git.git
18 years agorevamp git-clone (take #2).
Junio C Hamano [Tue, 21 Mar 2006 08:14:13 +0000 (00:14 -0800)]
revamp git-clone (take #2).

This builds on top of the previous one.

 * --use-separate-remote uses .git/refs/remotes/$origin/
   directory to keep track of the upstream branches.

 * The $origin above defaults to "origin" as usual, but the
   existing "-o $origin" option can be used to override it.

I am not yet convinced if we should make "$origin" the synonym to
"refs/remotes/$origin/$name" where $name is the primary branch
name of $origin upstream, nor if so how we should decide which
upstream branch is the primary one, but that is more or less
orthogonal to what the clone does here.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorevamp git-clone.
Junio C Hamano [Mon, 20 Mar 2006 08:21:10 +0000 (00:21 -0800)]
revamp git-clone.

This does two things.

 * A new flag --reference can be used to name a local repository
   that is to be used as an alternate.  This is in response to
   an inquiry by James Cloos in the message on the list
   <m3r74ykue7.fsf@lugabout.cloos.reno.nv.us>.

 * A new flag --use-separate-remote stops contaminating local
   branch namespace by upstream branch names.  The upstream
   branch heads are copied in .git/refs/remotes/ instead of
   .git/refs/heads/ and .git/remotes/origin file is set up to
   reflect this as well.  It requires to have fetch/pull update
   to understand .git/refs/remotes by Eric Wong to further
   update the repository cloned this way.

For the former change, git-fetch-pack is taught a new flag --all
to fetch from all the remote heads.  Nobody uses the git-clone-pack
with this change, so we could deprecate the command, but removal
of the command will be left to a separate round.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agofetch,parse-remote,fmt-merge-msg: refs/remotes/* support
Eric Wong [Fri, 10 Mar 2006 12:19:07 +0000 (04:19 -0800)]
fetch,parse-remote,fmt-merge-msg: refs/remotes/* support

We can now easily fetch and merge things from heads in the
refs/remotes/ hierarchy in remote repositories.

The refs/remotes/ hierarchy is likely to become the standard for
tracking foreign SCMs, as well as the location of Pull: targets
for tracking remote branches in newly cloned repositories.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd git-imap-send, derived from isync 1.0.1.
Mike McCormack [Fri, 10 Mar 2006 05:32:50 +0000 (14:32 +0900)]
Add git-imap-send, derived from isync 1.0.1.

git-imap-send drops a patch series generated by git-format-patch into an
IMAP folder. This allows patch submitters to send patches through their
own mail program.

git-imap-send uses the following values from the GIT repository
configuration:

The target IMAP folder:

[imap]
         Folder = "INBOX.Drafts"

A command to open an ssh tunnel to the imap mail server.

[imap]
         Tunnel = "ssh -q user@imap.server.com /usr/bin/imapd ./Maildir
2> /dev/null"

[imap]
         Host = imap.server.com
         User = bob
         Password = pwd
         Port = 143

18 years agorepack: prune loose objects when -d is given
Junio C Hamano [Sat, 11 Mar 2006 06:05:47 +0000 (22:05 -0800)]
repack: prune loose objects when -d is given

[jc: the request originally came from Martin Atukunda, which was
 improved further by Alex Riesen]

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotry_to_simplify_commit(): do not skip inspecting tree change at boundary.
Junio C Hamano [Sat, 11 Mar 2006 05:59:37 +0000 (21:59 -0800)]
try_to_simplify_commit(): do not skip inspecting tree change at boundary.

When git-rev-list (and git-log) collapsed ancestry chain to
commits that touch specified paths, we failed to inspect and
notice tree changes when we are about to hit uninteresting
parent.  This resulted in "git rev-list since.. -- file" to
always show the child commit after the lower bound, even if it
does not touch the file.  This commit fixes it.

Thanks for Catalin for reporting this.

See also:
461cf59f8924f174d7a0dcc3d77f576d93ed29a4

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix t1200 test for breakage caused by removal of full-stop at the end of fast-forward...
Junio C Hamano [Fri, 10 Mar 2006 06:51:21 +0000 (22:51 -0800)]
Fix t1200 test for breakage caused by removal of full-stop at the end of fast-forward message.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDescribe how to add extra mail header lines in mail generated by git-format-patch.
Mike McCormack [Fri, 10 Mar 2006 04:47:59 +0000 (13:47 +0900)]
Describe how to add extra mail header lines in mail generated by git-format-patch.

18 years agoDocument the --attach flag.
Mike McCormack [Fri, 10 Mar 2006 04:47:24 +0000 (13:47 +0900)]
Document the --attach flag.

18 years agoallow double click on current HEAD id after git-pull
Olaf Hering [Sat, 11 Feb 2006 11:26:30 +0000 (12:26 +0100)]
allow double click on current HEAD id after git-pull

Double click on to current HEAD commit id is not possible,
the dot has to go.

[jc: by popular requests.]

