git.git
18 years agoMerge with gitk
Junio C Hamano [Thu, 18 Aug 2005 04:09:15 +0000 (21:09 -0700)]
Merge with gitk

18 years ago[PATCH] Assorted changes to glossary
Johannes Schindelin [Thu, 18 Aug 2005 01:20:08 +0000 (03:20 +0200)]
[PATCH] Assorted changes to glossary

Based on the discussion on the git list, here are some important changes
to the glossary. (There is no cache, but an index. Use "object name"
rather than "SHA1". Reorder. Clarify.)

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAllow graph lines to jump through hyperspace.
Paul Mackerras [Wed, 17 Aug 2005 23:30:10 +0000 (09:30 +1000)]
Allow graph lines to jump through hyperspace.

When the graph gets too wide (as defined by the maxwidth variable,
which can be set in ~/.gitk), we can now terminate graph lines with
an arrow pointing downwards, and reintroduce them later with an
arrow pointing upwards when we need them.  This makes the graph much
less cluttered on large repositories such as the linux kernel.

Unfortunately this has made it slower; it takes about 10 seconds
user time on the linux-2.6 repository on my machine now, compared
to 6 seconds before.  I'll have to work on optimizing that.  Also
on the todo list are making the arrow heads active (so if you click
on them you jump to the other end) and improving the placement of
the null entry.

18 years agoMake rebase script saner.
Junio C Hamano [Wed, 17 Aug 2005 22:19:57 +0000 (15:19 -0700)]
Make rebase script saner.

It did not check to see if the working tree was clean and matched
the commit we were starting out as, resulting in the initial rebased
commit including whatever dirty state the working tree has had.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake sure alternates are carried over from the original repository.
Junio C Hamano [Wed, 17 Aug 2005 22:18:41 +0000 (15:18 -0700)]
Make sure alternates are carried over from the original repository.

When we create a cheap local clone by pointing at the object databse
of the original repository, we forgot to take the alternates the original
repository might have had into account.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix git-commit without paths.
Junio C Hamano [Wed, 17 Aug 2005 22:17:03 +0000 (15:17 -0700)]
Fix git-commit without paths.

The earlier one to grab output from diff-files --name-only has a grave
bug that when no paths are given it ended up doing the equivalent of
"git-commit --all", which was not what I intended.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Make git-update-cache take relative pathnames
Linus Torvalds [Wed, 17 Aug 2005 20:32:22 +0000 (13:32 -0700)]
[PATCH] Make git-update-cache take relative pathnames

This also makes "./filename" acceptable as a side effect, since the
pathname normalization handles that too.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Export relative path handling "prefix_path()" function
Linus Torvalds [Wed, 17 Aug 2005 20:31:49 +0000 (13:31 -0700)]
[PATCH] Export relative path handling "prefix_path()" function

Not all programs necessarily have a pathspec array of pathnames, some of
them (like git-update-cache) want to do things one file at a time.  So
export the single-path interface too.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] git-cvsimport - remove hardcoded reference to origin
Martin Langhoff [Tue, 16 Aug 2005 21:27:09 +0000 (09:27 +1200)]
[PATCH] git-cvsimport - remove hardcoded reference to origin

... in the newly introduced merge detection code.

Signed-off-by: Martin Langhoff <martin.langhoff@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add merge detection to git-cvsimport
Martin Langhoff [Tue, 16 Aug 2005 10:35:27 +0000 (22:35 +1200)]
[PATCH] Add merge detection to git-cvsimport

Added -m and -M flags for git-cvsimport to detect merge commits in cvs.
While this trusts the commit message, in repositories where merge commits
indicate 'merged from FOOBRANCH' the import works surprisingly well.

Even if some merges from CVS are bogus or incomplete, the resulting
branches are in better state to go forward (and merge) than without any
merge detection.

Signed-off-by: Martin Langhoff <martin.langhoff@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add GIT glossary
Johannes Schindelin [Wed, 17 Aug 2005 14:56:48 +0000 (16:56 +0200)]
[PATCH] Add GIT glossary

[jc: This is the version without asciidoc cross references;
Johannes says that the cross referenced one is generated from
this file using a Perl script, so I am placing this as the
source, and expecting to later receive the script and a Makefile
entry or two to massage this file into the final HTML or
whatever form.]

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Let git-format-patch-script write on stdout
Marco Costalba [Wed, 17 Aug 2005 12:17:27 +0000 (05:17 -0700)]
[PATCH] Let git-format-patch-script write on stdout

Avoid that git-format-patch writes out patch series
information on stderr when there are no errors

Signed-off-by: Marco Costalba <mcostalba@yahoo.it>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Be consistent in naming of remote heads
Kris Shannon [Wed, 17 Aug 2005 03:02:16 +0000 (13:02 +1000)]
[PATCH] Be consistent in naming of remote heads

The _remote_name variable used for messages does not need the
refs/heads/ prefix included.

Signed-off-by: Kris Shannon <kris.shannon@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] use it's and its correctly in documentation
Greg Louis [Wed, 17 Aug 2005 16:37:04 +0000 (12:37 -0400)]
[PATCH] use it's and its correctly in documentation

At one place in Documentation/tutorial.txt and several in the base
README, its was wrongly used in place of it's or vice versa.  One
instance remains somewhere in Documentation/howto/, which I didn't
correct because it's in a quotation.

