Code

git.git
19 years agogit-prune: quote possibly empty $dryrun as parameter to test
Junio C Hamano [Fri, 18 Nov 2005 19:15:40 +0000 (11:15 -0800)]
git-prune: quote possibly empty $dryrun as parameter to test

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-am: --binary; document --resume and --binary.
Junio C Hamano [Thu, 17 Nov 2005 00:46:24 +0000 (16:46 -0800)]
git-am: --binary; document --resume and --binary.

Now git-apply can grok binary replacement patches, give --binary
flag to git-am.  As a safety measure, this is not by default
enabled, so that you do not let malicious e-mailed patch to
replace an arbitrary path with just a couple of lines (diff
index lines, the filename and string "Binary files "...) by
accident.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoTeach "approxidate" about weekday syntax
Linus Torvalds [Thu, 17 Nov 2005 20:36:30 +0000 (12:36 -0800)]
Teach "approxidate" about weekday syntax

This allows people to use syntax like "last thursday" for the approxidate.

(Or, indeed, more complex "three thursdays ago", but I suspect that would
be pretty unusual).

NOTE! The parsing is strictly sequential, so if you do

"one day before last thursday"

it will _not_ do what you think it does. It will take the current time,
subtract one day, and then go back to the thursday before that. So to get
what you want, you'd have to write it the other way around:

"last thursday and one day before"

which is insane (it's usually the same as "last wednesday" _except_ if
today is Thursday, in which case "last wednesday" is yesterday, and "last
thursday and one day before" is eight days ago).

Similarly,

"last thursday one month ago"

will first go back to last thursday, and then go back one month from
there, not the other way around.

I doubt anybody would ever use insane dates like that, but I thought I'd
point out that the approxidate parsing is not exactly "standard English".

Side note 2: if you want to avoid spaces (because of quoting issues), you
can use any non-alphanumberic character instead. So

git log --since=2.days.ago

works without any quotes.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMake git-pack-redundant non-horribly slow on large sets of packs
Lukas Sandström [Thu, 17 Nov 2005 13:11:56 +0000 (14:11 +0100)]
Make git-pack-redundant non-horribly slow on large sets of packs

Change the smallest-set detection algortithm so that when
we have found a good set, we don't check any larger sets.

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-repack: Fix variable name
Ralf Baechle [Thu, 17 Nov 2005 14:34:47 +0000 (14:34 +0000)]
git-repack: Fix variable name

Three times remove_redandant -> remove_redundant.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years ago'make clean' forgot about some files
Nicolas Pitre [Thu, 17 Nov 2005 15:40:22 +0000 (10:40 -0500)]
'make clean' forgot about some files

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDeal with binary diff output from GNU diff 2.8.7
Junio C Hamano [Fri, 18 Nov 2005 04:46:29 +0000 (20:46 -0800)]
Deal with binary diff output from GNU diff 2.8.7

Some vintage of diff says just "Files X and Y differ\n", instead
of "Binary files X and Y differ\n", so catch both patterns.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'tojunio' of http://locke.catalyst.net.nz/git/git-martinlanghoff
Junio C Hamano [Thu, 17 Nov 2005 10:00:25 +0000 (02:00 -0800)]
Merge branch 'tojunio' of locke.catalyst.net.nz/git/git-martinlanghoff

19 years agoarchimport: allow for old style branch and public tag names
Martin Langhoff [Thu, 17 Nov 2005 08:20:45 +0000 (21:20 +1300)]
archimport: allow for old style branch and public tag names

This patch adds the -o switch, which lets old trees tracked by
git-archmirror continue working with their old branch and tag names
to make life easier for people tracking your tree.

Private tags that are only used internally by git-archimport continue to be
new-style, and automatically converted upon first run.

[ ml: rebased to skip import overhaul ]

Signed-off-by:: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
19 years agoAdd approxidate test calls.
Junio C Hamano [Tue, 15 Nov 2005 08:07:04 +0000 (00:07 -0800)]
Add approxidate test calls.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit's rev-parse.c function show_datestring presumes gnu date
Linus Torvalds [Tue, 15 Nov 2005 03:29:06 +0000 (19:29 -0800)]
git's rev-parse.c function show_datestring presumes gnu date

Ok. This is the insane patch to do this.

It really isn't very careful, and the reason I call it "approxidate()"
will become obvious when you look at the code. It is very liberal in what
it accepts, to the point where sometimes the results may not make a whole
lot of sense.

It accepts "last week" as a date string, by virtue of "last" parsing as
the number 1, and it totally ignoring superfluous fluff like "ago", so
"last week" ends up being exactly the same thing as "1 week ago". Fine so
far.

It has strange side effects: "last december" will actually parse as "Dec
1", which actually _does_ turn out right, because it will then notice that
it's not December yet, so it will decide that you must be talking about a
date last year. So it actually gets it right, but it's kind of for the
"wrong" reasons.