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorefs.c::do_for_each_ref(): Finish error message lines with "\n"
Junio C Hamano [Thu, 9 Mar 2006 20:59:16 +0000 (12:59 -0800)]
refs.c::do_for_each_ref(): Finish error message lines with "\n"

We used fprintf() to show an error message without terminating
it with LF; use error() for that.

cf. c401cb48e77459a4ccad76888ad31bef252facc5

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoNicer output from 'git'
Fredrik Kuivinen [Thu, 9 Mar 2006 16:24:19 +0000 (17:24 +0100)]
Nicer output from 'git'

[jc: with suggestions by Jan-Benedict Glaw]

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUse #define ARRAY_SIZE(x) (sizeof(x)/sizeof(x[0]))
Junio C Hamano [Thu, 9 Mar 2006 19:58:05 +0000 (11:58 -0800)]
Use #define ARRAY_SIZE(x) (sizeof(x)/sizeof(x[0]))

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRemove trailing dot after short description
Fredrik Kuivinen [Thu, 9 Mar 2006 16:24:50 +0000 (17:24 +0100)]
Remove trailing dot after short description

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix some inconsistencies in the docs
Fredrik Kuivinen [Thu, 9 Mar 2006 16:24:37 +0000 (17:24 +0100)]
Fix some inconsistencies in the docs

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: fix a harmless warning on rebuild (with old repos)
Eric Wong [Thu, 9 Mar 2006 11:52:48 +0000 (03:52 -0800)]
contrib/git-svn: fix a harmless warning on rebuild (with old repos)

It's only for repositories that were imported with very early
versions of git-svn.  Unfortunately, some of those repos are out
in the wild already, so fix this warning.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: remove the --no-stop-on-copy flag
Eric Wong [Thu, 9 Mar 2006 11:50:34 +0000 (03:50 -0800)]
contrib/git-svn: remove the --no-stop-on-copy flag

Output a big warning if somebody actually has a pre-1.0 version
of svn that doesn't support it.

Thanks to Yann Dirson for reminding me it still existed
and attempting to re-enable it :)

I think I subconciously removed support for it earlier...

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: fix svn compat and fetch args
Eric Wong [Thu, 9 Mar 2006 11:48:47 +0000 (03:48 -0800)]
contrib/git-svn: fix svn compat and fetch args

'svn info' doesn't work with URLs in svn <= 1.1.  Now we
only run svn info in local directories.

As a side effect, this should also work better for 'init' off
directories that are no longer in the latest revision of the
repository.

svn checkout -r<revision> arguments are fixed.
Newer versions of svn (1.2.x) seem to need URL@REV as well as
-rREV to checkout a particular revision...

Add an example in the manpage of how to track directory that has
been moved since its initial revision.

A huge thanks to Yann Dirson for the bug reporting and testing
my original patch.  Thanks also to Junio C Hamano for suggesting
a safer way to use git-rev-parse.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDon't recurse into parents marked uninteresting.
Matthias Urlichs [Thu, 9 Mar 2006 04:04:36 +0000 (05:04 +0100)]
Don't recurse into parents marked uninteresting.

revision.c:make_parents_uninteresting() is exponential with the number
of merges in the tree. That's fine -- unless some other part of git
already has pulled the whole commit tree into memory ...

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff-delta: bound hash list length to avoid O(m*n) behavior
Nicolas Pitre [Wed, 8 Mar 2006 19:32:50 +0000 (14:32 -0500)]
diff-delta: bound hash list length to avoid O(m*n) behavior

The diff-delta code can exhibit O(m*n) behavior with some patological
data set where most hash entries end up in the same hash bucket.

To prevent this, a limit is imposed to the number of entries that can
exist in the same hash bucket.

Because of the above the code is a tiny bit more expensive on average,
even if some small optimizations were added as well to atenuate the
overhead. But the problematic samples used to diagnoze the issue are now
orders of magnitude less expensive to process with only a slight loss in
compression.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotest-delta needs zlib to compile
Nicolas Pitre [Wed, 8 Mar 2006 18:19:19 +0000 (13:19 -0500)]
test-delta needs zlib to compile

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-fmt-merge-msg cleanup
Linus Torvalds [Thu, 9 Mar 2006 01:56:07 +0000 (17:56 -0800)]
git-fmt-merge-msg cleanup

Since I've started using the "merge.summary" flag in my repo, my merge
messages look nicer, but I dislike how I get notifications of merges
within merges.

So I'd suggest this trivial change..

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorepo-config: give value_ a sane default so regexec won't segfault
Jonas Fonseca [Mon, 6 Mar 2006 05:23:30 +0000 (06:23 +0100)]
repo-config: give value_ a sane default so regexec won't segfault

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUpdate http-push functionality
Nick Hengeveld [Wed, 8 Mar 2006 00:13:20 +0000 (16:13 -0800)]
Update http-push functionality

This brings http-push functionality more in line with the ssh/git version,
by borrowing bits from send-pack and rev-list to process refspecs and
revision history in more standard ways.  Also, the status of remote objects
is determined using PROPFIND requests for the object directory rather than
HEAD requests for each object - while it may be less efficient for small
numbers of objects, this approach is able to get the status of all remote
loose objects in a maximum of 256 requests.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsimport: Remove master-updating code
Matthias Urlichs [Tue, 7 Mar 2006 09:08:34 +0000 (10:08 +0100)]
cvsimport: Remove master-updating code

The code which tried to update the master branch was somewhat broken.
=> People should do that manually, with "git merge".