Signed-off-by: Greg Louis <glouis@dynamicro.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Also handle CVS branches with a '/' in their name
Johannes Schindelin [Wed, 17 Aug 2005 09:19:20 +0000 (11:19 +0200)]
[PATCH] Also handle CVS branches with a '/' in their name

I track a CVS project which has a branch with a '/' in the branch name.
Since git wants the branch name to be a file name at the same time,
substitute that character to a '-' by default (override with "-s <subst>").
This should work well, despite the fact that a division and a difference
are completely different :-)

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] When copying or renaming, keep the mode, please
Johannes Schindelin [Wed, 17 Aug 2005 07:01:07 +0000 (09:01 +0200)]
[PATCH] When copying or renaming, keep the mode, please

Without this patch, git-apply does not retain the mode when renaming or
copying files.

[jc: Good catch, Johannes.  I added a test case to demonstrate the
breackage in the original.]

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Change git-branch to list branches
Kalle Valo [Tue, 16 Aug 2005 17:58:10 +0000 (20:58 +0300)]
[PATCH] Change git-branch to list branches

If no argument provided to `git branch`, show available branches and
mark current branch with star.

This is based on patch written by Amos Waterland <apw@rossby.metr.ou.edu>.

[jc: and I changed it to handle subdirectories under refs/heads/ as well.]

Signed-off-by: Kalle Valo <Kalle.Valo@iki.fi>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit: pass explicit path to git-diff-files.
Junio C Hamano [Wed, 17 Aug 2005 01:08:19 +0000 (18:08 -0700)]
git-commit: pass explicit path to git-diff-files.

When running "git commit" with explicit path arguments, allow it to
take directory name.  This makes "git commit Documentation/" to commit
everything that is changed under Documentation/ directory.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoPass options to underlying git-rev-list from show-branches.
Junio C Hamano [Tue, 16 Aug 2005 20:01:39 +0000 (13:01 -0700)]
Pass options to underlying git-rev-list from show-branches.

This lets you say "git show-branches --max-count=30".

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd --symbolic flag to git-rev-parse.
Junio C Hamano [Tue, 16 Aug 2005 19:36:46 +0000 (12:36 -0700)]
Add --symbolic flag to git-rev-parse.

This is most useful with --all, --revs-only, --no-flags and --verify.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd graft support.
Paul Mackerras [Wed, 17 Aug 2005 11:27:55 +0000 (21:27 +1000)]
Add graft support.

We read .git/info/grafts and use the information in there to
override the list of parents we get from git-rev-list or
git-cat-file.

18 years agoTeach applymbox to keep the Subject: line.
Junio C Hamano [Wed, 17 Aug 2005 05:18:27 +0000 (22:18 -0700)]
Teach applymbox to keep the Subject: line.

This corresponds to the -k flag to git format-patch --mbox
option.  The option should probably not be used when applying a
real e-mail patch, but is needed when format-patch and applymbox
pair is used for cherrypicking.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUse LF and allow comments in objects/info/alternates file.
Junio C Hamano [Wed, 17 Aug 2005 01:22:05 +0000 (18:22 -0700)]
Use LF and allow comments in objects/info/alternates file.

Yes, using the same format for the file and the environment variable
was a big mistake.  This uses LF as the path separator, and allows
lines that begin with '#' to be comments.  ':' is no longer a separator
in objects/info/alternates file.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-format-patch fixes.
Junio C Hamano [Tue, 16 Aug 2005 22:23:50 +0000 (15:23 -0700)]
git-format-patch fixes.

Introduces --keep-subjects flag to tell it not to munge the
first line of the commit message.  Running "git applymbox" on
the output from "git format-patch -m -k" would preserve the
original commit information better this way.

At the same time, prefix Subject: on the first line of the
commit, to help people cut&copy.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Improve handling of "." and ".." in git-diff-*
Linus Torvalds [Wed, 17 Aug 2005 03:44:32 +0000 (20:44 -0700)]
[PATCH] Improve handling of "." and ".." in git-diff-*

This fixes up usage of ".." (without an ending slash) and "." (with or
without the ending slash) in the git diff family.

It also fixes pathspec matching for the case of an empty pathspec, since a
"." in the top-level directory (or enough ".." under subdirectories) will
result in an empty pathspec. We used to not match it against anything, but
it should in fact match everything.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Fix test failure due to overly strict .git directory tests
Linus Torvalds [Wed, 17 Aug 2005 02:50:37 +0000 (19:50 -0700)]
[PATCH] Fix test failure due to overly strict .git directory tests

We may not actually have a valid HEAD at all times, so relax the validity
tests for a .git subdirectory accordingly.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Make "git diff" work inside relative subdirectories
Linus Torvalds [Wed, 17 Aug 2005 01:06:34 +0000 (18:06 -0700)]
[PATCH] Make "git diff" work inside relative subdirectories

We always show the diff as an absolute path, but pathnames to diff are
taken relative to the current working directory (and if no pathnames are
given, the default ends up being all of the current working directory).

Note that "../xyz" also works, so you can do

cd linux/drivers/char
git diff ../block

and it will generate a diff of the linux/drivers/block changes.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add a bit more links to the commands to the main git(7) page.
Junio C Hamano [Wed, 17 Aug 2005 00:41:32 +0000 (17:41 -0700)]
[PATCH] Add a bit more links to the commands to the main git(7) page.

