Code

git.git
17 years agogit object hash cleanups v1.4.1
Linus Torvalds [Fri, 30 Jun 2006 18:20:33 +0000 (11:20 -0700)]
git object hash cleanups

This IMNSHO cleans up the object hashing.

The hash expansion is separated out into a function of its own, the hash
array (and size) names are made more obvious, and the code is generally
made to look a bit more like the object-ref hashing.

It also gets rid of "find_object()" returning an index (or negative
position if no object is found), since that is made redundant by the
simplified object rehashing. The basic operation is now "lookup_object()"
which just returns the object itself.

There's an almost unmeasurable speed increase, but more importantly, I
think the end result is more readable.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorevision.c: --full-history fix.
Linus Torvalds [Sat, 1 Jul 2006 03:21:59 +0000 (20:21 -0700)]
revision.c: --full-history fix.

With history simplification, we still show merges that are required
to make the history _complete_, i.e. say that you had:

  a
  |
  b
 / \
c   d
|   |

and neither "a" nor "b" actually changed the file, but both "c" and "d"
did: in this case we have to leave "b" around just because otherwise there
would be no way to show the _relationship_, even if "b" itself doesn't
actually change the tree in any way what-so-ever.

It would make sense to make that further simplification if the
"--parents" flag wasn't present.  In that case the user is
literally asking for a list of commits and is not interested in
the relationship between them.

This patch also fixes a real bug.  Without this patch, the
"--parents --full-history" combination (which you'd get if you
do something like

gitk --full-history Makefile

or similar) will actually _drop_ merges where all children are identical.
That's wrong in the --full-history case, because it means that the graph
ends up missing lots of entries.

In the process, this also should make

git-rev-list --full-history Makefile

give just the _true_ list of all commits that changed Makefile (and
properly ignore merges that were identical in one parent), because now
we're not asking for "--parent", so we don't need the unnecessary merge
commits to keep the history together.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix errno usage in connect.c
Petr Baudis [Sat, 1 Jul 2006 21:56:26 +0000 (23:56 +0200)]
Fix errno usage in connect.c

errno was used after it could've been modified by a subsequent library call.
Spotted by Morten Welinder.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMinor documentation fixup.
Robin Rosenberg [Sun, 2 Jul 2006 00:07:40 +0000 (02:07 +0200)]
Minor documentation fixup.

Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: allow a local target directory to be specified for init
Eric Wong [Sat, 1 Jul 2006 04:42:53 +0000 (21:42 -0700)]
git-svn: allow a local target directory to be specified for init

git-svn init url://to/the/repo local-repo

will create the local-repo dirrectory if doesn't exist yet and
populate it as expected.

Original patch by Luca Barbato, cleaned up and made to work for
the current version of git-svn by me (Eric Wong).

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agodon't load objects needlessly when repacking
Nicolas Pitre [Sat, 1 Jul 2006 02:55:30 +0000 (22:55 -0400)]
don't load objects needlessly when repacking

If no delta is attempted on some objects then it is useless to load them
in memory, neither create any delta index for them.  The best thing to
do is therefore to load and index them only when really needed.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoupload-pack.c: <sys/poll.h> includes <ctype.h> on OpenBSD 3.8
Junio C Hamano [Fri, 30 Jun 2006 18:30:29 +0000 (11:30 -0700)]
upload-pack.c: <sys/poll.h> includes <ctype.h> on OpenBSD 3.8

Merlyn reports that <sys/poll.h> on OpenBSD 3.8 includes <ctype.h>
and having our custom ctype (done in git-compat-util.h which is
included via cache.h) makes upload-pack.c uncompilable.  Try to
work it around by including the system headers first.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAbstract out accesses to object hash array
Linus Torvalds [Fri, 30 Jun 2006 04:38:55 +0000 (21:38 -0700)]
Abstract out accesses to object hash array

There are a few special places where some programs accessed the object
hash array directly, which bothered me because I wanted to play with some
simple re-organizations.

So this patch makes the object hash array data structures all entirely
local to object.c, and the few users who wanted to look at it now get to
use a function to query how many object index entries there can be, and to
actually access the array.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoconsider previous pack undeltified object state only when reusing delta data
Nicolas Pitre [Fri, 30 Jun 2006 03:44:52 +0000 (23:44 -0400)]
consider previous pack undeltified object state only when reusing delta data

Without this there would never be a chance to improve packing for
previously undeltified objects.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/test-3402'
Junio C Hamano [Fri, 30 Jun 2006 06:47:59 +0000 (23:47 -0700)]
Merge branch 'jc/test-3402'

* jc/test-3402:
  Racy GIT (part #3)

17 years agoDo not try futile object pairs when repacking.
Linus Torvalds [Thu, 29 Jun 2006 21:04:01 +0000 (14:04 -0700)]
Do not try futile object pairs when repacking.