Signed-off-by: Matthias Urlichs <smurf@smurf.noris.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'sp/checkout'
Junio C Hamano [Tue, 7 Mar 2006 04:58:17 +0000 (20:58 -0800)]
Merge branch 'sp/checkout'

* sp/checkout:
  Add --temp and --stage=all options to checkout-index.

18 years agoMerge branch 'fd/asciidoc'
Junio C Hamano [Tue, 7 Mar 2006 04:51:23 +0000 (20:51 -0800)]
Merge branch 'fd/asciidoc'

* fd/asciidoc:
  Tweak asciidoc output to work with broken docbook-xsl

18 years agoAllow format-patch to attach patches
Mike McCormack [Mon, 6 Mar 2006 13:12:12 +0000 (22:12 +0900)]
Allow format-patch to attach patches

The --attach patch to git-format-patch to attach patches instead of
inlining them.  Some mailers linewrap inlined patches (eg. Mozilla).

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAllow adding arbitary lines in the mail header generated by format-patch.
Mike McCormack [Mon, 6 Mar 2006 13:10:21 +0000 (22:10 +0900)]
Allow adding arbitary lines in the mail header generated by format-patch.

Entries may be added to the config file as follows:

[format]
         headers = "Organization: CodeWeavers\nTo: wine-patches
<wine-patches@winehq.org>\n"

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoannotate-blame: tests incomplete lines.
Junio C Hamano [Mon, 6 Mar 2006 08:41:17 +0000 (00:41 -0800)]
annotate-blame: tests incomplete lines.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoblame: unbreak "diff -U 0".
Junio C Hamano [Mon, 6 Mar 2006 08:32:50 +0000 (00:32 -0800)]
blame: unbreak "diff -U 0".

The commit 604c86d15bb319a1e93ba218fca48ce1c500ae52 changed the
original "diff -u0" to "diff -u -U 0" for portability.

A big mistake without proper testing.

The form "diff -u -U 0" shows the default 3-line contexts,
because -u and -U 0 contradicts with each other; "diff -U 0" (or
its longhand "diff --unified=0") is what we meant.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTweak asciidoc output to work with broken docbook-xsl
Francis Daly [Sun, 5 Mar 2006 23:13:36 +0000 (23:13 +0000)]
Tweak asciidoc output to work with broken docbook-xsl

docbook-xsl v1.68 incorrectly converts "<screen>" from docbook to
manpage by not rendering it verbatim. v1.69 handles it correctly, but
not many current popular distributions ship with it.

asciidoc by default converts "listingblock" to "<screen>". This change
causes asciidoc in git to convert "listingblock" to "<literallayout>", which
both old and new docbook-xsl handle correctly.

The difference can be seen in any manpage which includes a multi-line
example, such as git-branch.

[jc: the original patch was an disaster for html backends, so I made
 it apply only to docbook backends. ]

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoannotate-blame test: add evil merge.
Junio C Hamano [Mon, 6 Mar 2006 06:37:15 +0000 (22:37 -0800)]
annotate-blame test: add evil merge.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoannotate-blame test: don't "source", but say "."
Junio C Hamano [Mon, 6 Mar 2006 06:10:26 +0000 (22:10 -0800)]
annotate-blame test: don't "source", but say "."

Just I am old fashioned.  Source inclusion in bourne shell is
"." (dot), not "source" -- that's csh.

[jc: yes I know bash groks it, but I am old fashioned.]

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoannotate/blame tests updates.
Junio C Hamano [Mon, 6 Mar 2006 06:07:37 +0000 (22:07 -0800)]
annotate/blame tests updates.

This rewrites the result check code a bit.  The earlier one
using awk was splitting columns at any whitespace, which
confused lines attributed incorrectly to the merge made by the
default author "A U Thor <author@example.com>" with lines
attributed to author "A".

The latest test by Ryan to add the "starting from older commit"
test is also included, with another older commit test.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoannotate: Support annotation of files on other revisions.
Ryan Anderson [Mon, 6 Mar 2006 02:43:53 +0000 (21:43 -0500)]
annotate: Support annotation of files on other revisions.

This is a bug fix, and cleans up one or two other things spotted during the
course of tracking down the main bug here.

[jc: the part that updates test-suite is split out to the next
 one. Also I dropped "use Data::Dumper;" which seemed leftover
 from debugging session.]

Signed-off-by: Ryan Anderson <ryan@michonline.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit/Documentation: fix SYNOPSIS style bugs
Dmitry V. Levin [Mon, 6 Mar 2006 00:00:48 +0000 (03:00 +0300)]
git/Documentation: fix SYNOPSIS style bugs

This trivial patch fixes SYNOPSIS style bugs.

Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoblame: avoid "diff -u0".
Junio C Hamano [Sun, 5 Mar 2006 22:43:56 +0000 (14:43 -0800)]
blame: avoid "diff -u0".

As Linus suggests, use "diff -u -U 0" instead.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-blame: Use the same tests for git-blame as for git-annotate
Fredrik Kuivinen [Sun, 5 Mar 2006 11:13:34 +0000 (12:13 +0100)]
git-blame: Use the same tests for git-blame as for git-annotate

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoblame and annotate: show localtime with timezone.
Junio C Hamano [Sun, 5 Mar 2006 22:48:01 +0000 (14:48 -0800)]
blame and annotate: show localtime with timezone.