There are many programs like git-add not described at all, and the
organization of the list of commands may be suboptimal, but we have to
start somewhere.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Reformat git-show-branches-script documentation.
Junio C Hamano [Wed, 17 Aug 2005 00:24:34 +0000 (17:24 -0700)]
[PATCH] Reformat git-show-branches-script documentation.

... using ListingBlock of asciidoc.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Make do_each_ref() follow symlinks.
Matt Draisey [Tue, 16 Aug 2005 05:46:34 +0000 (01:46 -0400)]
[PATCH] Make do_each_ref() follow symlinks.

Because there is no reason not to, and this makes things a bit
safer when running "git prune".

[jc: I do not necessarily agree with the reasoning of the
original author that it is a good way to keep "git prune" from
stomping on objects to have a symlink that points at .git/refs
of the repository A in the .git/refs of the repository B when
repository A borrows object database from repository B.  For one
thing, the object database that everybody borrows objects from
may not even have its own .git/refs hierarchy.  Come to think of
it, maybe we should disallow symlink inside .git/refs hierarchy;
we update the files there by creat/rename pair, so having
symlinks would not work anyway when you do anything that would
update them.]

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add -k kill keyword expansion option to git-cvsimport - revised
Martin Langhoff [Tue, 16 Aug 2005 05:39:29 +0000 (17:39 +1200)]
[PATCH] Add -k kill keyword expansion option to git-cvsimport - revised

Early versions of git-cvsimport defaulted to using preexisting keyword
expansion settings. This change preserves compatibility with existing cvs
imports and allows new repository migrations to kill keyword expansion.

After exploration of the different -k modes in the cvs protocol, we use -kk
which kills keyword expansion wherever possible. Against the protocol
spec, -ko and -kb will sometimes expand keywords.

Should improve our chances of detecting merges and reduce imported
repository size.

Signed-off: Martin Langhoff <martin.langhoff@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Fix local pull that uses packs.
Junio C Hamano [Tue, 16 Aug 2005 05:48:09 +0000 (22:48 -0700)]
[PATCH] Fix local pull that uses packs.

The patch mixed up the sha1 for the object being pulled, and the
sha1 the pull engine internally use to represent the name of the
pack that contains the object being pulled.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Support packs in local-pull
Daniel Barkalow [Tue, 16 Aug 2005 04:10:32 +0000 (00:10 -0400)]
[PATCH] Support packs in local-pull

If it doesn't find an object, it looks for an index that contains it
and uses the same methods on that instead.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add function to read an index file from an arbitrary filename.
Daniel Barkalow [Tue, 16 Aug 2005 04:10:03 +0000 (00:10 -0400)]
[PATCH] Add function to read an index file from an arbitrary filename.

Note that the pack file has to be in the usual location if it gets
installed later.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Fixed two bugs in git-cvsimport-script.
iso-8859-1?Q?David_K=E5gedal [Mon, 15 Aug 2005 18:18:25 +0000 (20:18 +0200)]
[PATCH] Fixed two bugs in git-cvsimport-script.

The git-cvsimport-script had a copule of small bugs that prevented me
from importing a big CVS repository.

The first was that it didn't handle removed files with a multi-digit
primary revision number.

The second was that it was asking the CVS server for "F" messages,
although they were not handled.

I also updated the documentation for that script to correspond to
actual flags.

Signed-off-by: David K?5gedal <davidk@lysator.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] cvsgit fixes: spaces in filenames and CVS server dialog woes
Matthias Urlichs [Mon, 15 Aug 2005 09:28:19 +0000 (11:28 +0200)]
[PATCH] cvsgit fixes: spaces in filenames and CVS server dialog woes

Problems found while importing dasher's CVS:

* Allow spaces in filenames.
* cvsps may create unnamed branches with revisions that don't really
  exist, which causes the CVS server to return something we haven't
  hitherto expected.
* Report deleted files when being verbose.
* Also, report the commit date.
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake tools use DESTDIR as well (otherwise rpm build is broken).
Chris Wright [Tue, 16 Aug 2005 18:28:02 +0000 (11:28 -0700)]
Make tools use DESTDIR as well (otherwise rpm build is broken).

Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Make git-format-patch's signoff option more consistent
Johannes Schindelin [Tue, 16 Aug 2005 00:20:34 +0000 (02:20 +0200)]
[PATCH] Make git-format-patch's signoff option more consistent

Since git-commit-script has a "--signoff" option, use that in
git-format-patch-script, too (and since partial option names are
supported,"--sign" is still valid).

Also, if the message already contains the S-O-B line, silently ignore the
"--signoff" request.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTentative fix to git-commit-script
Junio C Hamano [Mon, 15 Aug 2005 22:37:37 +0000 (15:37 -0700)]
Tentative fix to git-commit-script

The recent change to give the multiple commit message source was not
carrying over the authorship information from -C/-c commits correctly.
The export of the environment variable happens only in the subprocess,
not the main process that eventually runs git-commit-tree.