In the repacking window, if both objects we are looking at already came
from the same (old) pack-file, don't bother delta'ing them against each
other.

That means that we'll still always check for better deltas for (and
against!) _unpacked_ objects, but assuming incremental repacks, you'll
avoid the delta creation 99% of the time.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRacy GIT (part #3)
Junio C Hamano [Thu, 29 Jun 2006 21:48:22 +0000 (14:48 -0700)]
Racy GIT (part #3)

Commit 29e4d3635709778bcc808dbad0477efad82f8d7e fixed the
underlying update-index races but git-commit was not careful
enough to preserve the index file timestamp when copying the
index file.  This caused t3402 test to occasionally fail.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/repack'
Junio C Hamano [Thu, 29 Jun 2006 06:43:48 +0000 (23:43 -0700)]
Merge branch 'jc/repack'

* jc/repack:
  git-repack: Be careful when updating the same pack as an existing one.

17 years agoMerge branch 'js/patch'
Junio C Hamano [Thu, 29 Jun 2006 06:42:40 +0000 (23:42 -0700)]
Merge branch 'js/patch'

* js/patch:
  diff.c: fix get_patch_id()
  t4014: fix test commit labels.
  format-patch: use clear_commit_marks() instead of some ad-hockery
  t4014: fix for whitespace from "wc -l"
  t4014: add format-patch --ignore-if-in-upstream test
  format-patch: introduce "--ignore-if-in-upstream"
  add diff_flush_patch_id() to calculate the patch id

17 years agodiff.c: fix get_patch_id()
Junio C Hamano [Thu, 29 Jun 2006 05:49:42 +0000 (22:49 -0700)]
diff.c: fix get_patch_id()

The function internally generated diff to get the patch id but
passed a wrong emit flags to the xdiff layer when it did so.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot4014: fix test commit labels.
Junio C Hamano [Thu, 29 Jun 2006 05:48:34 +0000 (22:48 -0700)]
t4014: fix test commit labels.

The commit tag and commit comments used in the test claimed that
the #1 commit was merged upstream where the test actually let the
upstream merge #2 commit.  Fix them.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocheckout -m: fix read-tree invocation v1.4.1-rc2
Junio C Hamano [Wed, 28 Jun 2006 18:47:28 +0000 (11:47 -0700)]
checkout -m: fix read-tree invocation

When we updated "read-tree -m -u" to be careful about not
removing untracked working tree files, we broke "checkout -m" to
switch between branches.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot/README: start testing porcelainish
Junio C Hamano [Wed, 28 Jun 2006 18:45:52 +0000 (11:45 -0700)]
t/README: start testing porcelainish

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoSave errno in handle_alias()
Johannes Schindelin [Wed, 28 Jun 2006 10:45:27 +0000 (12:45 +0200)]
Save errno in handle_alias()

git.c:main() relies on the value of errno being set by the last attempt to
execute the command. However, if something goes awry in handle_alias(),
that assumption is wrong. So restore errno before returning from
handle_alias().

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorebase: check for errors from git-commit
Eric Wong [Wed, 28 Jun 2006 10:24:23 +0000 (03:24 -0700)]
rebase: check for errors from git-commit

commit does not always succeed, so we'll have to check for
it in the absence of set -e.  This fixes a regression
introduced in 9e4bc7dd1bb9d92491c475cec55147fa0b3f954d

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocvsimport - cleanup of the multi-indexes handling
Martin Langhoff [Wed, 28 Jun 2006 10:13:23 +0000 (22:13 +1200)]
cvsimport - cleanup of the multi-indexes handling

Indexes are only needed when we are about preparing to commit. Prime them
inside commit() when we have all the info we need, and remove all the
redundant index setups.