It also accepts the numbers 1..10 in string format ("one" .. "ten"), so
you can do "ten weeks ago" or "ten hours ago" and it will do the right
thing.

But it will do some really strange thigns too: the string "this will last
forever", will not recognize anyting but "last", which is recognized as
"1", which since it doesn't understand anything else it will think is the
day of the month. So if you do

gitk --since="this will last forever"

the date will actually parse as the first day of the current month.

And it will parse the string "now" as "now", but only because it doesn't
understand it at all, and it makes everything relative to "now".

Similarly, it doesn't actually parse the "ago" or "from now", so "2 weeks
ago" is exactly the same as "2 weeks from now". It's the current date
minus 14 days.

But hey, it's probably better (and certainly faster) than depending on GNU
date. So now you can portably do things like

gitk --since="two weeks and three days ago"
git log --since="July 5"
git-whatchanged --since="10 hours ago"
git log --since="last october"

and it will actually do exactly what you thought it would do (I think). It
will count 17 days backwards, and it will do so even if you don't have GNU
date installed.

(I don't do "last monday" or similar yet, but I can extend it to that too
if people want).

It was kind of fun trying to write code that uses such totally relaxed
"understanding" of dates yet tries to get it right for the trivial cases.
The result should be mixed with a few strange preprocessor tricks, and be
submitted for the IOCCC ;)

Feel free to try it out, and see how many strange dates it gets right. Or
wrong.

And if you find some interesting (and valid - not "interesting" as in
"strange", but "interesting" as in "I'd be interested in actually doing
this) thing it gets wrong - usually by not understanding it and silently
just doing some strange things - please holler.

Now, as usual this certainly hasn't been getting a lot of testing. But my
code always works, no?

Linus

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDisambiguate the term 'branch' in Arch vs git
Eric Wong [Sat, 12 Nov 2005 09:29:20 +0000 (01:29 -0800)]
Disambiguate the term 'branch' in Arch vs git

Disambiguate the term 'branch' in Arch vs git,
and start using fully-qualified names.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
19 years agoarchimport: don't die on merge-base failure
Eric Wong [Sat, 12 Nov 2005 09:27:21 +0000 (01:27 -0800)]
archimport: don't die on merge-base failure

Don't die if we can't find a merge base, Arch allows arbitrary
cherry-picks between unrelated branches and we should not
die when that happens

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
19 years agoremove shellquote usage for tags
Eric Wong [Sat, 12 Nov 2005 09:25:33 +0000 (01:25 -0800)]
remove shellquote usage for tags

use ',' to encode '/' in "archivename/foo--bar--0.0" so we can allow
"--branch"-less trees which are valid in Arch ("archivename/foo--0.0")

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
19 years agodaemon.c: fix arg parsing bugs
Andreas Ericsson [Wed, 16 Nov 2005 23:38:29 +0000 (00:38 +0100)]
daemon.c: fix arg parsing bugs

Allow --init-timeout and --timeout to be specified without falling
through to usage().

Make sure openlog() is called even if implied by --inetd, or messages
will be sent to wherever LOG_USER ends up.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agotests: binary diff application.
Junio C Hamano [Wed, 16 Nov 2005 23:52:45 +0000 (15:52 -0800)]
tests: binary diff application.

This adds more tests to cover cases where binary diff
application succeeds.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agodiff: --full-index
Junio C Hamano [Tue, 15 Nov 2005 01:53:22 +0000 (17:53 -0800)]
diff: --full-index

A new option, --full-index, is introduced to diff family.  This
causes the full object name of pre- and post-images to appear on
the index line of patch formatted output, to be used in
conjunction with --allow-binary-replacement option of git-apply.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoapply: allow-binary-replacement.
Junio C Hamano [Tue, 15 Nov 2005 01:37:05 +0000 (17:37 -0800)]
apply: allow-binary-replacement.

A new option, --allow-binary-replacement, is introduced.

When you feed a diff that records full SHA1 name of pre- and
post-image blob on its index line to git-apply with this option,
the post-image blob replaces the path if what you have in the
working tree matches the pre-image _and_ post-image blob is
already available in the object directory.

Later we _might_ want to enhance the diff output to also include
the full binary data of the post-image, to make this more
useful, but this is good enough for local rebasing application.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-am: --resolved.
Junio C Hamano [Wed, 16 Nov 2005 08:19:32 +0000 (00:19 -0800)]
git-am: --resolved.