The right fix might be to teach git-commit-script to grok the From:
and Date: lines at the beginning of the commit message just like
git-applymbox knows how, but this has to do until that enhancement
happens.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add documentation for git repack and git-prune-packed.
Junio C Hamano [Mon, 15 Aug 2005 22:48:47 +0000 (15:48 -0700)]
[PATCH] Add documentation for git repack and git-prune-packed.

[jc: the patch forgot to update the main git.txt documentation,
making all these new documentation practically no-op, so I added
a minimum attempt linking them from there.]

Signed-off-by: Ryan Anderson <ryan@michonline.com>
18 years ago[PATCH] clean up git script
Junio C Hamano [Mon, 15 Aug 2005 22:37:37 +0000 (15:37 -0700)]
[PATCH] clean up git script

Makes git work with a pure POSIX shell (tested with bash --posix and ash).
Right now git causes ash to choke on the redundant shift on line two.

Reduces the number of system calls git makes just to do a usage
statement from 22610 to 1122, and the runtime for same from 349ms to
29ms on my x86 Linux box.

Presents a standard usage statement, and pretty prints the available
commands in a form that does not scroll off small terminals.

[jc: while shifting when $# was zero was a bug, the original
patch failed to shift when it needs to, which I fixed up.]

Signed-off-by: Amos Waterland <apw@rossby.metr.ou.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAudit rev-parse users.
Junio C Hamano [Mon, 15 Aug 2005 22:37:37 +0000 (15:37 -0700)]
Audit rev-parse users.

Make sure that we say --verify when we want to get a single SHA1
name.  Also when we say --verify, --revs-only is redundant.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoKeep excellent tutorial for using topic branches by Tony Luck
Junio C Hamano [Mon, 15 Aug 2005 22:36:52 +0000 (15:36 -0700)]
Keep excellent tutorial for using topic branches by Tony Luck

I would eventually like to move this to become a part of the tutorial,
but anyway, this was an excellent post that describes how topic
branches can be used to keep track of local changes.

18 years ago[PATCH] Run Ispell through git.txt
Yasushi SHOJI [Mon, 15 Aug 2005 15:23:06 +0000 (00:23 +0900)]
[PATCH] Run Ispell through git.txt

Signed-off-by: Yasushi SHOJI <yashi@atmark-techno.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoShow which branch you are on in git-status output.
Junio C Hamano [Mon, 15 Aug 2005 08:56:23 +0000 (01:56 -0700)]
Show which branch you are on in git-status output.

When not on the "master" branch, show which branch we are
working on in the git-status message.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd git-show-branches-script
Junio C Hamano [Mon, 15 Aug 2005 08:42:15 +0000 (01:42 -0700)]
Add git-show-branches-script

Often I find myself wanting to do quick branches check when I am
not in the windowing environment and cannot run gitk.

This stupid script shows commits leading to the heads of
interesting branches with indication which ones belong to which
branches, so that fork point is somewhat discernible without
using gitk.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Add some simple howtos, culled from the mailing list.
Ryan Anderson [Mon, 15 Aug 2005 08:25:53 +0000 (04:25 -0400)]
[PATCH] Add some simple howtos, culled from the mailing list.

I think these are useful, and I think putting them in a new "howto"
directory might help some users until we get to the point of splitting
up the tutorial to be easier to read.

Given the authorship, I think it's safe to put these in the repository.

Signed-off-by: Ryan Anderson <ryan@michonline.com>
18 years ago[PATCH] Fixed/Extended example for update hook
Josef Weidendorfer [Sat, 13 Aug 2005 20:39:41 +0000 (22:39 +0200)]
[PATCH] Fixed/Extended example for update hook

Add sample code to distinguish --force rebased head and simple
fast-forward.  At the same time fixes a real bug; the "new ref"
path was using a wrong parameter.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd cheap local clone '-s' flag to git-clone-script
Junio C Hamano [Mon, 15 Aug 2005 00:25:57 +0000 (17:25 -0700)]
Add cheap local clone '-s' flag to git-clone-script

Using the $GIT_OBJECT_DIRECTORY/info/alternates mechanism,
create a new repository that borrows objects from the original
repository when --shared flag is given in addition to --local.

It is worth pointing out that the "cloned" repository depends on
the original repository, so this should be used only when you
can reasonably trust that the original repository would not
disappear without your knowing.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAlternate object pool mechanism updates.
Junio C Hamano [Mon, 15 Aug 2005 00:25:57 +0000 (17:25 -0700)]
Alternate object pool mechanism updates.

It was a mistake to use GIT_ALTERNATE_OBJECT_DIRECTORIES
environment variable to specify what alternate object pools to
look for missing objects when working with an object database.
It is not a property of the process running the git commands,
but a property of the object database that is partial and needs
other object pools to complete the set of objects it lacks.

This patch allows you to have $GIT_OBJECT_DIRECTORY/info/alternates
whose contents is in exactly the same format as the environment
variable, to let an object database name alternate object pools
it depends on.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation updates.
Junio C Hamano [Mon, 15 Aug 2005 00:24:36 +0000 (17:24 -0700)]
Documentation updates.

Linus brought up that documentation for many commands have
incorrect attribution.  I started counting lines again, but
ended up adding a handful of missing manual pages.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] plug memory leak in diff.c::diff_free_filepair()
Yasushi SHOJI [Sat, 13 Aug 2005 10:58:56 +0000 (19:58 +0900)]
[PATCH] plug memory leak in diff.c::diff_free_filepair()

When I run git-diff-tree on big change, it seems the command eats so
much memory.  so I just put git under valgrind to see what's going on.
diff_free_filespec_data() doesn't free diff_filespec itself.

