worktree. But "checkout --to" is an
exception because it _creates_ a new worktree. Allow this option to
run on bare repositories.

People who check out from a bare repository should remember that
core.logallrefupdates is off by default and it should be turned back
on. `--to` cannot do this automatically behind the user's back because
some user may deliberately want no reflog.

For people interested in repository setup/discovery code,
is_bare_repository_cfg (aka "core.bare") is unchanged by this patch,
which means 'true' by default for bare repos. Fortunately when we get
the repo through a linked checkout, is_bare_repository_cfg is never
used. So all is still good.

[nd: commit message]

Signed-off-by: Dennis Kaarsemaker <dennis@kaarsemaker.net>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit_path(): keep "info/sparse-checkout" per work-tree
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:55 +0000 (15:24 +0700)]
git_path(): keep "info/sparse-checkout" per work-tree

Currently git_path("info/sparse-checkout") resolves to
$GIT_COMMON_DIR/info/sparse-checkout in multiple worktree mode. It
makes more sense for the sparse checkout patterns to be per worktree,
so you can have multiple checkouts with different parts of the tree.

With this, "git checkout --to <new>" on a sparse checkout will create
<new> as a full checkout. Which is expected, it's how a new checkout
is made. The user can reshape the worktree afterwards.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocount-objects: report unused files in $GIT_DIR/worktrees/...
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:54 +0000 (15:24 +0700)]
count-objects: report unused files in $GIT_DIR/worktrees/...

In linked checkouts, borrowed parts like config is taken from
$GIT_COMMON_DIR. $GIT_DIR/config is never used. Report them as
garbage.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogc: support prune --worktrees
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:53 +0000 (15:24 +0700)]
gc: support prune --worktrees

Helped-by: Marc Branchaud <marcnarc@xiplink.com>
Signed-off-by: Marc Branchaud <marcnarc@xiplink.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogc: factor out gc.pruneexpire parsing code
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:52 +0000 (15:24 +0700)]
gc: factor out gc.pruneexpire parsing code

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogc: style change -- no SP before closing parenthesis
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:51 +0000 (15:24 +0700)]
gc: style change -- no SP before closing parenthesis

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: clean up half-prepared directories in --to mode
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:50 +0000 (15:24 +0700)]
checkout: clean up half-prepared directories in --to mode

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: reject if the branch is already checked out elsewhere
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:49 +0000 (15:24 +0700)]
checkout: reject if the branch is already checked out elsewhere

One branch obviously can't be checked out at two places (but detached
heads are ok). Give the user a choice in this case: --detach, -b
new-branch, switch branch in the other checkout first or simply 'cd'
and continue to work there.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoprune: strategies for linked checkouts
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:48 +0000 (15:24 +0700)]
prune: strategies for linked checkouts

(alias R=$GIT_COMMON_DIR/worktrees/<id>)

 - linked checkouts are supposed to keep its location in $R/gitdir up
   to date. The use case is auto fixup after a manual checkout move.

 - linked checkouts are supposed to update mtime of $R/gitdir. If
   $R/gitdir's mtime is older than a limit, and it points to nowhere,
   worktrees/<id> is to be pruned.

 - If $R/locked exists, worktrees/<id> is not supposed to be pruned. If
   $R/locked exists and $R/gitdir's mtime is older than a really long
   limit, warn about old unused repo.

 - "git checkout --to" is supposed to make a hard link named $R/link
   pointing to the .git file on supported file systems to help detect
   the user manually deleting the checkout. If $R/link exists and its
   link count is greated than 1, the repo is kept.

Helped-by: Marc Branchaud <marcnarc@xiplink.com>
Helped-by: Eric Sunshine <sunshine@sunshineco.com>
Helped-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Marc Branchaud <marcnarc@xiplink.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocheckout: support checking out into a new working directory
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:47 +0000 (15:24 +0700)]
checkout: support checking out into a new working directory

"git checkout --to" sets up a new working directory with a .git file
pointing to $GIT_DIR/worktrees/<id>. It then executes "git checkout"
again on the new worktree with the same arguments except "--to" is
taken out. The second checkout execution, which is not contaminated
with any info from the current repository, will actually check out and
everything that normal "git checkout" does.

Helped-by: Marc Branchaud <marcnarc@xiplink.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agouse new wrapper write_file() for simple file writing
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:46 +0000 (15:24 +0700)]
use new wrapper write_file() for simple file writing

This fixes common problems in these code about error handling,
forgetting to close the file handle after fprintf() fails, or not
printing out the error string..

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agowrapper.c: wrapper to open a file, fprintf then close
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:45 +0000 (15:24 +0700)]
wrapper.c: wrapper to open a file, fprintf then close

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agosetup.c: support multi-checkout repo setup
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:44 +0000 (15:24 +0700)]
setup.c: support multi-checkout repo setup

The repo setup procedure is updated to detect $GIT_DIR/commondir and
set $GIT_COMMON_DIR properly.