After failed patch application, you can manually apply the patch
(this includes resolving the conflicted merge after git-am falls
back to 3-way merge) and run git-update-index on necessary paths
to prepare the index file in a shape a successful patch
application should have produced.  Then re-running git-am --resolved
would record the resulting index file along with the commit log
information taken from the patch e-mail.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-apply: fail if a patch cannot be applied.
Junio C Hamano [Wed, 16 Nov 2005 22:12:56 +0000 (14:12 -0800)]
git-apply: fail if a patch cannot be applied.

Recently we fixed 'git-apply --stat' not to barf on a binary
differences.  But it accidentally broke the error detection when
we actually attempt to apply them.

This commit fixes the problem and adds test cases.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-cvsexportcommit.perl: fix typos in output
Kevin Geiss [Mon, 14 Nov 2005 16:43:51 +0000 (09:43 -0700)]
git-cvsexportcommit.perl: fix typos in output

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-cvsexportcommit.perl: exit with non-0 status if patch fails.
Kevin Geiss [Mon, 14 Nov 2005 16:42:36 +0000 (09:42 -0700)]
git-cvsexportcommit.perl: exit with non-0 status if patch fails.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-cvsexportcommit.perl: use getopts to get binary flags
Kevin Geiss [Mon, 14 Nov 2005 16:41:43 +0000 (09:41 -0700)]
git-cvsexportcommit.perl: use getopts to get binary flags

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-cvsexportcommit.perl: Fix usage() output.
Kevin Geiss [Mon, 14 Nov 2005 16:45:05 +0000 (09:45 -0700)]
git-cvsexportcommit.perl: Fix usage() output.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocumentation/git-log.txt: trivial typo fix.
Alecs King [Wed, 16 Nov 2005 19:06:03 +0000 (03:06 +0800)]
Documentation/git-log.txt: trivial typo fix.

Signed-off-by: Alecs King <alecsk@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agosymref support for import scripts
Pavel Roskin [Wed, 16 Nov 2005 18:27:28 +0000 (13:27 -0500)]
symref support for import scripts

Fix git import script not to assume that .git/HEAD is a symlink.

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDisallow empty pattern in "git grep"
Linus Torvalds [Wed, 16 Nov 2005 17:38:46 +0000 (09:38 -0800)]
Disallow empty pattern in "git grep"

For some reason I've done a "git grep" twice with no pattern, which is
really irritating, since it just grep everything. If I actually wanted
that, I could do "git grep ^" or something.

So add a "usage" message if the pattern is empty.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit wrapper: basic fixes.
Junio C Hamano [Wed, 16 Nov 2005 07:13:30 +0000 (23:13 -0800)]
git wrapper: basic fixes.

Updates to fix the nits found during the list discussion.

 - Lose PATH_TO_MAN; just rely on execlp() to find whereever the
   "man" command is installed.

 - Do not randomly chdir(), but concatenate to the current
   working directory only if the given path is not absolute.

 - Lose use of glob(); read from exec_path and do sorting
   ourselves -- it is not that much more work.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoGive python a chance to find "backported" modules
Johannes Schindelin [Wed, 16 Nov 2005 02:33:44 +0000 (03:33 +0100)]
Give python a chance to find "backported" modules

python 2.2.1 is perfectly capable of executing git-merge-recursive,
provided that it finds heapq and sets. All you have to do is to steal
heapq.py and sets.py from python 2.3 or newer, and drop them in your
GIT_PYTHON_PATH.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix tests with new git in C
Johannes Schindelin [Wed, 16 Nov 2005 01:44:50 +0000 (02:44 +0100)]
Fix tests with new git in C

GIT_EXEC_PATH *has* to be set.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix llist_sorted_difference_inplace in git-pack-redundant
Lukas Sandström [Tue, 15 Nov 2005 21:24:02 +0000 (22:24 +0100)]
Fix llist_sorted_difference_inplace in git-pack-redundant

Simplify and actually make llist_sorted_difference_inplace work
by using llist_sorted_remove instead of duplicating parts of the
code.

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit --help COMMAND brings up the git-COMMAND man-page.
Andreas Ericsson [Tue, 15 Nov 2005 23:31:25 +0000 (00:31 +0100)]
git --help COMMAND brings up the git-COMMAND man-page.

It's by design a bit stupid (matching ^git rather than ^git-), so as
to work with 'gitk' and 'git' as well.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate git(7) man-page for the C wrapper.
Andreas Ericsson [Tue, 15 Nov 2005 23:31:25 +0000 (00:31 +0100)]
Update git(7) man-page for the C wrapper.

The program 'git' now has --exec-path which needs explaining.

Renamed old "DESCRIPTION" to "CORE GIT COMMANDS" to make room for
"OPTIONS" while following follow some sort of convention.

Also updated AUTHORS section to pat my own back a bit.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoC implementation of the 'git' program, take two.
Andreas Ericsson [Tue, 15 Nov 2005 23:31:25 +0000 (00:31 +0100)]
C implementation of the 'git' program, take two.