While we are at it, make sure that index handling is correct when opening
new branches, and on initial import.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoconnect.c: check the commit buffer boundary while parsing.
Junio C Hamano [Wed, 28 Jun 2006 10:51:00 +0000 (03:51 -0700)]
connect.c: check the commit buffer boundary while parsing.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoconnect.c: remove unused parameters from tcp_connect and proxy_connect
Junio C Hamano [Wed, 28 Jun 2006 10:50:33 +0000 (03:50 -0700)]
connect.c: remove unused parameters from tcp_connect and proxy_connect

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMake some strings const
Timo Hirvonen [Wed, 28 Jun 2006 09:04:39 +0000 (12:04 +0300)]
Make some strings const

Signed-off-by: Timo Hirvonen <tihirvon@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorebase: get rid of outdated MRESOLVEMSG
Eric Wong [Wed, 28 Jun 2006 09:11:06 +0000 (02:11 -0700)]
rebase: get rid of outdated MRESOLVEMSG

There was a time when rebase --skip didn't work when used with
--merge, but that is no more so we don't need that message
anymore.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit wrapper: fix command name in an error message.
Andreas Ericsson [Wed, 28 Jun 2006 09:17:21 +0000 (02:17 -0700)]
git wrapper: fix command name in an error message.

When the command execution by execv_git_cmd() fails with an errno
other than ENOENT, we used an uninitialized variable instead of
the string that holds the command name to report what failed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: be verbose by default on fetch/commit, add -q/--quiet option
Eric Wong [Wed, 28 Jun 2006 02:39:14 +0000 (19:39 -0700)]
git-svn: be verbose by default on fetch/commit, add -q/--quiet option

Slower connections can make git-svn look as if it's doing
nothing for a long time; leaving the user wondering if we're
actually doing anything.  Now we print some file progress just
to assure the user that something is going on while they're
waiting.

Added the -q/--quiet option to users to revert to the old method
if they preferred it.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: add --follow-parent and --no-metadata options to fetch
Eric Wong [Wed, 28 Jun 2006 02:39:13 +0000 (19:39 -0700)]
git-svn: add --follow-parent and --no-metadata options to fetch

--follow-parent:
  This is especially helpful when we're tracking a directory
  that has been moved around within the repository, or if we
  started tracking a branch and never tracked the trunk it was
  descended from.

  This relies on the SVN::* libraries to work.  We can't
  reliably parse path info from the svn command-line client
  without relying on XML, so it's better just to have the SVN::*
  libs installed.

  This also removes oldvalue verification when calling update-ref

  In SVN, branches can be deleted, and then recreated under the
  same path as the original one with different ancestry
  information, causing parent information to be mismatched /
  misordered.

  Also force the current ref, if existing, to be a parent,
  regardless of whether or not it was specified.

--no-metadata:
  This gets rid of the git-svn-id: lines at the end of every commit.

  With this, you lose the ability to use the rebuild command.  If
  you ever lose your .git/svn/git-svn/.rev_db file, you won't be
  able to fetch again, either.  This is fine for one-shot imports.

  Also fix some issues with multi-fetch --follow-parent that were
  exposed while testing this.  Additionally, repack checking is
  simplified greatly.

  git-svn log will not work on repositories using this, either.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: add the commit-diff command
Eric Wong [Wed, 28 Jun 2006 02:39:12 +0000 (19:39 -0700)]
git-svn: add the commit-diff command

This is intended for interoperability with git-svnimport.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: several graft-branches improvements
Eric Wong [Wed, 28 Jun 2006 02:39:11 +0000 (19:39 -0700)]
git-svn: several graft-branches improvements

The 'graft-branches' command can now analyze tree matches for
merge detection after commits are done, when --branch or
--branch-all-refs options are used.

We ensure that tree joins (--branch and --branch-all-refs
options) during commit time only add SVN parents that occurred
before the commit we're importing

Also fixed branch detection via merge messages, this manner of
merge detection (a la git-svnimport) is really all fuzzy, but at
least it actually works now :)

Add some new tests to go along with these fixes, too.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: SVN 1.1.x library compatibility
Eric Wong [Wed, 28 Jun 2006 10:07:14 +0000 (03:07 -0700)]
git-svn: SVN 1.1.x library compatibility

Tested on a plain Ubuntu Hoary installation
using subversion 1.1.1-2ubuntu3

1.1.x issues I had to deal with:

* Avoid the noisy command-line client compatibility check if we
  use the libraries.

* get_log() arguments differ (now using a nice wrapper from
  Junio's suggestion)

* get_file() is picky about what kind of file handles it gets,
  so I ended up redirecting STDOUT.  I'm probably overflushing
  my file handles, but that's the safest thing to do...

* BDB kept segfaulting on me during tests, so svnadmin will use FSFS
  whenever we can.

* If somebody used an expanded CVS $Id$ line inside a file, then
  propsetting it to use svn:keywords will cause the original CVS
  $Id$ to be retained when asked for the original file.  As far as
  I can see, this is a server-side issue.  We won't care in the
  test anymore, as long as it's not expanded by SVN, a static
  CVS $Id$ line is fine.