The core.worktree is ignored when $GIT_COMMON_DIR is set. This is
because the config file is shared in multi-checkout setup, but
checkout directories _are_ different. Making core.worktree effective
in all checkouts mean it's back to a single checkout.

Helped-by: Johannes Sixt <j6t@kdbg.org>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agosetup.c: detect $GIT_COMMON_DIR check_repository_format_gently()
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:43 +0000 (15:24 +0700)]
setup.c: detect $GIT_COMMON_DIR check_repository_format_gently()

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agosetup.c: convert check_repository_format_gently to use strbuf
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:42 +0000 (15:24 +0700)]
setup.c: convert check_repository_format_gently to use strbuf

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agosetup.c: detect $GIT_COMMON_DIR in is_git_directory()
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:41 +0000 (15:24 +0700)]
setup.c: detect $GIT_COMMON_DIR in is_git_directory()

If the file "$GIT_DIR/commondir" exists, it contains the value of
$GIT_COMMON_DIR.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agosetup.c: convert is_git_directory() to use strbuf
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:40 +0000 (15:24 +0700)]
setup.c: convert is_git_directory() to use strbuf

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-stash: avoid hardcoding $GIT_DIR/logs/....
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:39 +0000 (15:24 +0700)]
git-stash: avoid hardcoding $GIT_DIR/logs/....

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years ago*.sh: avoid hardcoding $GIT_DIR/hooks/...
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:38 +0000 (15:24 +0700)]
*.sh: avoid hardcoding $GIT_DIR/hooks/...

If $GIT_COMMON_DIR is set, it should be $GIT_COMMON_DIR/hooks/, not
$GIT_DIR/hooks/. Just let rev-parse --git-path handle it.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit-sh-setup.sh: use rev-parse --git-path to get $GIT_DIR/objects
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:37 +0000 (15:24 +0700)]
git-sh-setup.sh: use rev-parse --git-path to get $GIT_DIR/objects

If $GIT_COMMON_DIR is set, $GIT_OBJECT_DIRECTORY should be
$GIT_COMMON_DIR/objects, not $GIT_DIR/objects. Just let rev-parse
--git-path handle it.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years ago$GIT_COMMON_DIR: a new environment variable
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:36 +0000 (15:24 +0700)]
$GIT_COMMON_DIR: a new environment variable

This variable is intended to support multiple working directories
attached to a repository. Such a repository may have a main working
directory, created by either "git init" or "git clone" and one or more
linked working directories. These working directories and the main
repository share the same repository directory.

In linked working directories, $GIT_COMMON_DIR must be defined to point
to the real repository directory and $GIT_DIR points to an unused
subdirectory inside $GIT_COMMON_DIR. File locations inside the
repository are reorganized from the linked worktree view point:

 - worktree-specific such as HEAD, logs/HEAD, index, other top-level
   refs and unrecognized files are from $GIT_DIR.

 - the rest like objects, refs, info, hooks, packed-refs, shallow...
   are from $GIT_COMMON_DIR (except info/sparse-checkout, but that's
   a separate patch)

Scripts are supposed to retrieve paths in $GIT_DIR with "git rev-parse
--git-path", which will take care of "$GIT_DIR vs $GIT_COMMON_DIR"
business.

The redirection is done by git_path(), git_pathdup() and
strbuf_git_path(). The selected list of paths goes to $GIT_COMMON_DIR,
not the other way around in case a developer adds a new
worktree-specific file and it's accidentally promoted to be shared
across repositories (this includes unknown files added by third party
commands)

The list of known files that belong to $GIT_DIR are:

ADD_EDIT.patch BISECT_ANCESTORS_OK BISECT_EXPECTED_REV BISECT_LOG
BISECT_NAMES CHERRY_PICK_HEAD COMMIT_MSG FETCH_HEAD HEAD MERGE_HEAD
MERGE_MODE MERGE_RR NOTES_EDITMSG NOTES_MERGE_WORKTREE ORIG_HEAD
REVERT_HEAD SQUASH_MSG TAG_EDITMSG fast_import_crash_* logs/HEAD
next-index-* rebase-apply rebase-merge rsync-refs-* sequencer/*
shallow_*

Path mapping is NOT done for git_path_submodule(). Multi-checkouts are
not supported as submodules.