This patch provides a C implementation of the 'git' program and
introduces support for putting the git-* commands in a directory
of their own. It also saves some time on executing those commands
in a tight loop and it prints the currently available git commands
in a nicely formatted list.

The location of the GIT_EXEC_PATH (name discussion's closed, thank gods)
can be obtained by running

git --exec-path

which will hopefully give porcelainistas ample time to adapt their
heavy-duty loops to call the core programs directly and thus save
the extra fork() / execve() overhead, although that's not really
necessary any more.

The --exec-path value is prepended to $PATH, so the git-* programs
should Just Work without ever requiring any changes to how they call
other programs in the suite.

Some timing values for 10000 invocations of git-var >&/dev/null:
git.sh: 24.194s
git.c:   9.044s
git-var: 7.377s

The git-<tab><tab> behaviour can, along with the someday-to-be-deprecated
git-<command> form of invocation, be indefinitely retained by adding
the following line to one's .bash_profile or equivalent:

PATH=$PATH:$(git --exec-path)

Experimental libraries can be used by either setting the environment variable
GIT_EXEC_PATH, or by using

git --exec-path=/some/experimental/exec-path

Relative paths are properly grok'ed as exec-path values.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agosha1_file.c::add_packed_git(): fix type mismatch.
Junio C Hamano [Tue, 15 Nov 2005 20:51:02 +0000 (12:51 -0800)]
sha1_file.c::add_packed_git(): fix type mismatch.

An object name is 20-byte 'unsigned char', not 'char'.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agodiff: make default rename detection limit configurable.
Junio C Hamano [Tue, 15 Nov 2005 20:48:08 +0000 (12:48 -0800)]
diff: make default rename detection limit configurable.

A while ago, a rename-detection limit logic was implemented as a
response to this thread:

http://marc.theaimsgroup.com/?l=git&m=112413080630175

where gitweb was found to be using a lot of time and memory to
detect renames on huge commits.  git-diff family takes -l<num>
flag, and if the number of paths that are rename destination
candidates (i.e. new paths with -M, or modified paths with -C)
are larger than that number, skips rename/copy detection even
when -M or -C is specified on the command line.

This commit makes the rename detection limit easier to use.  You
can have:

[diff]
renamelimit = 30

in your .git/config file to specify the default rename detection
limit.  You can override this from the command line; giving 0
means 'unlimited':

git diff -M -l0

We might want to change the default behaviour, when you do not
have the configuration, to limit it to say 20 paths or so.  This
would also help the diffstat generation after a big 'git pull'.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd config variable core.symrefsonly
Johannes Schindelin [Tue, 15 Nov 2005 18:24:19 +0000 (19:24 +0100)]
Add config variable core.symrefsonly

This allows you to force git to avoid symlinks for refs. Just add
something like

[core]
symrefsonly = true

to .git/config.

Don´t forget to "git checkout your_branch", or it does not do anything...

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRework object refs tracking to reduce memory usage
Sergey Vlasov [Tue, 15 Nov 2005 16:08:08 +0000 (19:08 +0300)]
Rework object refs tracking to reduce memory usage

Store pointers to referenced objects in a variable sized array instead
of linked list.  This cuts down memory usage of utilities which use
object references; e.g., git-fsck-objects --full on the git.git
repository consumes about 2 MB of memory tracked by Massif instead of
7 MB before the change.  Object refs are still the biggest consumer of
memory (57%), but the malloc overhead for a single block instead of a
linked list is substantially smaller.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-fsck-objects: Free tree entries after use
Sergey Vlasov [Tue, 15 Nov 2005 16:07:15 +0000 (19:07 +0300)]
git-fsck-objects: Free tree entries after use

The Massif tool of Valgrind revealed that parsed tree entries occupy
more than 60% of memory allocated by git-fsck-objects.  These entries
can be freed immediately after use, which significantly decreases
memory consumption.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix git(1) link to git-index-pack
Jonas Fonseca [Tue, 15 Nov 2005 12:29:18 +0000 (13:29 +0100)]
Fix git(1) link to git-index-pack

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocumentation: do not blindly run 'cat' .git/HEAD, or echo into it.
Junio C Hamano [Tue, 15 Nov 2005 09:31:04 +0000 (01:31 -0800)]
Documentation: do not blindly run 'cat' .git/HEAD, or echo into it.

Many places in the documentation we still talked about reading
what commit is recorded in .git/HEAD or writing the new head
information into it, both assuming .git/HEAD is a symlink.  That
is not necessarily so.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix bunch of fd leaks in http-fetch
Petr Baudis [Fri, 11 Nov 2005 23:55:16 +0000 (00:55 +0100)]
Fix bunch of fd leaks in http-fetch

The current http-fetch is rather careless about fd leakage, causing
problems while fetching large repositories. This patch does not reserve
exhaustiveness, but I covered everything I spotted. I also left some
safeguards in place in case I missed something, so that we get to know,
sooner or later.