[jc: I ended up doing things slightly differently from Yasushi's
patch.  The original idea was to use free_filespec_data() only to
free the data portion and keep useing the filespec itself, but
no existing code seems to do things that way, so I just yanked
that part out.]

Signed-off-by: Yasushi SHOJI <yashi@atmark-techno.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] fix null TZ problems on os/x
Brad Roberts [Sat, 13 Aug 2005 10:22:46 +0000 (03:22 -0700)]
[PATCH] fix null TZ problems on os/x

It seems that the localtime() libc routine doesn't care for finding a TZ
that's empty.  It's ok with TZ not being set.  Setting the TZ to GMT allowed
these tests to pass.

$ uname -v
Darwin Kernel Version 7.9.0: Wed Mar 30 20:11:17 PST 2005; root:xnu/xnu-517.12.7.obj~1/RELEASE_PPC

Signed-off-by: Brad Roberts <braddr@puremagic.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] fix git-checkout-cache segfault on parisc
James Bottomley [Sat, 13 Aug 2005 15:29:32 +0000 (10:29 -0500)]
[PATCH] fix git-checkout-cache segfault on parisc

The index cleanup code is executed via atexit() which is *after* main
has completed, so the stack allocated cache_file has gone out of scope.
Parisc seems to use stack in the destructor functions, so cache_file
gets partially overwritten leading to the predictable disastrous
consequences.

[jc: Just to make sure, I audited other users of the function
 hold_index_file_for_update() to make sure they do not have this
 problem; everybody else uses non-stack cache_file structure and
 is fine.  Thanks, James.]

Signed-off-by: James Bottomley <James.Bottomley@SteelEye.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Make get_sha1_basic() more careful
Linus Torvalds [Sat, 13 Aug 2005 18:05:25 +0000 (11:05 -0700)]
[PATCH] Make get_sha1_basic() more careful

The "get_sha1_hex()" function is designed to work with SHA1 hex strings
that may be followed by arbitrary crud. However, that's not acceptable for
"get_sha1()" which is used for command line arguments etc: we don't want
to silently allow random characters after the end of the SHA1.

So verify that the hex string is all we have.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Make sure git-resolve-script always works on commits
Linus Torvalds [Sat, 13 Aug 2005 17:50:56 +0000 (10:50 -0700)]
[PATCH] Make sure git-resolve-script always works on commits

You can resolve a tag, and it does the right thing except that it might
end up writing the tag itself into the resulting HEAD, which will confuse
subsequent operations no end.

This makes sure that when we resolve two heads, we will have turned them
into proper commits before we start acting on them.

This also fixes the parsing of "treeish^0", which would incorrectly
resolve to "treeish" instead of causing an error.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd SubmittingPatches
Junio C Hamano [Sat, 13 Aug 2005 06:48:09 +0000 (23:48 -0700)]
Add SubmittingPatches

Not that I have stricter patch submission standard than ordinary
projects, I wanted to have it to make sure people understand
what they are doing when they add their own Signed-off-by line.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd --signoff and --verify option to git commit.
Junio C Hamano [Sat, 13 Aug 2005 06:39:15 +0000 (23:39 -0700)]
Add --signoff and --verify option to git commit.

As brought up in the discussion which followed a patch to add a
signed-off-by line with the --sign flag to format-patch from
Johannes Schindelin, add --signoff to the git commit command.

Also add --verify to make sure the lines you introduced are
clean, which is more useful in commit but not very much in
format-patch as it was originally implemented, because finding
botches at format-patch time is too late.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDebianization updates
Matthias Urlichs [Sat, 13 Aug 2005 00:38:15 +0000 (02:38 +0200)]
Debianization updates

git-tk should be architecture independent.
git-core forgot to depend on perl.

Signed-Off-By: Matthias Urlichs <smurf@debian.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix debianization: mark git-tk architecture neutral.
Junio C Hamano [Fri, 12 Aug 2005 20:32:02 +0000 (13:32 -0700)]
Fix debianization: mark git-tk architecture neutral.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdjust binary packaging to dest=>DESTDIR change.
Junio C Hamano [Fri, 12 Aug 2005 19:25:34 +0000 (12:25 -0700)]
Adjust binary packaging to dest=>DESTDIR change.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake tools/ directory first-class citizen.
Junio C Hamano [Fri, 12 Aug 2005 17:37:33 +0000 (10:37 -0700)]
Make tools/ directory first-class citizen.

Tools directory being separate is just a historical
coincidence.  Build and install together with the main
directory, just like the clean target does.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Unify Makefile indentation
Petr Baudis [Thu, 4 Aug 2005 22:52:46 +0000 (00:52 +0200)]
[PATCH] Unify Makefile indentation

Use <tab> instead of two spaces uniformly in the Makefile, even in the
ifdefs. Gives it a nice consistent look.

[jc: At the same time I indented the nested ifdefs to make them
slightly easier to read.]

Signed-off-by: Petr Baudis <pasky@ucw.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Use $DESTDIR instead of $dest
Petr Baudis [Thu, 4 Aug 2005 23:56:38 +0000 (01:56 +0200)]
[PATCH] Use $DESTDIR instead of $dest

$DESTDIR is more usual during the build than $dest and is what
is usually used in the makefiles, so let's use it too.