Helped-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agocommit: use SEQ_DIR instead of hardcoding "sequencer"
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:35 +0000 (15:24 +0700)]
commit: use SEQ_DIR instead of hardcoding "sequencer"

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agofast-import: use git_path() for accessing .git dir instead of get_git_dir()
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:34 +0000 (15:24 +0700)]
fast-import: use git_path() for accessing .git dir instead of get_git_dir()

This allows git_path() to redirect info/fast-import to another place
if needed

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoreflog: avoid constructing .lock path with git_path
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:33 +0000 (15:24 +0700)]
reflog: avoid constructing .lock path with git_path

Among pathnames in $GIT_DIR, e.g. "index" or "packed-refs", we want to
automatically and silently map some of them to the $GIT_DIR of the
repository we are borrowing from via $GIT_COMMON_DIR mechanism.  When
we formulate the pathname for its lockfile, we want it to be in the
same location as its final destination.  "index" is not shared and
needs to remain in the borrowing repository, while "packed-refs" is
shared and needs to go to the borrowed repository.

git_path() could be taught about the ".lock" suffix and map
"index.lock" and "packed-refs.lock" the same way their basenames are
mapped, but instead the caller can help by asking where the basename
(e.g. "index") is mapped to git_path() and then appending ".lock"
after the mapping is done.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years ago*.sh: respect $GIT_INDEX_FILE
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:32 +0000 (15:24 +0700)]
*.sh: respect $GIT_INDEX_FILE

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit_path(): be aware of file relocation in $GIT_DIR
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:31 +0000 (15:24 +0700)]
git_path(): be aware of file relocation in $GIT_DIR

We allow the user to relocate certain paths out of $GIT_DIR via
environment variables, e.g. GIT_OBJECT_DIRECTORY, GIT_INDEX_FILE and
GIT_GRAFT_FILE. Callers are not supposed to use git_path() or
git_pathdup() to get those paths. Instead they must use
get_object_directory(), get_index_file() and get_graft_file()
respectively. This is inconvenient and could be missed in review (for
example, there's git_path("objects/info/alternates") somewhere in
sha1_file.c).

This patch makes git_path() and git_pathdup() understand those
environment variables. So if you set GIT_OBJECT_DIRECTORY to /foo/bar,
git_path("objects/abc") should return /foo/bar/abc. The same is done
for the two remaining env variables.

"git rev-parse --git-path" is the wrapper for script use.

This patch kinda reverts a0279e1 (setup_git_env: use git_pathdup
instead of xmalloc + sprintf - 2014-06-19) because using git_pathdup
here would result in infinite recursion:

  setup_git_env() -> git_pathdup("objects") -> .. -> adjust_git_path()
  -> get_object_directory() -> oops, git_object_directory is NOT set
  yet -> setup_git_env()

I wanted to make git_pathdup_literal() that skips adjust_git_path().
But that won't work because later on when $GIT_COMMON_DIR is
introduced, git_pathdup_literal("objects") needs adjust_git_path() to
replace $GIT_DIR with $GIT_COMMON_DIR.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agopath.c: group git_path(), git_pathdup() and strbuf_git_path() together
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:30 +0000 (15:24 +0700)]
path.c: group git_path(), git_pathdup() and strbuf_git_path() together

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agopath.c: rename vsnpath() to do_git_path()
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:29 +0000 (15:24 +0700)]
path.c: rename vsnpath() to do_git_path()

The name vsnpath() gives an impression that this is general path
handling function. It's not. This is the underlying implementation of
git_path(), git_pathdup() and strbuf_git_path() which will prefix
$GIT_DIR in the result string.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogit_snpath(): retire and replace with strbuf_git_path()
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:28 +0000 (15:24 +0700)]
git_snpath(): retire and replace with strbuf_git_path()

In the previous patch, git_snpath() is modified to allocate a new
strbuf buffer because vsnpath() needs that. But that makes it
awkward because git_snpath() receives a pre-allocated buffer from
outside and has to copy data back. Rename it to strbuf_git_path()
and make it receive strbuf directly.

Using git_path() in update_refs_for_switch() which used to call
git_snpath() is safe because that function and all of its callers do
not keep any pointer to the round-robin buffer pool allocated by
get_pathname().

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agopath.c: make get_pathname() call sites return const char *
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:27 +0000 (15:24 +0700)]
path.c: make get_pathname() call sites return const char *

Before the previous commit, get_pathname returns an array of PATH_MAX
length. Even if git_path() and similar functions does not use the
whole array, git_path() caller can, in theory.

After the commit, get_pathname() may return a buffer that has just
enough room for the returned string and git_path() caller should never
write beyond that.

Make git_path(), mkpath() and git_path_submodule() return a const
buffer to make sure callers do not write in it at all.