Earlier they showed gmtime and timezone, which was inconsistent
with the way our commits and tags are pretty-printed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoblame: avoid -lm by not using log().
Junio C Hamano [Sun, 5 Mar 2006 22:46:36 +0000 (14:46 -0800)]
blame: avoid -lm by not using log().

... as suggested on the list.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-blame: Make the output human readable
Fredrik Kuivinen [Sun, 5 Mar 2006 11:03:51 +0000 (12:03 +0100)]
git-blame: Make the output human readable

The default output mode is slightly different from git-annotate's.
However, git-annotate's output mode can be obtained by using the
'-c' flag.

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoget_revision(): do not dig deeper when we know we are at the end.
Linus Torvalds [Sun, 5 Mar 2006 17:53:52 +0000 (09:53 -0800)]
get_revision(): do not dig deeper when we know we are at the end.

This resurrects the special casing for "rev-list -n 1" which
avoided reading parents unnecessarily.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodocumentation: add 'see also' sections to git-rm and git-add
Jeff Muizelaar [Sun, 5 Mar 2006 21:18:19 +0000 (16:18 -0500)]
documentation: add 'see also' sections to git-rm and git-add

Pair up git-add and git-rm by adding a 'see also' section that
references the opposite command to each of their documentation files.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/emacs/Makefile: Provide tool for byte-compiling files.
Mark Wooding [Sun, 5 Mar 2006 16:14:31 +0000 (16:14 +0000)]
contrib/emacs/Makefile: Provide tool for byte-compiling files.

Signed-off-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogitignore: Ignore some more boring things.
Mark Wooding [Sun, 5 Mar 2006 12:25:43 +0000 (12:25 +0000)]
gitignore: Ignore some more boring things.

Signed-off-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoConst tightening.
Junio C Hamano [Sun, 5 Mar 2006 10:47:29 +0000 (02:47 -0800)]
Const tightening.

Mark Wooding noticed there was a type mismatch warning in git.c; this
patch does things slightly differently (mostly tightening const) and
was what I was holding onto, waiting for the setup-revisions change
to be merged into the master branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation/Makefile: Some `git-*.txt' files aren't manpages.
Mark Wooding [Sun, 5 Mar 2006 09:36:33 +0000 (09:36 +0000)]
Documentation/Makefile: Some `git-*.txt' files aren't manpages.

In particular, git-tools.txt isn't a manpage, and my Asciidoc gets upset
by it.  The simplest fix is to Remove articles from the list of manpages
the Makefile.

Signed-off-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: updated documentation
Martin Langhoff [Sun, 5 Mar 2006 10:11:00 +0000 (23:11 +1300)]
cvsserver: updated documentation

... and stripped trailing whitespace to appease the Gods...

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd --temp and --stage=all options to checkout-index.
Shawn Pearce [Sun, 5 Mar 2006 08:24:15 +0000 (03:24 -0500)]
Add --temp and --stage=all options to checkout-index.

Sometimes it is convient for a Porcelain to be able to checkout all
unmerged files in all stages so that an external merge tool can be
executed by the Porcelain or the end-user.  Using git-unpack-file
on each stage individually incurs a rather high penalty due to the
need to fork for each file version obtained.  git-checkout-index -a
--stage=all will now do the same thing, but faster.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocosmetics: change from 'See-Also' to 'See Also'
Jeff Muizelaar [Sun, 5 Mar 2006 06:20:07 +0000 (01:20 -0500)]
cosmetics: change from 'See-Also' to 'See Also'

Changes the documentation that uses 'See-Also' to the more common
'See Also' form.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit --amend: allow empty commit.
Junio C Hamano [Sun, 5 Mar 2006 04:36:28 +0000 (20:36 -0800)]
git-commit --amend: allow empty commit.

When amending a commit only to update the commit log message, git-status
would rightly say "Nothing to commit."  Do not let this prevent commit to
be made.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoCauterize dropped or duplicate bits from next.
Junio C Hamano [Sun, 5 Mar 2006 00:18:43 +0000 (16:18 -0800)]
Cauterize dropped or duplicate bits from next.

I am very sorry to do this, but without this funky octopus, "git
log --no-merges master..next" will show commits already merged
into "master" forever.

There are some commits on the next branch (which is never to be
rewound) that are reverts of other commits on the next branch.
They are to revert the finer grained delta experiments that
turned out to have undesirable performance effects.  Also there
are some other commits that were first done as a merge into
"next" (a pull request based on next) and then cherry picked
into master.  Since they are not going to be merged into
"master" ever, they will stay forever in "log master..next".

Yuck.

So this commit records the fact that the commits currently shown
by "git log --no-merges master..next" to be merged into "master"
are already in the master, either because they really are (in
the case of git-cvsserver bits, which needed cherry-picking into
"master"), or because they are fully reverted in "next" (in the
case of finer-grained delta bits).

Here is the way I made this commit:

 (1) Inspect "gitk --no-merges --parents master..next"

     This shows what git thinks are missing from master.  It
     shows chain of commits that are already merged and chain of
     commits whose net effect should amount to a no-op.

     Look at each commits and make sure they are either unwanted
     or already merged by cherry-picking.