Reported by Becky Bruce <becky.bruce@freescale.com>.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-daemon: --inetd implies --syslog
Andreas Ericsson [Mon, 14 Nov 2005 16:41:01 +0000 (17:41 +0100)]
git-daemon: --inetd implies --syslog

Otherwise nothing is logged anywhere, which is a Bad Thing.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-core-foo -> git-foo, except the core package
H. Peter Anvin [Tue, 15 Nov 2005 06:12:17 +0000 (22:12 -0800)]
git-core-foo -> git-foo, except the core package

This patch renames the tarball "git" rather than "git-core", and changes
the names of various packages from git-core-foo to git-foo.  git-core is
still the true core package; an empty RPM package named "git" pulls in
ALL the git packages -- this makes updates work correctly, and allows
"yum install git" to do the obvious thing.

It also renames the git-(core-)tk package to gitk.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate topo-order test.
Junio C Hamano [Tue, 15 Nov 2005 02:15:30 +0000 (18:15 -0800)]
Update topo-order test.

The recently we updated rev-list --topo-order to show the heads
in date order, but we had a test that expected to see the old
behaviour.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoallow git-update-ref create refs with slashes in names
Alex Riesen [Mon, 14 Nov 2005 22:10:59 +0000 (23:10 +0100)]
allow git-update-ref create refs with slashes in names

Make git-update-ref create references with slashes in them. git-branch
and git-checkout already support such reference names.

git-branch can use git-update-ref to create the references in a more
formal manner now.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocumentation nitpicking
Nikolai Weibull [Mon, 14 Nov 2005 23:20:01 +0000 (00:20 +0100)]
Documentation nitpicking

This patch fixes some small problems with the documentation.

Signed-off-by: Nikolai Weibull <nikolai@bitwi.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocument a couple of missing command-line options.
Nikolai Weibull [Mon, 14 Nov 2005 23:20:01 +0000 (00:20 +0100)]
Document a couple of missing command-line options.

This patch adds documentation to quite a few command-line options.

Signed-off-by: Nikolai Weibull <nikolai@bitwi.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocument the -n command-line option to git-unpack-objects
Nikolai Weibull [Mon, 14 Nov 2005 23:20:01 +0000 (00:20 +0100)]
Document the -n command-line option to git-unpack-objects

This patch documents the -n command-line option to git-unpack-objects,
as it was previously undocumented.

Signed-off-by: Nikolai Weibull <nikolai@bitwi.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoapply: fix binary patch detection.
Junio C Hamano [Tue, 15 Nov 2005 01:15:07 +0000 (17:15 -0800)]
apply: fix binary patch detection.

The comparison to find "Binary files " string was looking at a
wrong place when offset != 0.

Also, we may have the full 40-byte textual sha1 on the index
line; two off-by-one errors prevented it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix git-rev-list "date order" with --topo-order
Linus Torvalds [Mon, 14 Nov 2005 18:01:26 +0000 (10:01 -0800)]
Fix git-rev-list "date order" with --topo-order

This fixes git-rev-list so that when there are multiple branches, we still
sort the heads in proper approximate date order even when sorting the
output topologically.

This makes things like

gitk --all -d

work sanely and show the branches in date order (where "date order" is
obviously modified by the paren-child dependency requirements of the
topological sort).

The trivial fix is to just build the "work" list in date order rather than
inserting the new work entries at the beginning.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-branch: Mention -d and -D in man-page.
Andreas Ericsson [Mon, 14 Nov 2005 16:53:42 +0000 (17:53 +0100)]
git-branch: Mention -d and -D in man-page.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'svnup' of http://netz.smurf.noris.de/git/git
Junio C Hamano [Mon, 14 Nov 2005 21:50:05 +0000 (13:50 -0800)]
Merge branch 'svnup' of netz.smurf.noris.de/git/git

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDepend on asciidoc 7 (at least).
Matthias Urlichs [Mon, 14 Nov 2005 16:41:31 +0000 (17:41 +0100)]
Depend on asciidoc 7 (at least).

19 years agoRemove git-rename. git-mv does the same
Josef Weidendorfer [Sun, 13 Nov 2005 14:08:00 +0000 (15:08 +0100)]
Remove git-rename. git-mv does the same

Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoBugfix: stop if directory already exists
Josef Weidendorfer [Sun, 13 Nov 2005 14:03:31 +0000 (15:03 +0100)]
Bugfix: stop if directory already exists

Fix a typo: We do not want to run the directory as command,
and want to terminate if the directory exists
Additionally, update the usage message

Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSeparate LDFLAGS and CFLAGS.
Junio C Hamano [Sun, 13 Nov 2005 09:46:13 +0000 (01:46 -0800)]
Separate LDFLAGS and CFLAGS.