This could have been part of the previous commit, but the "const"
conversion is too much distraction from the core changes in path.c.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agopath.c: make get_pathname() return strbuf instead of static buffer
Nguyễn Thái Ngọc Duy [Sun, 30 Nov 2014 08:24:26 +0000 (15:24 +0700)]
path.c: make get_pathname() return strbuf instead of static buffer

We've been avoiding PATH_MAX whenever possible. This patch makes
get_pathname() return a strbuf and updates the callers to take
advantage of this. The code is simplified as we no longer need to
worry about buffer overflow.

vsnpath() behavior is changed slightly: previously it always clears
the buffer before writing, now it just appends. Fortunately this is a
static function and all of its callers prepare the buffer properly:
git_path() gets the buffer from get_pathname() which resets the
buffer, the remaining call sites start with STRBUF_INIT'd buffer.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoGit 2.2 v2.2.0
Junio C Hamano [Wed, 26 Nov 2014 21:18:34 +0000 (13:18 -0800)]
Git 2.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoRelNotes: spelling & grammar tweaks
Marc Branchaud [Fri, 21 Nov 2014 23:10:04 +0000 (18:10 -0500)]
RelNotes: spelling & grammar tweaks

Signed-off-by: Marc Branchaud <marcnarc@xiplink.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoGit 2.2.0-rc3 v2.2.0-rc3
Junio C Hamano [Fri, 21 Nov 2014 20:10:56 +0000 (12:10 -0800)]
Git 2.2.0-rc3

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agol10n: remove a superfluous translation for push.c
Jiang Xin [Thu, 20 Nov 2014 08:12:34 +0000 (16:12 +0800)]
l10n: remove a superfluous translation for push.c

Ralf reported that '--recurse-submodules' option in push.c should not be
translated [1].  Before his commit is merged, remove superfluous
translations for push.c.

[1] http://www.spinics.net/lists/git/msg241964.html

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
9 years agol10n: de.po: translate 2 messages
Ralf Thielow [Thu, 20 Nov 2014 06:15:15 +0000 (07:15 +0100)]
l10n: de.po: translate 2 messages

Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
9 years agol10n: de.po: translate 2 new messages
Ralf Thielow [Tue, 18 Nov 2014 18:06:51 +0000 (19:06 +0100)]
l10n: de.po: translate 2 new messages

Signed-off-by: Phillip Sz <phillip.szelat@gmail.com>
Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
9 years agol10n: batch updates for one trivial change
Jiang Xin [Thu, 20 Nov 2014 02:53:48 +0000 (10:53 +0800)]
l10n: batch updates for one trivial change

In order to catch up with the release of Git 2.2.0 final, make a batch
l10n update for the new l10n change brought by commit d52adf1 (trailer:
display a trailer without its trailing newline).

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
9 years agol10n: git.pot: v2.2.0 round 2 (1 updated)
Jiang Xin [Thu, 20 Nov 2014 02:03:10 +0000 (10:03 +0800)]
l10n: git.pot: v2.2.0 round 2 (1 updated)

Generate po/git.pot from v2.2.0-rc2-23-gca0107e for git v2.2.0 l10n
round 2.

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
9 years agoMerge branch 'sv/submitting-final-patch'
Junio C Hamano [Wed, 19 Nov 2014 21:48:01 +0000 (13:48 -0800)]
Merge branch 'sv/submitting-final-patch'

* sv/submitting-final-patch:
  SubmittingPatches: final submission is To: maintainer and CC: list

9 years agoMerge branch 'sn/tutorial-status-output-example'
Junio C Hamano [Wed, 19 Nov 2014 21:47:59 +0000 (13:47 -0800)]
Merge branch 'sn/tutorial-status-output-example'

* sn/tutorial-status-output-example:
  gittutorial: fix output of 'git status'

9 years agoMerge branch 'mh/doc-remote-helper-xref'
Junio C Hamano [Wed, 19 Nov 2014 21:47:55 +0000 (13:47 -0800)]
Merge branch 'mh/doc-remote-helper-xref'

* mh/doc-remote-helper-xref:
  doc: add some crossrefs between manual pages

9 years agoMerge branch 'tb/no-relative-file-url'
Junio C Hamano [Wed, 19 Nov 2014 21:47:53 +0000 (13:47 -0800)]
Merge branch 'tb/no-relative-file-url'

* tb/no-relative-file-url:
  t5705: the file:// URL should be absolute

9 years agoMerge branch 'cc/interpret-trailers'
Junio C Hamano [Wed, 19 Nov 2014 21:47:49 +0000 (13:47 -0800)]
Merge branch 'cc/interpret-trailers'

Small fixes to a new experimental command already in 'master'.

* cc/interpret-trailers:
  trailer: display a trailer without its trailing newline
  trailer: ignore comment lines inside the trailers

9 years agoMerge git://github.com/git-l10n/git-po
Junio C Hamano [Tue, 18 Nov 2014 18:27:46 +0000 (10:27 -0800)]
Merge git://github.com/git-l10n/git-po

* 'master' of git://github.com/git-l10n/git-po:
  l10n: Update Catalan translation

9 years agoMerge branch 'jc/doc-commit-only'
Junio C Hamano [Tue, 18 Nov 2014 18:19:38 +0000 (10:19 -0800)]
Merge branch 'jc/doc-commit-only'

* jc/doc-commit-only:
  Documentation/git-commit: clarify that --only/--include records the working tree contents

9 years agoMerge branch 'ta/tutorial-modernize'
Junio C Hamano [Tue, 18 Nov 2014 18:18:28 +0000 (10:18 -0800)]
Merge branch 'ta/tutorial-modernize'

* ta/tutorial-modernize:
  gittutorial.txt: remove reference to ancient Git version

9 years agoMerge branch 'da/difftool'
Junio C Hamano [Tue, 18 Nov 2014 18:16:54 +0000 (10:16 -0800)]
Merge branch 'da/difftool'

Fix-up to a new feature in 'master'.