While we're at making ourselves more compatible, avoid grep
along with the -q flag, which is GNU-specific. (grep avoidance
tip from Junio, too)

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocombine-diff.c: type sanity
Junio C Hamano [Wed, 28 Jun 2006 08:38:19 +0000 (01:38 -0700)]
combine-diff.c: type sanity

- combine_diff() took cnt (count) which is unsigned in nature but the
  parameter type was declared as "int";
- find_next() took "uninteresting" parameter, which masked a static
  function of the same name;
- show_parent_lno() took an unused parameter "cnt";
- show_patch_diff() used a local variable in nested inner scope with
  the same name with different type, masking the one in the outer scope;
- the last loop in show_patch_diff iterated over lines so it should use
  the local variable "lno"

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoquote.c: silence compiler warnings from EMIT macro
Jeff King [Wed, 28 Jun 2006 05:59:23 +0000 (01:59 -0400)]
quote.c: silence compiler warnings from EMIT macro

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoformat-patch: use clear_commit_marks() instead of some ad-hockery
Johannes Schindelin [Tue, 27 Jun 2006 20:38:04 +0000 (22:38 +0200)]
format-patch: use clear_commit_marks() instead of some ad-hockery

It is cleaner, and it describes better what the idea behind the code is.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix expr usage for FreeBSD
Dennis Stosberg [Tue, 27 Jun 2006 16:54:26 +0000 (18:54 +0200)]
Fix expr usage for FreeBSD

Some implementations of "expr" (e.g. FreeBSD's) fail, if an
argument starts with a dash.

Signed-off-by: Dennis Stosberg <dennis@stosberg.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot4014: fix for whitespace from "wc -l"
Johannes Schindelin [Tue, 27 Jun 2006 08:12:12 +0000 (10:12 +0200)]
t4014: fix for whitespace from "wc -l"

Some "wc" insist on putting a TAB in front of the number.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot4014: add format-patch --ignore-if-in-upstream test
Junio C Hamano [Mon, 26 Jun 2006 22:40:09 +0000 (15:40 -0700)]
t4014: add format-patch --ignore-if-in-upstream test

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoformat-patch: introduce "--ignore-if-in-upstream"
Johannes Schindelin [Sun, 25 Jun 2006 01:52:01 +0000 (03:52 +0200)]
format-patch: introduce "--ignore-if-in-upstream"

With this flag, format-patch will try very hard not to output patches which
are already in the upstream branch.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoadd diff_flush_patch_id() to calculate the patch id
Johannes Schindelin [Sun, 25 Jun 2006 01:51:08 +0000 (03:51 +0200)]
add diff_flush_patch_id() to calculate the patch id

Call it like this:

unsigned char id[20];
if (diff_flush_patch_id(diff_options, id))
printf("And the patch id is: %s\n", sha1_to_hex(id));

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/squash'
Junio C Hamano [Mon, 26 Jun 2006 21:36:10 +0000 (14:36 -0700)]
Merge branch 'jc/squash'

* jc/squash:
  git-merge --squash

17 years agoMerge branch 'jc/diff'
Junio C Hamano [Mon, 26 Jun 2006 21:36:02 +0000 (14:36 -0700)]
Merge branch 'jc/diff'

* jc/diff:
  diff --color: use $GIT_DIR/config

17 years agoMerge branch 'ml/cvsimport'
Junio C Hamano [Mon, 26 Jun 2006 21:35:33 +0000 (14:35 -0700)]
Merge branch 'ml/cvsimport'

* ml/cvsimport:
  cvsimport: always set $ENV{GIT_INDEX_FILE} to $index{$branch}
  cvsimport: setup indexes correctly for ancestors and incremental imports

17 years agoMerge branch 'js/diff'
Junio C Hamano [Mon, 26 Jun 2006 21:28:42 +0000 (14:28 -0700)]
Merge branch 'js/diff'

* js/diff:
  Teach diff about -b and -w flags

17 years agoMerge branch 'ew/rebase'
Junio C Hamano [Mon, 26 Jun 2006 21:05:13 +0000 (14:05 -0700)]
Merge branch 'ew/rebase'

* ew/rebase:
  rebase: allow --skip to work with --merge
  rebase: cleanup rebasing with --merge
  rebase: allow --merge option to handle patches merged upstream