Signed-off-by: Petr Baudis <pasky@ucw.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake CFLAGS overridable from make command line.
Pavel Roskin [Sat, 6 Aug 2005 05:36:15 +0000 (01:36 -0400)]
Make CFLAGS overridable from make command line.

This patch renames COPTS to CFLAGS, because it's COPTS that was user
overridable.  Also, -Wall is moved there because it's optional.  What
was CFLAGS is now ALL_CFLAGS, which users should not override.

Defines are added to DEFINES.  Since ALL_CFLAGS is recursively expanded,
it uses the final value of DEFINES.

Implicit rules are made explicit since the implicit rules use CFLAGS
rather than ALL_CFLAGS.  I believe that serious projects should not rely
on implicit rules anyway.  Percent rules are used because they are used
already and because they don't need the .SUFFIXES target.

[jc: in addition to updating the patch for 0.99.4, I fixed up a
glitch in Pavel's original patch which compiled sha1.o out of
mozilla-sha1/sha1.c, where it should have left the resulting
object file in mozilla-sha1 directory for later "ar".]

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoClean generated files a bit more, to cope with Debian build droppings.
Junio C Hamano [Fri, 12 Aug 2005 08:03:07 +0000 (01:03 -0700)]
Clean generated files a bit more, to cope with Debian build droppings.

Also attempt to build a source package for debian.

18 years ago[PATCH] Add "--sign" option to git-format-patch-script
Johannes Schindelin [Fri, 12 Aug 2005 13:32:42 +0000 (15:32 +0200)]
[PATCH] Add "--sign" option to git-format-patch-script

This adds the option "--sign" to git-format-patch-script which adds
a Signed-off-by: line automatically.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agofetch-pack: start multi-head pulling.
Junio C Hamano [Fri, 12 Aug 2005 09:08:29 +0000 (02:08 -0700)]
fetch-pack: start multi-head pulling.

This is a beginning of resurrecting the multi-head pulling support
for git-fetch-pack command.  The git-fetch-script wrapper still
only knows about fetching a single head, without renaming, so it is
not very useful unless you directly call git-fetch-pack itself yet.

It also fixes a longstanding obsolete description of how the command
discovers the list of local commits.

18 years agoUpdate unpack-objects usage and documentation.
Junio C Hamano [Fri, 12 Aug 2005 08:45:52 +0000 (01:45 -0700)]
Update unpack-objects usage and documentation.

It long supported -q flag to suppress progress meter without
properly being documented.

18 years ago[PATCH] Also parse objects we already have
Daniel Barkalow [Fri, 12 Aug 2005 03:17:55 +0000 (23:17 -0400)]
[PATCH] Also parse objects we already have

In the case where we don't know from context what type an object is, but
we don't have to fetch it, we need to parse it to determine the type
before processing it.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Fix parallel pull dependancy tracking.
Daniel Barkalow [Thu, 11 Aug 2005 23:38:09 +0000 (19:38 -0400)]
[PATCH] Fix parallel pull dependancy tracking.

It didn't refetch an object it already had (good), but didn't process
it, either (bad). Synchronously process anything you already have.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoClean generated deb files.
Junio C Hamano [Fri, 12 Aug 2005 01:52:34 +0000 (18:52 -0700)]
Clean generated deb files.

Do not forgot that we have a separate git-tk package these days.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Trapping exit in tests, using return for errors: further fixes.
Pavel Roskin [Thu, 11 Aug 2005 16:00:40 +0000 (12:00 -0400)]
[PATCH] Trapping exit in tests, using return for errors: further fixes.

"return" from a test would leave the exit trap set, which could cause a
spurious error message if it's the last test in the script or
--immediate is used.

The easiest solution would be to have a global trap that is set when
test-lib.sh is sourced and unset either by test_done(), error() or by
test_failure_() with --immediate.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Trapping exit in tests, using return for errors
Pavel Roskin [Thu, 11 Aug 2005 03:56:21 +0000 (23:56 -0400)]
[PATCH] Trapping exit in tests, using return for errors

I have noticed that "make test" fails without any explanations when the
"merge" utility is missing.  I don't think tests should be silent in
case of failure.

It turned out that the particular test was using "exit" to interrupt the
test in case of an error.  This caused the whole test script to exit.
No further tests would be run even if "--immediate" wasn't specified.
No error message was printed.

This patch does following:

All instances of "exit", "exit 1" and "(exit 1)" in tests have been
replaced with "return 1".  In fact, "(exit 1)" had no effect.

File descriptor 5 is duplicated from file descriptor 1.  This is needed
to print important error messages from tests.