* da/difftool:
  difftool: honor --trust-exit-code for builtin tools

9 years agol10n: Update Catalan translation
Alex Henrie [Tue, 18 Nov 2014 03:22:48 +0000 (20:22 -0700)]
l10n: Update Catalan translation

Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
9 years agoMerge branch 'master' of git://github.com/git-l10n/git-po
Junio C Hamano [Mon, 17 Nov 2014 17:28:23 +0000 (09:28 -0800)]
Merge branch 'master' of git://github.com/git-l10n/git-po

* 'master' of git://github.com/git-l10n/git-po:
  l10n: de.po: translate 62 new messages
  l10n: de.po: Fixup one translation
  l10n: de.po: use imperative form for command options

9 years agol10n: de.po: translate 62 new messages
Ralf Thielow [Thu, 30 Oct 2014 08:00:47 +0000 (09:00 +0100)]
l10n: de.po: translate 62 new messages

Translate 62 new messages came from git.pot update in 16742b0
(l10n: git.pot: proposed updates for v2.2.0 (+62)).

Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
9 years agol10n: de.po: Fixup one translation
Stefan Beller [Tue, 23 Sep 2014 12:54:46 +0000 (14:54 +0200)]
l10n: de.po: Fixup one translation

English grammar with German words doesn't make it a German translation. ;)

Signed-off-by: Stefan Beller <stefanbeller@gmail.com>
Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
9 years agodifftool: honor --trust-exit-code for builtin tools
David Aguilar [Fri, 14 Nov 2014 21:33:55 +0000 (13:33 -0800)]
difftool: honor --trust-exit-code for builtin tools

run_merge_tool() was not setting $status, which prevented the
exit code for builtin tools from being forwarded to the caller.

Capture the exit status and add a test to guarantee the behavior.

Reported-by: Adria Farres <14farresa@gmail.com>
Signed-off-by: David Aguilar <davvid@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoGit 2.2.0-rc2 v2.2.0-rc2
Junio C Hamano [Fri, 14 Nov 2014 21:31:15 +0000 (13:31 -0800)]
Git 2.2.0-rc2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agol10n: de.po: use imperative form for command options
Ralf Thielow [Fri, 19 Sep 2014 16:38:13 +0000 (18:38 +0200)]
l10n: de.po: use imperative form for command options

Signed-off-by: Ralf Thielow <ralf.thielow@gmail.com>
9 years agogittutorial: fix output of 'git status'
Stefan Naewe [Thu, 13 Nov 2014 10:40:07 +0000 (10:40 +0000)]
gittutorial: fix output of 'git status'

'git status' doesn't output leading '#'s these days.

Signed-off-by: Stefan Naewe <stefan.naewe@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agot5705: the file:// URL should be absolute
Torsten Bögershausen [Thu, 13 Nov 2014 07:36:07 +0000 (08:36 +0100)]
t5705: the file:// URL should be absolute

The test misused a URL "file://." to mean "relative to here",
which we no longer accept.

In a file:// URL, typically there is no host, and RFC1738 says that
file:///<path> should be used.

Update t5705 to use a working URL.

Reported-by: Michael Blume <blume.mike@gmail.com>
Signed-off-by: Torsten Bögershausen <tboegi@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoSubmittingPatches: final submission is To: maintainer and CC: list
Slavomir Vlcek [Wed, 12 Nov 2014 23:18:39 +0000 (00:18 +0100)]
SubmittingPatches: final submission is To: maintainer and CC: list

In an earlier part there is:

  "re-send it with "To:" set to the maintainer [*1*] and "cc:" the list [*2*]"

for the final submission, but later we see

  "Send it to the list and cc the maintainer."

Fix the later one to match the previous.

Signed-off-by: Slavomir Vlcek <svlc@inventati.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoUpdate draft release notes to 2.2
Junio C Hamano [Wed, 12 Nov 2014 20:12:50 +0000 (12:12 -0800)]
Update draft release notes to 2.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoSync with 'maint'
Junio C Hamano [Wed, 12 Nov 2014 20:13:25 +0000 (12:13 -0800)]
Sync with 'maint'

9 years agoMerge branch 'nd/gitignore-trailing-whitespace' into maint
Junio C Hamano [Wed, 12 Nov 2014 20:13:11 +0000 (12:13 -0800)]
Merge branch 'nd/gitignore-trailing-whitespace' into maint

* nd/gitignore-trailing-whitespace:
  gitignore.txt: fix spelling of "backslash"

9 years agoMerge branch 'jk/fetch-reflog-df-conflict'
Junio C Hamano [Wed, 12 Nov 2014 19:59:58 +0000 (11:59 -0800)]
Merge branch 'jk/fetch-reflog-df-conflict'

Fix-up a test for portability.