17 years agoFix pkt-line.h to compile with a non-GCC compiler
Dennis Stosberg [Mon, 26 Jun 2006 08:27:07 +0000 (10:27 +0200)]
Fix pkt-line.h to compile with a non-GCC compiler

pkt-line.h uses GCC's __attribute__ extension but does not include
git-compat-util.h.  So it will not compile with a compiler that does
not support this extension.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoSolaris needs inclusion of signal.h for signal()
Dennis Stosberg [Mon, 26 Jun 2006 08:26:13 +0000 (10:26 +0200)]
Solaris needs inclusion of signal.h for signal()

Currently the compilation fails in connect.c and merge-index.c

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocorrect documentation for git grep
Matthias Lederhofer [Sun, 25 Jun 2006 16:07:15 +0000 (18:07 +0200)]
correct documentation for git grep

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-repack: Be careful when updating the same pack as an existing one.
Junio C Hamano [Sun, 25 Jun 2006 12:28:58 +0000 (05:28 -0700)]
git-repack: Be careful when updating the same pack as an existing one.

After a clone, packfiles are read-only by default and "mv" to
replace the pack with a new one goes interactive, asking if the
user wants to replace it.  If one is successfully moved and the
other is not, the pack and its idx would become out-of-sync and
corrupts the repository.

Recovering is straightforward -- it is just the matter of
finding the remaining .tmp-pack-* and make sure they are both
moved -- but we should be extra careful not to do something so
alarming to the users.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agodiff --color: use $GIT_DIR/config
Junio C Hamano [Sat, 24 Jun 2006 11:06:23 +0000 (04:06 -0700)]
diff --color: use $GIT_DIR/config

This lets you use something like this in your $GIT_DIR/config
file.

[diff]
color = auto

[diff.color]
new = blue
old = yellow
frag = reverse

When diff.color is set to "auto", colored diff is enabled when
the standard output is the terminal.  Other choices are "always",
and "never".  Usual boolean true/false can also be used.

The colormap entries can specify colors for the following slots:

plain - lines that appear in both old and new file (context)
meta - diff --git header and extended git diff headers
frag - @@ -n,m +l,k @@ lines (hunk header)
old - lines deleted from old file
new - lines added to new file

The following color names can be used:

normal, bold, dim, l, blink, reverse, reset,
black, red, green, yellow, blue, magenta, cyan,
white

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorebase: allow --skip to work with --merge
Eric Wong [Sun, 25 Jun 2006 01:29:49 +0000 (18:29 -0700)]
rebase: allow --skip to work with --merge

Now that we control the merge base selection, we won't be forced
into rolling things in that we wanted to skip beforehand.

Also, add a test to ensure this all works as intended.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorebase: cleanup rebasing with --merge
Eric Wong [Sun, 25 Jun 2006 01:29:48 +0000 (18:29 -0700)]
rebase: cleanup rebasing with --merge

We no longer have to recommit each patch to remove the parent
information we're rebasing since we're using the low-level merge
strategies directly instead of git-merge.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorebase: allow --merge option to handle patches merged upstream
Eric Wong [Sun, 25 Jun 2006 01:29:47 +0000 (18:29 -0700)]
rebase: allow --merge option to handle patches merged upstream

Enhance t3401-rebase-partial to test with --merge as well as
the standard am -3 strategy.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-commit: filter out log message lines only when editor was run.
Yann Dirson [Fri, 23 Jun 2006 22:04:05 +0000 (00:04 +0200)]
git-commit: filter out log message lines only when editor was run.

The current behaviour strips out lines starting with a # even when fed
through stdin or -m.  This is particularly bad when importing history from
another SCM (tailor 0.9.23 uses git-commit).  In the best cases all lines
are stripped and the commit fails with a confusing "empty log message"
error, but in many cases the commit is done, with loss of information.

Note that it is quite peculiar to just have "#" handled as a leading
comment char here.  One commonly meet CVS: or CG: or STG: as prefixes, and
using GIT: would be more robust as well as consistent with other commit
tools.  However, that would break any tool relying on the # (if any).

Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRename safe_strncpy() to strlcpy().
Peter Eriksen [Sat, 24 Jun 2006 14:01:25 +0000 (16:01 +0200)]
Rename safe_strncpy() to strlcpy().

This cleans up the use of safe_strncpy() even more.  Since it has the
same semantics as strlcpy() use this name instead.  Also move the
definition from inside path.c to its own file compat/strlcpy.c, and use
it conditionally at compile time, since some platforms already has
strlcpy().  It's included in the same way as compat/setenv.c.