Stuffing -L flag and friends meant for the linking phase into
ALL_CFLAGS is not right; honor LDFLAGS and introduce ALL_LDFLAGS
to separate them out.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'svnup' of http://netz.smurf.noris.de/git/git
Junio C Hamano [Mon, 14 Nov 2005 08:25:48 +0000 (00:25 -0800)]
Merge branch 'svnup' of netz.smurf.noris.de/git/git

19 years agoRemove trailing slashes
Matthias Urlichs [Mon, 14 Nov 2005 07:31:00 +0000 (08:31 +0100)]
Remove trailing slashes

SVN dies a messy death when passed a path with trailing slashes.

19 years agoRename lost+found to lost-found.
Junio C Hamano [Sun, 13 Nov 2005 10:07:02 +0000 (02:07 -0800)]
Rename lost+found to lost-found.

Because we use "lost-found" as the directory name to hold
dangling object names, it is confusing to call the command
git-lost+found, although it makes sense and is even cute ;-).

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix compilation warnings in pack-redundant.c
Kai Ruemmler [Sat, 12 Nov 2005 16:33:24 +0000 (17:33 +0100)]
Fix compilation warnings in pack-redundant.c

This fixes compilation warnings where "%ld" was used to print values of
type size_t.

Signed-off-by: Kai Ruemmler <kai.ruemmler@gmx.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDebian: build-depend on libexpat-dev.
Junio C Hamano [Sun, 13 Nov 2005 09:57:33 +0000 (01:57 -0800)]
Debian: build-depend on libexpat-dev.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSplit gitk into seperate RPM package
Thomas Matysik [Sun, 13 Nov 2005 04:58:02 +0000 (17:58 +1300)]
Split gitk into seperate RPM package

I don't want to have to install x11-libs and all it's dependencies on
my headless machines, so this patch splits gitk out of the RPM.

The .deb already appears to have gitk split out.

Signed-off-by: Thomas Matysik <thomas@matysik.co.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd expat and expat-devel dependencies (for http-push) to RPM spec.
Thomas Matysik [Sun, 13 Nov 2005 04:56:55 +0000 (17:56 +1300)]
Add expat and expat-devel dependencies (for http-push) to RPM spec.

Signed-off-by: Thomas Matysik <thomas@matysik.co.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix fd leak in http-fetch
Nick Hengeveld [Sat, 12 Nov 2005 17:38:28 +0000 (09:38 -0800)]
Fix fd leak in http-fetch

Added a call to finish_request to clean up resources if the server
returned a 404 and there are no alternates left to try.

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix for multiple alternates requests in http-fetch
Nick Hengeveld [Sat, 12 Nov 2005 17:11:32 +0000 (09:11 -0800)]
Fix for multiple alternates requests in http-fetch

Stop additional alternates requests from starting if one is already in
progress.  This adds an optional callback which is processed after a slot
has finished running.

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agodebian packaging: git-cvs needs cvsps
Matthias Urlichs [Sat, 12 Nov 2005 22:15:50 +0000 (23:15 +0100)]
debian packaging: git-cvs needs cvsps

19 years agoIgnore built git-lost+found.
Junio C Hamano [Sat, 12 Nov 2005 06:28:03 +0000 (22:28 -0800)]
Ignore built git-lost+found.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocumentation: asciidoc sources are utf-8
Junio C Hamano [Sat, 12 Nov 2005 06:16:59 +0000 (22:16 -0800)]
Documentation: asciidoc sources are utf-8

git-pack-redundant documentation was encoded in latin1, which caused
asciidoc to barf, which expected to see utf-8.  Run tcs to re-encode
it in utf-8.

Also just for fun try my name in Japanese in git-lost+found
documentation ;-)

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRename .git/lost+found to .git/lost-found.
Junio C Hamano [Fri, 11 Nov 2005 03:16:26 +0000 (19:16 -0800)]
Rename .git/lost+found to .git/lost-found.

Just to avoid confusion that scripts poorly written by somebody
else ;-) might mistake this as a mount point, or backup tools
ignoring the directory.  The latter is probably not a big loss,
however, considering that this directory's contents are to be
used while fresh anyway.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-lost+found
Junio C Hamano [Mon, 7 Nov 2005 23:15:34 +0000 (15:15 -0800)]
git-lost+found

This command helps you resurrect accidentally lost tags and commits.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-prune: prune redundant packs
Junio C Hamano [Fri, 11 Nov 2005 18:41:53 +0000 (10:41 -0800)]
git-prune: prune redundant packs

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMake git-pack-redundant consider alt-odbs
Lukas_Sandström [Fri, 11 Nov 2005 00:25:04 +0000 (01:25 +0100)]
Make git-pack-redundant consider alt-odbs

This patch changes git-pack-redundant so that packfiles
in alternate object directories also are considered when
deciding which objects are redundant.