 (2) Record the tip of branches that I do not want.  In this
     case, the following were unwanted:

     cfcbd3427e67056a00ec832645b057eaf33888d9 cvsserver
     c436eb8cf1efa3fe2c70100ae0cbc48f0feaf5af diff-delta
     38fd0721d0a2a1a723bc28fc0817e3571987b1ef diff-delta
     f0bcd511ee3a00b7fd3975a386aa1165c07a0721 cvsserver
     2b8d9347aa1a11f1ac13591f89ca9f984d467c77 diff-delta

 (3) Shorten the list by finding independent ones from the
     above.

     $ git show-branch --independent $the $above $tips

     cfcbd3427e67056a00ec832645b057eaf33888d9
     c436eb8cf1efa3fe2c70100ae0cbc48f0feaf5af

 (4) Checkout "master" and cauterize them with "ours" strategy:

     $ git merge -s ours "`cat $this-file`" HEAD cfcbd3 c436eb

18 years agoMerge part of 'sp/checkout'
Junio C Hamano [Sat, 4 Mar 2006 22:58:11 +0000 (14:58 -0800)]
Merge part of 'sp/checkout'

18 years agoAsciiDoc fix for tutorial
Francis Daly [Sat, 4 Mar 2006 16:35:27 +0000 (16:35 +0000)]
AsciiDoc fix for tutorial

RE \^.+\^ becomes <sup>. Not wanted here

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit.el: Added customize support for all parameters.
Alexandre Julliard [Sat, 4 Mar 2006 16:38:58 +0000 (17:38 +0100)]
git.el: Added customize support for all parameters.

Also fixed quoting of git-log-msg-separator.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit.el: Added support for Signed-off-by.
Alexandre Julliard [Sat, 4 Mar 2006 16:38:41 +0000 (17:38 +0100)]
git.el: Added support for Signed-off-by.

If `git-append-signed-off-by' is non-nil, automatically append a
sign-off line to the log message when editing it.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit.el: Automatically update .gitignore status.
Alexandre Julliard [Sat, 4 Mar 2006 16:38:20 +0000 (17:38 +0100)]
git.el: Automatically update .gitignore status.

Update .gitignore files in the status list as they are created or
modified.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit.el: Set default directory before running the status mode setup hooks.
Alexandre Julliard [Sat, 4 Mar 2006 16:38:05 +0000 (17:38 +0100)]
git.el: Set default directory before running the status mode setup hooks.

Also set the list-buffers-directory variable for nicer buffer list
display.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit.el: Portability fixes for XEmacs and Emacs CVS.
Alexandre Julliard [Sat, 4 Mar 2006 16:37:42 +0000 (17:37 +0100)]
git.el: Portability fixes for XEmacs and Emacs CVS.

Fixed octal constants for XEmacs.
Added highlighting support in log-edit buffer for Emacs CVS.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/emacs: Add an Emacs VC backend.
Alexandre Julliard [Sat, 4 Mar 2006 14:32:19 +0000 (15:32 +0100)]
contrib/emacs: Add an Emacs VC backend.

Add a basic Emacs VC backend. It currently supports the following
commands: checkin, checkout, diff, log, revert, and annotate. There is
only limited support for working with revisions other than HEAD.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'fk/blame'
Junio C Hamano [Sat, 4 Mar 2006 21:22:01 +0000 (13:22 -0800)]
Merge branch 'fk/blame'

* fk/blame:
  git-blame, take 2
  Merge part of 'lt/rev-list' into 'fk/blame'
  Add git-blame, a tool for assigning blame.

18 years agoMerge branch 'lt/rev-list'
Junio C Hamano [Sat, 4 Mar 2006 21:21:17 +0000 (13:21 -0800)]
Merge branch 'lt/rev-list'

* lt/rev-list:
  setup_revisions(): handle -n<n> and -<n> internally.
  git-log (internal): more options.
  git-log (internal): add approxidate.
  Rip out merge-order and make "git log <paths>..." work again.
  Tie it all together: "git log"
  Introduce trivial new pager.c helper infrastructure
  git-rev-list libification: rev-list walking
  Splitting rev-list into revisions lib, end of beginning.
  rev-list split: minimum fixup.
  First cut at libifying revlist generation

18 years agoAdd a Documentation/git-tools.txt
Marco Costalba [Sat, 4 Mar 2006 06:42:09 +0000 (07:42 +0100)]
Add a Documentation/git-tools.txt

A brief survey of useful git tools, including third-party
and external projects.

Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: anonymous cvs via pserver support
Martin Langhoff [Sat, 4 Mar 2006 07:30:04 +0000 (20:30 +1300)]
cvsserver: anonymous cvs via pserver support

git-cvsserver now knows how to do the pserver auth chat when the user
is anonymous. To get it to work, add a line to your inetd.conf like

  cvspserver stream tcp nowait nobody git-cvsserver pserver

(On some inetd implementations you may have to put the pserver parameter twice.)

Commits are blocked. Naively, git-cvsserver assumes non-malicious users. Please
review the code before setting this up on an internet-accessible server.

NOTE: the <nobody> user above will need write access to the .git directory
to maintain the sqlite database. Updating of the sqlite database should be
put in an update hook to avoid this problem, so that it is maintained by
users with write access.