Signed-off-by: Peter Eriksen <s022018@student.dtu.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoapply: replace NO_ACCURATE_DIFF with --inaccurate-eof runtime flag.
Johannes Schindelin [Sat, 24 Jun 2006 20:10:11 +0000 (22:10 +0200)]
apply: replace NO_ACCURATE_DIFF with --inaccurate-eof runtime flag.

It does not make much sense to build git whose behaviour is
different depending on the brokenness of diff implementation of
the platform because the brokenness of the patch that is applied
with the tool depends on brokenness of the diff the person who
generates the patch uses.  So we do not use NO_ACCURATE_DIFF
anymore, but help people to apply patches that do not record
incomplete lines correctly with a runtime flag.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoClean up diff.c
Timo Hirvonen [Sat, 24 Jun 2006 17:20:32 +0000 (20:20 +0300)]
Clean up diff.c

Signed-off-by: Timo Hirvonen <tihirvon@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocvsimport: always set $ENV{GIT_INDEX_FILE} to $index{$branch}
Johannes Schindelin [Sat, 24 Jun 2006 19:42:20 +0000 (21:42 +0200)]
cvsimport: always set $ENV{GIT_INDEX_FILE} to $index{$branch}

Also, make sure that the initial git-read-tree is performed.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
17 years agocvsimport: setup indexes correctly for ancestors and incremental imports
Martin Langhoff [Sat, 24 Jun 2006 11:13:08 +0000 (23:13 +1200)]
cvsimport: setup indexes correctly for ancestors and incremental imports

Two bugs had slipped in the "keep one index per branch during import"
patch. Both incremental imports and new branches would see an
empty tree for their initial commit. Now we cover all the relevant
cases, checking whether we actually need to setup the index before
preparing the actual commit, and doing it.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorepo-config: fix printing of bool
Junio C Hamano [Sat, 24 Jun 2006 12:19:30 +0000 (05:19 -0700)]
repo-config: fix printing of bool

When a bool variable appears without any value, it means true.
However, replacing the NULL value with an empty string, an earlier
commit f067a13745fbeae1aa357876348a00e5edd0a629 broke show-config.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agodiff --color: use reset sequence when we mean reset.
Junio C Hamano [Sat, 24 Jun 2006 11:20:39 +0000 (04:20 -0700)]
diff --color: use reset sequence when we mean reset.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-repack -- respect -q and be quiet
Martin Langhoff [Sat, 24 Jun 2006 09:41:25 +0000 (21:41 +1200)]
git-repack -- respect -q and be quiet

git-repack was passing the -q along to pack-objects but ignoring it
itself. Correct the oversight.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-merge --squash
Junio C Hamano [Fri, 23 Jun 2006 08:37:02 +0000 (01:37 -0700)]
git-merge --squash

Some people tend to do many little commits on a topic branch,
recording all the trials and errors, and when the topic is
reasonably cooked well, would want to record the net effect of
the series as one commit on top of the mainline, removing the
cruft from the history.  The topic is then abandoned or forked
off again from that point at the mainline.

The barebone porcelainish that comes with core git tools does
not officially support such operation, but you can fake it by
using "git pull --no-merge" when such a topic branch is not a
strict superset of the mainline, like this:

git checkout mainline
git pull --no-commit . that-topic-branch
: fix conflicts if any
rm -f .git/MERGE_HEAD
        git commit -a -m 'consolidated commit log message'
git branch -f that-topic-branch ;# now fully merged

This however does not work when the topic branch is a fast
forward of the mainline, because normal "git pull" will never
create a merge commit in such a case, and there is nothing
special --no-commit could do to begin with.

This patch introduces a new option, --squash, to support such a
workflow officially in both fast-forward case and true merge
case.  The user-level operation would be the same in both cases:

git checkout mainline
        git pull --squash . that-topic-branch
        : fix conflicts if any -- naturally, there would be
        : no conflict if fast forward.
git commit -a -m  'consolidated commit log message'
git branch -f that-topic-branch ;# now fully merged

When the current branch is already up-to-date with respect to
the other branch, there truly is nothing to do, so the new
option does not have any effect.