This functionality is controlled by the flag '--alt-odb'.

Also convert the other flags to the long form, and update
docs and git-repack accordingly.

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRename git-pack-intersect to git-pack-redundant
Lukas_Sandström [Wed, 9 Nov 2005 23:16:13 +0000 (00:16 +0100)]
Rename git-pack-intersect to git-pack-redundant

This patch renames git-pack-intersect to git-pack-redundant
as suggested by Petr Baudis. The new name reflects what the
program does, rather than how it does it.

Also fix a small argument parsing bug.

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMake git-repack use git-pack-intersect.
Lukas_Sandström [Wed, 9 Nov 2005 01:25:48 +0000 (02:25 +0100)]
Make git-repack use git-pack-intersect.

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd git-pack-intersect to .gitignore
Lukas_Sandström [Wed, 9 Nov 2005 01:24:51 +0000 (02:24 +0100)]
Add git-pack-intersect to .gitignore

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd documentation for git-pack-intersect
Lukas_Sandström [Wed, 9 Nov 2005 01:23:55 +0000 (02:23 +0100)]
Add documentation for git-pack-intersect

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd git-pack-intersect
Lukas_Sandström [Wed, 9 Nov 2005 01:22:40 +0000 (02:22 +0100)]
Add git-pack-intersect

This patch adds the program git-pack-intersect. It is
used to find redundant packs in git repositories.

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-one-file: use empty- or common-base condintionally in two-stage merge.
Junio C Hamano [Fri, 11 Nov 2005 03:30:23 +0000 (19:30 -0800)]
merge-one-file: use empty- or common-base condintionally in two-stage merge.

If two sides added the same path completely different thing, it is
easier to see the merge pivoting on /dev/null.  So check the size of
the common section we have found, and empty it if it is too small.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocumentation: git-apply --no-add
Junio C Hamano [Thu, 10 Nov 2005 04:53:54 +0000 (20:53 -0800)]
Documentation: git-apply --no-add

This is a specialized hack to help no-base merges, but other
people might find it useful, so let's document it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-one-file: use common as base, instead of emptiness.
Junio C Hamano [Thu, 10 Nov 2005 04:38:33 +0000 (20:38 -0800)]
merge-one-file: use common as base, instead of emptiness.

Unlike the previous round that merged the path added differently
in each branches using emptiness as the base, compute a common
version and use it as input to 'merge' program.

This would show the resulting (still conflicting) file left in
the working tree as:

common file contents...
<<<<<< FILENAME
version from our branch...
======
version from their branch...
>>>>>> .merge_file_XXXXXX
more common file contents...

when both sides added similar contents.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge with /dev/null as base, instead of punting O==empty case
Junio C Hamano [Tue, 8 Nov 2005 06:03:46 +0000 (22:03 -0800)]
Merge ... /dev/null as base, instead of punting O==empty case

Instead of leaving the path unmerged in a case where each side
adds different version of the same path, attempt to merge it
with empty base and leave "our" version in the index file, just
like we do for the case in conflicting merge.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoShow URL in the "Getting <foo> list" http-fetch messages
Petr Baudis [Sat, 12 Nov 2005 00:49:59 +0000 (01:49 +0100)]
Show URL in the "Getting <foo> list" http-fetch messages

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-recursive: Use '~' instead of '_' to separate file names from branch names
Fredrik Kuivinen [Fri, 11 Nov 2005 23:55:36 +0000 (00:55 +0100)]
merge-recursive: Use '~' instead of '_' to separate file names from branch names

Makes it less probable that we get a clash with an existing file,
furthermore Cogito already uses '~' for this purpose.

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-recursive: Add copyright notice
Fredrik Kuivinen [Fri, 11 Nov 2005 23:54:55 +0000 (00:54 +0100)]
merge-recursive: Add copyright notice

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-recursive: Indent the output properly
Fredrik Kuivinen [Fri, 11 Nov 2005 23:53:07 +0000 (00:53 +0100)]
merge-recursive: Indent the output properly

If we have multiple common ancestors and have to recursively merge
them then the output will be much more readable with this commit.