* jk/fetch-reflog-df-conflict:
  t1410: fix breakage on case-insensitive filesystems

9 years agodoc: add some crossrefs between manual pages
Max Horn [Tue, 11 Nov 2014 20:17:07 +0000 (21:17 +0100)]
doc: add some crossrefs between manual pages

In particular, git-fast-import and -export link to each
other, and gitremote-helpers links to existing remote
helpers, and vice versa. Also link to fast-import from the
remote helper spec, as this is relevant for remote helpers
using the fast-import format.

Signed-off-by: Max Horn <max@quendi.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogittutorial.txt: remove reference to ancient Git version
Thomas Ackermann [Tue, 11 Nov 2014 19:13:36 +0000 (20:13 +0100)]
gittutorial.txt: remove reference to ancient Git version

Signed-off-by: Thomas Ackermann <th.acker@arcor.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoSync with maint
Junio C Hamano [Tue, 11 Nov 2014 20:45:48 +0000 (12:45 -0800)]
Sync with maint

* maint:

9 years agoMerge branch 'rs/clean-menu-item-defn' into maint
Junio C Hamano [Tue, 11 Nov 2014 18:20:13 +0000 (10:20 -0800)]
Merge branch 'rs/clean-menu-item-defn' into maint

* rs/clean-menu-item-defn:
  clean: use f(void) instead of f() to declare a pointer to a function without arguments

9 years agorun-command: use void to declare that functions take no parameters
René Scharfe [Mon, 10 Nov 2014 21:17:00 +0000 (22:17 +0100)]
run-command: use void to declare that functions take no parameters

Explicitly declare that git_atexit_dispatch() and git_atexit_clear()
take no parameters instead of leaving their parameter list empty and
thus unspecified.

Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoMerge branch 'master' of git://github.com/git-l10n/git-po
Junio C Hamano [Mon, 10 Nov 2014 19:59:30 +0000 (11:59 -0800)]
Merge branch 'master' of git://github.com/git-l10n/git-po

* 'master' of git://github.com/git-l10n/git-po:
  l10n: Updated Bulgarian translation of git (2296t,0f,0u)
  l10n: zh_CN: translations for git v2.2.0-rc0
  l10n: sv.po: Update Swedish translation (2296t0f0u)
  l10n: fr.po (2296t) update for version 2.2.0
  l10n: vi.po: Update new message strings
  l10n: git.pot: v2.2.0 round 1 (62 new, 23 removed)

9 years agoSync with maint
Junio C Hamano [Mon, 10 Nov 2014 19:26:18 +0000 (11:26 -0800)]
Sync with maint

* maint:
  Documentation/config.txt: fix minor typo
  config.txt: fix typo

9 years agoMerge branch 'js/diff-highlight-avoid-sigpipe'
Junio C Hamano [Mon, 10 Nov 2014 19:26:09 +0000 (11:26 -0800)]
Merge branch 'js/diff-highlight-avoid-sigpipe'

* js/diff-highlight-avoid-sigpipe:
  diff-highlight: exit when a pipe is broken

9 years agoDocumentation/config.txt: fix minor typo
Thomas Quinot [Sat, 8 Nov 2014 10:45:39 +0000 (11:45 +0100)]
Documentation/config.txt: fix minor typo

Add a missing article at the beginning of a sentence, and rephrase
slightly.

Signed-off-by: Thomas Quinot <thomas@quinot.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoconfig.txt: fix typo
Nicolas Dermine [Sun, 9 Nov 2014 16:19:33 +0000 (17:19 +0100)]
config.txt: fix typo

Signed-off-by: Nicolas Dermine <nicolas.dermine@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agotrailer: use CHILD_PROCESS_INIT in apply_command()
René Scharfe [Sun, 9 Nov 2014 13:49:58 +0000 (14:49 +0100)]
trailer: use CHILD_PROCESS_INIT in apply_command()

Initialize the struct child_process variable cp at declaration time.
This is shorter, saves a function call and prevents using the variable
before initialization by mistake.

Suggested-by: Jeff King <peff@peff.net>
Signed-off-by: Rene Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agotrailer: display a trailer without its trailing newline
Christian Couder [Sun, 9 Nov 2014 09:23:40 +0000 (10:23 +0100)]
trailer: display a trailer without its trailing newline

Trailers passed to the parse_trailer() function often have
a trailing newline. When erroring out, we should display
the invalid trailer properly, that means without any
trailing newline.