This was brought up in #git IRC channel recently.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-pull: abort when fmt-merge-msg fails.
Junio C Hamano [Sat, 24 Jun 2006 08:10:27 +0000 (01:10 -0700)]
git-pull: abort when fmt-merge-msg fails.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'pb/error'
Junio C Hamano [Sat, 24 Jun 2006 08:06:51 +0000 (01:06 -0700)]
Merge branch 'pb/error'

* pb/error:
  usage: minimum type fix.
  Customizable error handlers
  git-merge: Don't use -p when outputting summary
  git-commit: allow -e option anywhere on command line
  patch-id: take "commit" prefix as well as "diff-tree" prefix

17 years agousage: minimum type fix.
Junio C Hamano [Sat, 24 Jun 2006 05:44:33 +0000 (22:44 -0700)]
usage: minimum type fix.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoCustomizable error handlers
Petr Baudis [Sat, 24 Jun 2006 02:34:38 +0000 (04:34 +0200)]
Customizable error handlers

This patch makes the usage(), die() and error() handlers customizable.
Nothing in the git code itself uses that but many other libgit users
(like Git.pm) will.

This is implemented using the mutator functions primarily because you
cannot directly modifying global variables of libgit from a program that
dlopen()ed it, apparently. But having functions for that is a better API
anyway.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-merge: Don't use -p when outputting summary
Timo Hirvonen [Fri, 23 Jun 2006 21:45:40 +0000 (00:45 +0300)]
git-merge: Don't use -p when outputting summary

-p is not needed and we only want diffstat and summary.

Signed-off-by: Timo Hirvonen <tihirvon@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTeach diff about -b and -w flags
Johannes Schindelin [Wed, 14 Jun 2006 15:40:23 +0000 (17:40 +0200)]
Teach diff about -b and -w flags

This adds -b (--ignore-space-change) and -w (--ignore-all-space) flags to
diff. The main part of the patch is teaching libxdiff about it.

[jc: renamed xdl_line_match() to xdl_recmatch() since the former is used
 for different purposes in xpatchi.c which is in the parts of the upstream
 source we do not use.]

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-commit: allow -e option anywhere on command line
Jeff King [Fri, 23 Jun 2006 13:43:38 +0000 (09:43 -0400)]
git-commit: allow -e option anywhere on command line

Previously, the command 'git-commit -e -m foo' would ignore the '-e' option
because the '-m' option overwrites the no_edit flag during sequential
option parsing. Now we cause -e to reset the no_edit flag after all
options are parsed.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agopatch-id: take "commit" prefix as well as "diff-tree" prefix
Johannes Schindelin [Fri, 23 Jun 2006 15:36:21 +0000 (17:36 +0200)]
patch-id: take "commit" prefix as well as "diff-tree" prefix

Some time ago we changed git-log in a massive way, and one consequence is
that the keyword changed. Adjust patch-id for that.

[jc: as Linus suggests, allowing both old and new prefix.]

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMakefile: do not recompile main programs when libraries have changed.
Junio C Hamano [Thu, 22 Jun 2006 22:43:47 +0000 (15:43 -0700)]
Makefile: do not recompile main programs when libraries have changed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoadd GIT-CFLAGS to .gitignore
Matthias Kestenholz [Thu, 22 Jun 2006 21:06:39 +0000 (23:06 +0200)]
add GIT-CFLAGS to .gitignore

Signed-off-by: Matthias Kestenholz <matthias@spinlock.ch>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTweak diff colors
Linus Torvalds [Thu, 22 Jun 2006 20:53:31 +0000 (13:53 -0700)]
Tweak diff colors

This patch does:

 - always reset the color _before_ printing out the newline.

   This is actually important. You (and Johannes) didn't see it, because
   it only matters if you set the background, but if you don't do this,
   you get some random and funky behaviour if you pick a color with a
   non-default background (which still potentially has problems with tabs
   etc, but less so).

 - allow people to have a different color for the "file headers"
   (DIFF_METAINFO) and for the "fragment header" (DIFF_FRAGINFO). Also,
   make a difference between "normal color" and "reset colors"

 - default to red/green for old/new lines. That's the norm, I'd think.

 - instead of that eye-popping (and eye-ball-with-a-fondue-fork-popping)
   purple color for metadata, use bold-face for file headers, and cyan for
   the frag headers. I actually prefer the "gray background" for that, but
   it only works well in xterms, so COLOR_CYAN it is..

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'master' into next v1.4.1-rc1
Junio C Hamano [Thu, 22 Jun 2006 17:36:26 +0000 (10:36 -0700)]
Merge branch 'master' into next

* master:
  git-svn: fix commit --edit flag when using SVN:: libraries
  Makefile: do not force unneeded recompilation upon GIT_VERSION changes
  Check and document the options to prevent mistakes.
  Pass -DDEFAULT_GIT_TEMPLATE_DIR only where actually used.