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoChange 'cache' to 'index' in the docs
Lukas_Sandström [Fri, 11 Nov 2005 01:12:27 +0000 (02:12 +0100)]
Change 'cache' to 'index' in the docs

This patch makes the documentation refer to the index
as index instead of cache, but some references still
remain. (e.g. git-update-index.txt)

Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoINSTALL: duplicate python requirements from Makefile
Junio C Hamano [Fri, 11 Nov 2005 19:27:03 +0000 (11:27 -0800)]
INSTALL: duplicate python requirements from Makefile

... and refer the reader to Makefile for other things that can be
tweaked.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branches 'jc/sb' and 'jc/mb'
Junio C Hamano [Fri, 11 Nov 2005 18:52:59 +0000 (10:52 -0800)]
Merge branches 'jc/sb' and 'jc/mb'

19 years agot1200: use --topo-order to keep the show-branch output stable.
Junio C Hamano [Fri, 11 Nov 2005 02:27:44 +0000 (18:27 -0800)]
t1200: use --topo-order to keep the show-branch output stable.

Because a batch-oriented script creates many commits within a second
on a fast machine, show-branch output of the test results are unstable
without topo-order.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFully detect uninteresting commits.
Junio C Hamano [Thu, 10 Nov 2005 23:47:58 +0000 (15:47 -0800)]
Fully detect uninteresting commits.

With the change in the previous round, we are guaranteed to come up
with the list of all relevant merge bases, but sometimes we do not
fully mark unintersting ones due to a horizon effect.  Add a phase to
postprocess, so that we mark all ancestor of "interesting" commit.

This also changes the default ordering of shown commits back to
chronological order, and adds --topo-order flag to show them in
topological order.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-show-branch: tighten merge-base computation.
Junio C Hamano [Thu, 10 Nov 2005 07:36:15 +0000 (23:36 -0800)]
git-show-branch: tighten merge-base computation.

This makes the merge-base computation resistant to the pathological
case discussed on the list earlier, by doing the same logic as
git-merge-base.  As a side effect, it breaks the command's primary
function to list non-merge commit sequences, which needs to be fixed
separately.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd test case for merge-base.
Junio C Hamano [Fri, 11 Nov 2005 09:00:52 +0000 (01:00 -0800)]
Add test case for merge-base.

Although it was shown that the "full contamination" was not really full
during the list discussion, the series improves things without incurring
extra parsing cost, and here is a test to check that.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-base: avoid unnecessary postprocessing.
Junio C Hamano [Fri, 11 Nov 2005 06:41:44 +0000 (22:41 -0800)]
merge-base: avoid unnecessary postprocessing.

When we have only one merge-base candidates in the result list,
there is no point going back to mark the reachable commits
again.  And that is the most common case, so try not to waste
time on it.  Suggested by Linus.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomerge-base: fully contaminate the well.
Junio C Hamano [Fri, 11 Nov 2005 01:21:54 +0000 (17:21 -0800)]
merge-base: fully contaminate the well.

The discussion on the list demonstrated a pathological case where
an ancestor of a merge-base can be left interesting.  This commit
introduces a postprocessing phase to fix it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRPM: arch submodule needs tla.
Junio C Hamano [Fri, 11 Nov 2005 17:31:38 +0000 (09:31 -0800)]
RPM: arch submodule needs tla.

Noticed by Horst von Brand.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-clone: Allow cloning into directories other than child of current dir.
Andreas Ericsson [Thu, 10 Nov 2005 11:58:08 +0000 (12:58 +0100)]
git-clone: Allow cloning into directories other than child of current dir.

This patch adds -p to mkdir and an explicit check to see if the target
directory exists (since mkdir -p doesn't throw an error if it does).

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-clone: Keep remote names when cloning unless explicitly told not to.
Andreas Ericsson [Thu, 10 Nov 2005 11:58:08 +0000 (12:58 +0100)]
git-clone: Keep remote names when cloning unless explicitly told not to.

With this patch the following commands all clone into the local directory
"repo". If repo exists, it will still barf.

git-clone git://host.xz/repo.git
git-clone /path/to/repo/.git
git-clone host.xz:repo.git

I ended up doing the same source-to-target sed'ing for all our company
projects, so it was easier to add it directly to git-clone.

Signed-off-by: Andreas Ericsson <ae@op5.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agospecfile cleanups
Chris Wright [Fri, 11 Nov 2005 09:10:29 +0000 (01:10 -0800)]
specfile cleanups

Some specfile cleanups after the split.
- zlib dependency fix, current method is inconsistent, you can
  potentially build a package that you can't install on machine you
  built it on
- Add proper defattr
- Remove trailing '.' in summary
- Add docs to split up packages
- Add git-core dependency for each subpackage
- Move arch import to separate package as well

Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoarchimport: handle pika escaping
Martin Langhoff [Fri, 11 Nov 2005 05:00:57 +0000 (18:00 +1300)]
archimport: handle pika escaping

Arch uses pika escaping in some places (but not all!). Specifically, commits of
the type 'patch' use pika escaping in the log entries, which we parse to know
what to add/delete and what to commit.

This patch checks for hints of pika escaping and asks tla to unescape for us.

Originally implemented by Penny Leach <penny@catalyst.net.nz>

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>