New function test_run_() has been introduced.  Any "return" in the test
would merely cause that function to return without skipping calls to
test_failure_() and test_ok_().  The new function also traps "exit" and
treats it like a fatal error (in case somebody reintroduces "exit" in
the tests).

test_expect_failure() and test_expect_success() check both the result of
eval and the return value of test_run_().  If the later is not 0, it's
always a failure because it indicates the the test didn't complete.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Need to set PAGER in tests
Pavel Roskin [Thu, 11 Aug 2005 02:10:01 +0000 (22:10 -0400)]
[PATCH] Need to set PAGER in tests

"t5400-send-pack.sh --verbose" stops waiting for user input.  It happens
because "git log" uses less for output now.  To prevent this, PAGER
should be set to cat.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Missing test_done
Pavel Roskin [Thu, 11 Aug 2005 02:15:02 +0000 (22:15 -0400)]
[PATCH] Missing test_done

All test scripts should end with test_done, which reports the test
results.  In the future, it could be used for other purposes, e.g. to
distinguish graceful end from "exit" in a test.  This patch fixes
scripts that don't call test_done.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoString comparison of test is done with '=', not '=='.
Junio C Hamano [Thu, 11 Aug 2005 05:53:27 +0000 (22:53 -0700)]
String comparison of test is done with '=', not '=='.

Caught this during a test setting /bin/sh to (d)ash.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDebian packaging fixes.
Matthias Urlichs [Wed, 10 Aug 2005 23:54:36 +0000 (01:54 +0200)]
Debian packaging fixes.

- Split gitk off to its own package;
  it needs tk installed, but nothing else does.
- Refer to GPL properly, don't install COPYING.
- Fix maintainer.
- Use dh_movefiles instead of dh_install;
  we don't want to list everything *except* gitk.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agomerge-base.c: pathological case fix.
Junio C Hamano [Fri, 12 Aug 2005 01:13:55 +0000 (18:13 -0700)]
merge-base.c: pathological case fix.

Also add some illustration requested by Linus.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago[PATCH] Speed up git-merge-base a lot
Linus Torvalds [Wed, 10 Aug 2005 23:26:32 +0000 (16:26 -0700)]
[PATCH] Speed up git-merge-base a lot

In commit 4f7eb2e5a351e0d1f19fd4eab7e92834cc4528c2 I fixed git-merge-base
getting confused by datestamps that caused it to traverse things in a
non-obvious order.

However, my fix was a very brute-force one, and it had some really
horrible implications for more complex trees with lots of parallell
development. It might end up traversing all the way to the root commit.

Now, normally that isn't that horrible: it's used mainly for merging, and
the bad cases really tend to happen fairly rarely, so if it takes a few
seconds, we're not in too bad shape.

However, gitk will also do the git-merge-base for every merge it shows,
because it basically re-does the trivial merge in order to show the
"interesting" parts. And there we'd really like the result to be
instantaneous.

This patch does that by walking the tree more completely, and using the
same heuristic as git-rev-list to decide "ok, the rest is uninteresting".

In one - hopefully fairly extreme - case, it made a git-merge-base go from
just under five seconds(!) to a tenth of a second on my machine.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge with gitk.
Junio C Hamano [Fri, 12 Aug 2005 01:24:35 +0000 (18:24 -0700)]
Merge with gitk.

This merges commit
    b664550c066810b770ad3e19cafe2fbdd42c6793
from gitk into our head commit
    bf570303153902ec3d85570ed24515bcf8948848

Sincerely,
    jit-merge command.

18 years agoRefine the update heuristic to improve responsiveness a bit.
Paul Mackerras [Wed, 10 Aug 2005 23:56:23 +0000 (09:56 +1000)]
Refine the update heuristic to improve responsiveness a bit.

The previous commit improved performance a lot but also meant that
we waited longer to see something drawn.  This refines the heuristics
for when to call update so that (1) when we have finished processing
a bufferfull of information from git-rev-list, we call update if
enough time has elapsed, regardless of how many commits we've drawn,
and (2) the number of commits drawn between updates scales with the
total number of commits drawn: 1 for 1-99 commits, 10 for 100-9999
commits, or 100 for >= 10000 commits.

18 years agoGIT 0.99.4. v0.99.4
Junio C Hamano [Wed, 10 Aug 2005 21:43:08 +0000 (14:43 -0700)]
GIT 0.99.4.

Mark it official.  Finally.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge with master.
Junio C Hamano [Wed, 10 Aug 2005 20:20:08 +0000 (13:20 -0700)]
Merge with master.

This merges commit
    e8c80638abc99928dba9ac832589749a531e2e21
from master into our head commit
    bdb71a41caa9ffed2e3207736c21fffbc2007670

Sincerely,
    jit-merge command.

18 years agoMerge with gitk.
Junio C Hamano [Wed, 10 Aug 2005 20:18:11 +0000 (13:18 -0700)]
Merge with gitk.

This merges commit
    466e4fdd6696c89700294e1a54fa73e1fb94643b
from gitk into our head commit
    ef0bd2e6e643f09a8294937caf99463328b9215a

Sincerely,
    jit-merge command.

18 years agoMerge with master.
Junio C Hamano [Wed, 10 Aug 2005 17:55:54 +0000 (10:55 -0700)]
Merge with master.

This merges commit
    ef0bd2e6e643f09a8294937caf99463328b9215a
from master into our head commit
    f69714c38c6f3296a4bfba0d057e0f1605373f49

Sincerely,
    jit-merge command.

18 years ago[PATCH] Fix git-rev-parse's parent handling
Johannes Schindelin [Wed, 10 Aug 2005 03:07:36 +0000 (05:07 +0200)]
[PATCH] Fix git-rev-parse's parent handling

git-rev-parse HEAD^1 would fail, because of an off-by-one bug (but HEAD^
would yield the expected result). Also, when the parent does not exist, do
not silently return an incorrect SHA1. Of course, this no longer applies
to git-rev-parse alone, but every user of get_sha1().