Helped-by: Junio C Hamano <gitster@pobox.com>
Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agotrailer: ignore comment lines inside the trailers
Christian Couder [Sun, 9 Nov 2014 09:23:39 +0000 (10:23 +0100)]
trailer: ignore comment lines inside the trailers

Otherwise trailers that are commented out might be
processed. We would also error out if the comment line
char is also a separator.

This means that comments inside a trailer block will
disappear, but that was already the case anyway.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agot1410: fix breakage on case-insensitive filesystems
Jeff King [Sun, 9 Nov 2014 01:59:18 +0000 (20:59 -0500)]
t1410: fix breakage on case-insensitive filesystems

Two tests recently added to t1410 create branches "a" and
"a/b" to test d/f conflicts on reflogs. Earlier, unrelated
tests in that script create the path "A/B" in the working
tree.  There's no conflict on a case-sensitive filesystem,
but on a case-insensitive one, "git log" will complain that
"a/b" is both a revision and a working tree path.

We could fix this by using a "--" to disambiguate, but we
are probably better off using names that are less confusing
to make it more clear that they are unrelated to the working
tree files.  This patch turns "a/b" into "one/two".

Reported-by: Michael Blume <blume.mike@gmail.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoGit 2.2.0-rc1 v2.2.0-rc1
Junio C Hamano [Fri, 7 Nov 2014 20:01:01 +0000 (12:01 -0800)]
Git 2.2.0-rc1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoDocumentation/git-commit: clarify that --only/--include records the working tree...
Junio C Hamano [Fri, 7 Nov 2014 19:55:40 +0000 (11:55 -0800)]
Documentation/git-commit: clarify that --only/--include records the working tree contents

With the original phrasing, it is possible to misunderstand as if
the contents in the index for only the specified paths are made into
the new commit.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoMerge branch 'maint'
Junio C Hamano [Thu, 6 Nov 2014 18:52:51 +0000 (10:52 -0800)]
Merge branch 'maint'

* maint:
  docs/credential-store: s/--store/--file/

9 years agoMerge branch 'nd/gitignore-trailing-whitespace'
Junio C Hamano [Thu, 6 Nov 2014 18:52:40 +0000 (10:52 -0800)]
Merge branch 'nd/gitignore-trailing-whitespace'

Documentation update.

* nd/gitignore-trailing-whitespace:
  gitignore.txt: fix spelling of "backslash"

9 years agoMerge branch 'tm/line-log-first-parent'
Junio C Hamano [Thu, 6 Nov 2014 18:52:36 +0000 (10:52 -0800)]
Merge branch 'tm/line-log-first-parent'

"git log --first-parent -L..." used to crash.

* tm/line-log-first-parent:
  line-log: fix crash when --first-parent is used

9 years agoMerge branch 'jk/fetch-reflog-df-conflict'
Junio C Hamano [Thu, 6 Nov 2014 18:52:31 +0000 (10:52 -0800)]
Merge branch 'jk/fetch-reflog-df-conflict'

Corner-case bugfixes for "git fetch" around reflog handling.

* jk/fetch-reflog-df-conflict:
  ignore stale directories when checking reflog existence
  fetch: load all default config at startup

9 years agoMerge branch 'rs/use-child-process-init-more'
Junio C Hamano [Thu, 6 Nov 2014 18:52:23 +0000 (10:52 -0800)]
Merge branch 'rs/use-child-process-init-more'

* rs/use-child-process-init-more:
  bundle: split out ref writing from bundle_create
  bundle: split out a helper function to compute and write prerequisites
  bundle: split out a helper function to create pack data
  use child_process_init() to initialize struct child_process variables

9 years agoMerge branch 'jk/cache-tree-protect-from-broken-libgit2'
Junio C Hamano [Thu, 6 Nov 2014 18:51:35 +0000 (10:51 -0800)]
Merge branch 'jk/cache-tree-protect-from-broken-libgit2'

The code to use cache-tree trusted the on-disk data too much
and fell into an infinite loop.

* jk/cache-tree-protect-from-broken-libgit2:
  cache-tree: avoid infinite loop on zero-entry tree

9 years agodocs/credential-store: s/--store/--file/
Jeff King [Thu, 6 Nov 2014 07:40:32 +0000 (02:40 -0500)]
docs/credential-store: s/--store/--file/

The option name "--store" was used early in development, but
never even made it into an applied patch, let alone a
released version of git. I forgot to update the matching
documentation at the time, though.

Noticed-by: Jesse Hopkins <jesse.hopkins@lmco.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agogitignore.txt: fix spelling of "backslash"
Ben North [Tue, 4 Nov 2014 22:18:33 +0000 (22:18 +0000)]
gitignore.txt: fix spelling of "backslash"

Signed-off-by: Ben North <ben@redfrontdoor.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agodiff-highlight: exit when a pipe is broken
John Szakmeister [Tue, 4 Nov 2014 20:01:12 +0000 (15:01 -0500)]
diff-highlight: exit when a pipe is broken

While using diff-highlight with other tools, I have discovered that Python
ignores SIGPIPE by default.  Unfortunately, this also means that tools
attempting to launch a pager under Python--and don't realize this is
happening--means that the subprocess inherits this setting.  In this case, it
means diff-highlight will be launched with SIGPIPE being ignored.  Let's work
with those broken scripts by restoring the default SIGPIPE handler.