18 years agoMerge branch 'tl/anno'
Junio C Hamano [Sat, 4 Mar 2006 06:29:17 +0000 (22:29 -0800)]
Merge branch 'tl/anno'

* tl/anno:
  annotate should number lines starting with 1

18 years agocvsserver: better error messages
Martin Langhoff [Sat, 4 Mar 2006 04:47:22 +0000 (17:47 +1300)]
cvsserver: better error messages

We now have different error messages when the repo is not found vs repo is
not configured to allow gitcvs. Should help users during initial checkouts.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: nested directory creation fixups for Eclipse clients
Martin Langhoff [Sat, 4 Mar 2006 04:47:29 +0000 (17:47 +1300)]
cvsserver: nested directory creation fixups for Eclipse clients

To create nested directories without (or before) sending file entries
is rather tricky. Most clients just work. Eclipse, however, expects
a very specific sequence of events. With this patch, cvsserver meets
those expectations.

Note: we may want to reuse prepdir() in req_update -- should move it
outside of req_co. Right now prepdir() is tied to how req_co() works.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'maint'
Junio C Hamano [Sat, 4 Mar 2006 05:36:52 +0000 (21:36 -0800)]
Merge branch 'maint'

* maint:
  tar-tree: file/dirmode fix.

18 years agotar-tree: file/dirmode fix.
Junio C Hamano [Sat, 4 Mar 2006 05:34:14 +0000 (21:34 -0800)]
tar-tree: file/dirmode fix.

This fixes two bugs introduced when we switched to generic tree
traversal code.

 (1) directory mode recorded silently became 0755, not 0777

 (2) if passed a tree object (not a commit), it emitted an
     alarming error message (but proceeded anyway).

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoannotate should number lines starting with 1
Luck, Tony [Thu, 2 Mar 2006 23:27:31 +0000 (15:27 -0800)]
annotate should number lines starting with 1

C programmers are well used to counting from zero, but every
other text file tool starts counting from 1.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: fix a copied-tree bug in an overzealous assertion
Eric Wong [Fri, 3 Mar 2006 21:35:48 +0000 (13:35 -0800)]
contrib/git-svn: fix a copied-tree bug in an overzealous assertion

I thought passing --stop-on-copy to svn would save us from all
the trouble svn-arch-mirror had with directory (project) copies.
I was wrong, there was one thing I overlooked.

If a tree was moved from /foo/trunk to /bar/foo/trunk with no
other changes in r10, but the last change was done in r5, the
Last Changed Rev (from svn info) in /bar/foo/trunk will still be
r5, even though the copy in the repository didn't exist until
r10.

Now, if we ever detect that the Last Changed Rev isn't what
we're expecting, we'll run svn diff and only croak if there are
differences between them.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoshow-branch --topics: omit more uninteresting commits.
Junio C Hamano [Fri, 3 Mar 2006 22:34:40 +0000 (14:34 -0800)]
show-branch --topics: omit more uninteresting commits.

When inspecting contents of topic branches for yet-to-be-merged
commits, a commit that is in the release/master branch is
uninteresting.  Previous round still showed them, especially,
the ones before a topic branch that was forked from the
release/master later than other topic branches.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoworkaround fat/ntfs deficiencies for t3600-rm.sh (git-rm)
Alex Riesen [Fri, 3 Mar 2006 10:20:18 +0000 (11:20 +0100)]
workaround fat/ntfs deficiencies for t3600-rm.sh (git-rm)

Signed-off-by: Alex Riesen <ariesen@harmanbecker.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-mv: fix moves into a subdir from outside
Josef Weidendorfer [Fri, 3 Mar 2006 16:23:32 +0000 (17:23 +0100)]
git-mv: fix moves into a subdir from outside

git-mv needs to be run from the base directory so that
the check if a file is under revision also covers files
outside of a subdirectory. Previously, e.g. in the git repo,

  cd Documentation; git-mv ../README .

produced the error

  Error: '../README' not under version control

The test is extended for this case; it previously only tested
one direction.

Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosend-email: accept --no-signed-off-by-cc as the documentation states
Eric Wong [Fri, 3 Mar 2006 09:28:48 +0000 (01:28 -0800)]
send-email: accept --no-signed-off-by-cc as the documentation states

--no-signed-off-cc is still supported, for backwards compatibility

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: better documenting of CLI switches
Eric Wong [Fri, 3 Mar 2006 09:20:09 +0000 (01:20 -0800)]
contrib/git-svn: better documenting of CLI switches

Also, fix a asciidoc formatting error

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: add --id/-i=$GIT_SVN_ID command-line switch
Eric Wong [Fri, 3 Mar 2006 09:20:09 +0000 (01:20 -0800)]
contrib/git-svn: add --id/-i=$GIT_SVN_ID command-line switch

I ended up using GIT_SVN_ID far more than I ever thought I
would.  Typing less is good.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: avoid re-reading the repository uuid, it never changes
Eric Wong [Fri, 3 Mar 2006 09:20:09 +0000 (01:20 -0800)]
contrib/git-svn: avoid re-reading the repository uuid, it never changes

If it does change, we're screwed anyways as SVN will refuse to
commit or update.  We also never access more than one SVN
repository per-invocation, so we can store it as a global, too.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: create a more recent master if one does not exist
Eric Wong [Fri, 3 Mar 2006 09:20:09 +0000 (01:20 -0800)]
contrib/git-svn: create a more recent master if one does not exist

In a new repository, the initial fetch creates a master branch
if one does not exist so HEAD has something to point to.