While at it, add a test.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoOnly do an update every 100 commits when drawing the graph.
Paul Mackerras [Wed, 10 Aug 2005 12:50:28 +0000 (22:50 +1000)]
Only do an update every 100 commits when drawing the graph.

On a large repository with > 60,000 commits, each call to the Tk
update primitive (which gives Tk a chance to respond to events and
redraw the screen) was taking up to 0.2 seconds.  Because the logic
was to call update after drawing a commit if 0.1 seconds had passed
since the last update call, we were calling it for every commit,
which was slowing us down enormously.  Now we also require that we
have drawn 100 commits since the last update (as well as it being
at least 0.1 seconds since the last update).  Drawing 100 commits
takes around 0.1 - 0.2 seconds (even in this large repo) on my G5.

18 years agoMerge with master.
Junio C Hamano [Wed, 10 Aug 2005 07:10:18 +0000 (00:10 -0700)]
Merge with master.

This merges commit
    c35a7b8d806317dc1762e36561cbd31c2530dd9c
from master into our head commit
    edee414c3e5a546aae3dd1529f397df949713305

Sincerely,
    jit-merge command.

18 years agoSkip merges in format-patch.
Junio C Hamano [Wed, 10 Aug 2005 07:06:16 +0000 (00:06 -0700)]
Skip merges in format-patch.

18 years agoAdd -v option to git-cherry.
Junio C Hamano [Wed, 10 Aug 2005 06:58:35 +0000 (23:58 -0700)]
Add -v option to git-cherry.

18 years agoTeach git push .git/branches shorthand
Junio C Hamano [Mon, 8 Aug 2005 05:55:45 +0000 (22:55 -0700)]
Teach git push .git/branches shorthand

Although it is uncertain if we would keep .git/branches for
long, the shorthand stored there can be used for pushing if it
is host:path/to/git format, so let's make use of it.  This does
not use git-parse-remote because that script will be rewritten
quite a bit for updated pulling.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agols-remote: drop storing operation and add documentation.
Junio C Hamano [Mon, 8 Aug 2005 08:20:03 +0000 (01:20 -0700)]
ls-remote: drop storing operation and add documentation.

The store operation was never useful because we needed to fetch
the objects needed to complete the reference.  Remove it.

The fetch command fetch multiple references shortly to
replace the lost "store" functionality in more a generic way.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTeach format-patch, rebase and cherry a..b format
Junio C Hamano [Sun, 7 Aug 2005 22:51:09 +0000 (15:51 -0700)]
Teach format-patch, rebase and cherry a..b format

Although these commands take only begin and end, not necessarily
generic SHA1 expressions rev-parse supports, supporting a..b
notation is good for consistency.  This commit adds such without
breaking backward compatibility.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoStash away the original head in ORIG_HEAD when resetting.
Junio C Hamano [Sun, 7 Aug 2005 03:59:47 +0000 (20:59 -0700)]
Stash away the original head in ORIG_HEAD when resetting.

When rewinding the head, stash away the value of the original
HEAD in ORIG_HEAD, just like git-resolve-script does.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-revert: revert an existing commit.
Junio C Hamano [Wed, 10 Aug 2005 05:30:17 +0000 (22:30 -0700)]
git-revert: revert an existing commit.

Given one existing commit, revert the change the patch
introduces, and record a new commit that records it.  This
requires your working tree to be clean (no modifications from
the HEAD commit).

This is based on what Linus posted to the list, with
enhancements he suggested, including the use of -M to attempt
reverting renames.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoIntroduce --pretty=oneline format.
Junio C Hamano [Tue, 9 Aug 2005 05:15:40 +0000 (22:15 -0700)]
Introduce --pretty=oneline format.

This introduces --pretty=oneline to git-rev-tree and
git-rev-list commands to show only the first line of the commit
message, without frills.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocument "git commit"
Junio C Hamano [Tue, 9 Aug 2005 00:39:18 +0000 (17:39 -0700)]
Document "git commit"

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit: log parameter updates.
Junio C Hamano [Tue, 9 Aug 2005 00:03:14 +0000 (17:03 -0700)]
git-commit: log parameter updates.

While moving '-m' to make room for CVS compatible "here is the
log message", enhance source of log parameters.

  -m 'message': a command line parameter.
  -F <file>   : a file (use '-' to read from stdin).
  -C <commit> : message in existing commit.
  -c <commit> : message in existing commit (allows further editing).

Longer option names for these options are also available.

While we are at it, get rid of shell array bashism.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUpdate rev-parse flags list.
Junio C Hamano [Tue, 9 Aug 2005 02:31:37 +0000 (19:31 -0700)]
Update rev-parse flags list.

I haven't audited the rev-parse users, but I am having a feeling
that many of them would choke when they expect a couple of SHA1
object names and malicious user feeds them "--max-count=6" or
somesuch to shoot himself in the foot.  Anyway, this adds a
couple of missing parameters that affect the list of revs to be
returned from rev-list, not the flags that affect how they are
presented by rev-list.  I think that is the intention, but I am
not quite sure.

Signed-off-by: Junio C Hamano <junkio@cox.net>