Signed-off-by: John Szakmeister <john@szakmeister.net>
Acked-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoDocumentation: typofixes
Thomas Ackermann [Mon, 3 Nov 2014 20:37:07 +0000 (21:37 +0100)]
Documentation: typofixes

In addition to fixing trivial and obvious typos, be careful about
the following points:

 - Spell ASCII, URL and CRC in ALL CAPS;
 - Spell Linux as Capitalized;
 - Do not omit periods in "i.e." and "e.g.".

Signed-off-by: Thomas Ackermann <th.acker@arcor.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoline-log: fix crash when --first-parent is used
Tzvetan Mikov [Tue, 4 Nov 2014 20:33:37 +0000 (12:33 -0800)]
line-log: fix crash when --first-parent is used

line-log tries to access all parents of a commit, but only the first
parent has been loaded if "--first-parent" is specified, resulting
in a crash.

Limit the number of parents to one if "--first-parent" is specified.

Reported-by: Eric N. Vander Weele <ericvw@gmail.com>
Signed-off-by: Tzvetan Mikov <tmikov@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoignore stale directories when checking reflog existence
Jeff King [Tue, 4 Nov 2014 13:24:53 +0000 (08:24 -0500)]
ignore stale directories when checking reflog existence

When we update a ref, we have two rules for whether or not
we actually update the reflog:

  1. If the reflog already exists, we will always append to
     it.

  2. If log_all_ref_updates is set, we will create a new
     reflog file if necessary.

We do the existence check by trying to open the reflog file,
either with or without O_CREAT (depending on log_all_ref_updates).
If it fails, then we check errno to see what happened.

If we were not using O_CREAT and we got ENOENT, the file
doesn't exist, and we return success (there isn't a reflog
already, and we were not told to make a new one).

If we get EISDIR, then there is likely a stale directory
that needs to be removed (e.g., there used to be "foo/bar",
it was deleted, and the directory "foo" was left. Now we
want to create the ref "foo"). If O_CREAT is set, then we
catch this case, try to remove the directory, and retry our
open. So far so good.

But if we get EISDIR and O_CREAT is not set, then we treat
this as any other error, which is not right. Like ENOENT,
EISDIR is an indication that we do not have a reflog, and we
should silently return success (we were not told to create
it). Instead, the current code reports this as an error, and
we fail to update the ref at all.

Note that this is relatively unlikely to happen, as you
would have to have had reflogs turned on, and then later
turned them off (it could also happen due to a bug in fetch,
but that was fixed in the previous commit). However, it's
quite easy to fix: we just need to treat EISDIR like ENOENT
for the non-O_CREAT case, and silently return (note that
this early return means we can also simplify the O_CREAT
case).

Our new tests cover both cases (O_CREAT and non-O_CREAT).
The first one already worked, of course.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agofetch: load all default config at startup
Jeff King [Tue, 4 Nov 2014 13:11:19 +0000 (08:11 -0500)]
fetch: load all default config at startup

When we start the git-fetch program, we call git_config to
load all config, but our callback only processes the
fetch.prune option; we do not chain to git_default_config at
all.

This means that we may not load some core configuration
which will have an effect. For instance, we do not load
core.logAllRefUpdates, which impacts whether or not we
create reflogs in a bare repository.

Note that I said "may" above. It gets even more exciting. If
we have to transfer actual objects as part of the fetch,
then we call fetch_pack as part of the same process. That
function loads its own config, which does chain to
git_default_config, impacting global variables which are
used by the rest of fetch. But if the fetch is a pure ref
update (e.g., a new ref which is a copy of an old one), we
skip fetch_pack entirely. So we get inconsistent results
depending on whether or not we have actual objects to
transfer or not!

Let's just load the core config at the start of fetch, so we
know we have it (we may also load it again as part of
fetch_pack, but that's OK; it's designed to be idempotent).

Our tests check both cases (with and without a pack). We
also check similar behavior for push for good measure, but
it already works as expected.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agoRelNotes/2.2.0.txt: fix minor typos
Matthieu Moy [Mon, 3 Nov 2014 15:12:00 +0000 (16:12 +0100)]
RelNotes/2.2.0.txt: fix minor typos

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
9 years agol10n: Updated Bulgarian translation of git (2296t,0f,0u)
Alexander Shopov [Fri, 17 Oct 2014 19:39:45 +0000 (22:39 +0300)]
l10n: Updated Bulgarian translation of git (2296t,0f,0u)

Signed-off-by: Alexander Shopov <ash@kambanaria.org>