It now creates a master at the end of the initial fetch run,
pointing to the latest revision.  Previously it pointed to the
first revision imported, which is generally less useful.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: cleanup option parsing
Eric Wong [Fri, 3 Mar 2006 09:20:08 +0000 (01:20 -0800)]
contrib/git-svn: cleanup option parsing

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: allow --authors-file to be specified
Eric Wong [Fri, 3 Mar 2006 09:20:08 +0000 (01:20 -0800)]
contrib/git-svn: allow --authors-file to be specified

Syntax is compatible with git-svnimport and git-cvsimport:

normalperson = Eric Wong <normalperson@yhbt.net>

If this option is specified and git-svn encounters an SVN
committer name that it cannot parse, it git-svn will abort.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: strip 'git-svn-id:' when commiting to SVN
Eric Wong [Fri, 3 Mar 2006 09:20:08 +0000 (01:20 -0800)]
contrib/git-svn: strip 'git-svn-id:' when commiting to SVN

We regenerate and use git-svn-id: whenever we fetch or otherwise
commit to remotes/git-svn.  We don't actually know what revision
number we'll commit to SVN at commit time, so this is useless.
It won't throw off things like 'rebuild', though, which knows to
only use the last instance of git-svn-id: in a log message

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: several small bug fixes and changes
Eric Wong [Fri, 3 Mar 2006 09:20:07 +0000 (01:20 -0800)]
contrib/git-svn: several small bug fixes and changes

 * Fixed manually-edited commit messages not going to
   remotes/git-svn on sequential commits after the sequential
   commit optimization.
 * format help correctly after adding 'show-ignore'
 * sha1_short regexp matches down to 4 hex characters
   (from git-rev-parse --short documentation)
 * Print the first line of the commit message when we commit to
   SVN next to the sha1.
 * Document 'T' (type change) in the comments

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: add -b/--branch switch for branch detection
Eric Wong [Fri, 3 Mar 2006 09:20:07 +0000 (01:20 -0800)]
contrib/git-svn: add -b/--branch switch for branch detection

I've said I don't like branches in Subversion, and I still don't.
This is a bit more flexible, though, as the argument for -b is any
arbitrary git head/tag reference.

This makes some things easier:
 * Importing git history into a brand new SVN branch.
 * Tracking multiple SVN branches via GIT_SVN_ID, even from multiple
   repositories.
 * Adding tags from SVN (still need to use GIT_SVN_ID, though).
 * Even merge tracking is supported, if and only the heads end up with
   100% equivalent tree objects.  This is more stricter but more robust
   and foolproof than parsing commit messages, imho.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoPrevent --index-info from ignoring -z.
Shawn Pearce [Thu, 2 Mar 2006 17:21:33 +0000 (12:21 -0500)]
Prevent --index-info from ignoring -z.

If git-update-index --index-info -z is used only the first
record given to the process will actually be updated as
the -z option is ignored until after all index records
have been read and processed.  This meant that multiple
null terminated records were seen as a single record which
was lacking a trailing LF, however since the first record
ended in a null the C string handling functions ignored the
trailing garbage.  So --index-info should be required to be
the last command line option, much as --stdin is required
to be the last command line option.  Because --index-info
implies --stdin this isn't an issue as the user shouldn't
be passing --stdin when also passing --index-info.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agomanpages: insert two missing [verse] markers for multi-line SYNOPSIS
Jonas Fonseca [Thu, 2 Mar 2006 16:10:49 +0000 (17:10 +0100)]
manpages: insert two missing [verse] markers for multi-line SYNOPSIS

Found with:

for i in *.txt; do
grep -A 2 "SYNOPSIS" "$i" | grep -q "^\[verse\]$" && continue
multiline=$(grep -A 3 "SYNOPSIS" "$i" | tail -n 1)
test -n "$multiline" && echo "$i: $multiline"
done

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogitview: pass the missing argument _show_clicked_cb.
Aneesh Kumar K.V [Thu, 2 Mar 2006 11:47:39 +0000 (17:17 +0530)]
gitview: pass the missing argument _show_clicked_cb.

In our last update to use the encoding while showing the commit
diff we added a new argument to this function. But we missed
updating all the callers.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix test case for some sed
Johannes Schindelin [Thu, 2 Mar 2006 11:43:51 +0000 (12:43 +0100)]
Fix test case for some sed

Some versions of sed lack the "-i" option.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-branch: add -r switch to list refs/remotes/*
Eric Wong [Thu, 2 Mar 2006 20:23:17 +0000 (12:23 -0800)]
git-branch: add -r switch to list refs/remotes/*

If we decide to use refs/remotes/, having a convenient way to
list them would be nice.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: fix checkouts with -d <somedir>
Martin Langhoff [Fri, 3 Mar 2006 03:57:03 +0000 (16:57 +1300)]
cvsserver: fix checkouts with -d <somedir>

A recent Eclipse compat fix broke checkouts with -d. Fix it so that the server
sends the correct module name instead of the destination directory name.

18 years agocvsserver: checkout faster by sending files in a sensible order
Martin Langhoff [Fri, 3 Mar 2006 03:38:03 +0000 (16:38 +1300)]
cvsserver: checkout faster by sending files in a sensible order

Just by sending the files in an ordered fashion, clients can process them
much faster. And we can optimize our check of whether we created this
directory already -- faster.

Timings for a checkout on a commandline cvs client for a project with
~13K files totalling ~100MB:

Unsorted:
  603.12 real        16.89 user        42.88 sys

Sorted:
  298.19 real        26.37 user        42.42 sys

18 years agogit-commit: make sure we protect against races.
Junio C Hamano [Fri, 3 Mar 2006 05:13:24 +0000 (21:13 -0800)]
git-commit: make sure we protect against races.

An earlier commit 8098a178b26dc7a158d129a092a5b78da6d12b72
accidentally lost race protection from git-commit command.
This commit reinstates it.  When something else updates HEAD
pointer while you were editing your commit message, the command
would notice and abort the commit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit --amend
Junio C Hamano [Fri, 3 Mar 2006 05:04:05 +0000 (21:04 -0800)]
git-commit --amend

The new flag is used to amend the tip of the current branch.  Prepare
the tree object you would want to replace the latest commit as usual
(this includes the usual -i/-o and explicit paths), and the commit log
editor is seeded with the commit message from the tip of the current
branch.  The commit you create replaces the current tip -- if it was a
merge, it will have the parents of the current tip as parents -- so the
current top commit is discarded.

It is a rough equivalent for:

$ git reset --soft HEAD^
$ ... do something else to come up with the right tree ...
$ git commit -c ORIG_HEAD

but can be used to amend a merge commit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: fix checkouts with -d <somedir>
Martin Langhoff [Fri, 3 Mar 2006 03:57:03 +0000 (16:57 +1300)]
cvsserver: fix checkouts with -d <somedir>

A recent Eclipse compat fix broke checkouts with -d. Fix it so that the server
sends the correct module name instead of the destination directory name.

18 years agocvsserver: checkout faster by sending files in a sensible order
Martin Langhoff [Fri, 3 Mar 2006 03:38:03 +0000 (16:38 +1300)]
cvsserver: checkout faster by sending files in a sensible order

Just by sending the files in an ordered fashion, clients can process them
much faster. And we can optimize our check of whether we created this
directory already -- faster.

Timings for a checkout on a commandline cvs client for a project with
~13K files totalling ~100MB:

Unsorted:
  603.12 real        16.89 user        42.88 sys

Sorted:
  298.19 real        26.37 user        42.42 sys

18 years agoshow-branch --topics
Junio C Hamano [Fri, 3 Mar 2006 01:14:00 +0000 (17:14 -0800)]
show-branch --topics

This adds a new flag, --topics, to help managing topic
branches.  When you have topic branches forked some time ago
from your primary line of development, show-branch would show
many "uninteresting" things that happend on the primary line of
development when trying to see what are still not merged from
the topic branches.

With this flag, the first ref given to show-branch is taken as
the primary branch, and the rest are taken as the topic
branches.  Output from the command is modified so that commits
only on the primary branch are not shown.  In other words,

$ git show-branch --topics master topic1 topic2 ...

shows an (almost) equivalent of

$ git rev-list ^master topic1 topic2 ...

The major differences are that (1) you can tell which commits
are on which branch, and (2) the commit at the fork point is
shown.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoGIT-VERSION-GEN: squelch unneeded error from "cat version"
Junio C Hamano [Thu, 2 Mar 2006 22:38:44 +0000 (14:38 -0800)]
GIT-VERSION-GEN: squelch unneeded error from "cat version"

Now this is really a corner case, but if you have the git source
tree from somewhere other than the official tarball, you do not
have version file.  And if git-describe does not work for you
(maybe you do not have git yet), we spilled an error message
from "cat version".

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosetup_revisions(): handle -n<n> and -<n> internally.
Junio C Hamano [Thu, 2 Mar 2006 23:24:01 +0000 (15:24 -0800)]
setup_revisions(): handle -n<n> and -<n> internally.

This moves the handling of max-count shorthand from the internal
implementation of "git log" to setup_revisions() so other users
of setup_revisions() can use it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-blame, take 2
Fredrik Kuivinen [Thu, 2 Mar 2006 21:28:17 +0000 (22:28 +0100)]
git-blame, take 2

Here is an updated version of git-blame. The main changes compared to
the first version are:

* Use the new revision.h interface to do the revision walking
* Do the right thing in a lot of more cases than before. In particular
  parallel development tracks are hopefully handled sanely.
* Lots of clean-up

It still won't follow file renames though.

There are still some differences in the output between git-blame and
git-annotate. For example, in 'Makefile' git-blame assigns lines
354-358 to 455a7f3275d264f6e66045b92c83747ec461dda5 and git-annotate
assigns the same lines to 79a9d8ea0d88a3667ad19be8e705405ab5d896f1.

I think git-blame is correct in this case. This patterns occur in
several other places, git-annotate seems to sometimes assign lines to
merge commits when the lines actually changed in some other commit
which precedes the merge.

[jc: I have conned Ryan into doing test cases, so that it would
help development and fixes on both implementations.  Let the
battle begin! ;-) ]

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>