17 years agoMerge branch 'js/lsfix'
Junio C Hamano [Thu, 22 Jun 2006 17:35:13 +0000 (10:35 -0700)]
Merge branch 'js/lsfix'

* js/lsfix:
  Initialize lock_file struct to all zero.
  Make git-update-ref a builtin
  Make git-update-index a builtin
  Make git-stripspace a builtin
  Make git-mailinfo a builtin
  Make git-mailsplit a builtin
  Make git-write-tree a builtin

17 years agoMerge branch 'ew/rebase'
Junio C Hamano [Thu, 22 Jun 2006 17:34:02 +0000 (10:34 -0700)]
Merge branch 'ew/rebase'

* ew/rebase:
  rebase --merge: fix for rebasing more than 7 commits.
  rebase: error out for NO_PYTHON if they use recursive merge
  Add renaming-rebase test.
  rebase: Allow merge strategies to be used when rebasing

17 years agoMerge branch 'jn/web'
Junio C Hamano [Thu, 22 Jun 2006 17:33:34 +0000 (10:33 -0700)]
Merge branch 'jn/web'

* jn/web:
  gitweb: whitespace cleanup around '='
  gitweb: Use $hash_base as $search_hash if possible
  gitweb: Make use of $PATH_INFO for project parameter
  Move $gitbin earlier in gitweb.cgi
  Add git version to gitweb output
  gitweb: whitespace cleanup
  gitweb: style done with stylesheet
  gitweb: A couple of page title tweaking
  Fix: Support for the standard mime.types map in gitweb
  gitweb: add type="text/css" to stylesheet link
  Make CSS file gitweb/gitweb.css more readable
  Fix gitweb stylesheet
  Support for the standard mime.types map in gitweb
  gitweb: text files for 'blob_plain' action without charset by default
  gitweb: safely output binary files for 'blob_plain' action
  Move gitweb style to gitweb.css

17 years agoMerge early parts of branch 'js/diff'
Junio C Hamano [Thu, 22 Jun 2006 17:32:58 +0000 (10:32 -0700)]
Merge early parts of branch 'js/diff'

17 years agoMerge branch 'jc/upload-corrupt'
Junio C Hamano [Thu, 22 Jun 2006 17:25:51 +0000 (10:25 -0700)]
Merge branch 'jc/upload-corrupt'

* jc/upload-corrupt:
  daemon: send stderr to /dev/null instead of closing.
  upload-pack/fetch-pack: support side-band communication
  Retire git-clone-pack
  upload-pack: prepare for sideband message support.
  upload-pack: avoid sending an incomplete pack upon failure

17 years agoMerge branch 'ew/rebase' into next
Junio C Hamano [Thu, 22 Jun 2006 09:29:58 +0000 (02:29 -0700)]
Merge branch 'ew/rebase' into next

* ew/rebase:
  rebase --merge: fix for rebasing more than 7 commits.

17 years agoMerge branch 'jn/web' into next
Junio C Hamano [Thu, 22 Jun 2006 09:29:50 +0000 (02:29 -0700)]
Merge branch 'jn/web' into next

* jn/web:
  gitweb: whitespace cleanup around '='
  gitweb: Use $hash_base as $search_hash if possible

17 years agogitweb: whitespace cleanup around '='
Jakub Narebski [Thu, 22 Jun 2006 06:52:57 +0000 (08:52 +0200)]
gitweb: whitespace cleanup around '='

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Use $hash_base as $search_hash if possible
Timo Hirvonen [Tue, 20 Jun 2006 13:41:05 +0000 (16:41 +0300)]
gitweb: Use $hash_base as $search_hash if possible

$hash (h parameter) does not always point to a commit. Use $hash_base as
$search_hash when it is defined.

Signed-off-by: Timo Hirvonen <tihirvon@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/waitpid'
Junio C Hamano [Thu, 22 Jun 2006 09:19:08 +0000 (02:19 -0700)]
Merge branch 'jc/waitpid'

* jc/waitpid:
  Restore SIGCHLD to SIG_DFL where we care about waitpid().

17 years agoMerge branch 'ff/c99'
Junio C Hamano [Thu, 22 Jun 2006 09:18:51 +0000 (02:18 -0700)]
Merge branch 'ff/c99'

* ff/c99:
  Remove all void-pointer arithmetic.
  Change types used in bitfields to be `int's.
  Don't use empty structure initializers.
  Cast pointers to `void *' when used in a format.
  Don't instantiate structures with FAMs.
  Initialize FAMs using `FLEX_ARRAY'.
  Remove ranges from switch statements.

17 years agoMerge branch 'pb/config'
Junio C Hamano [Thu, 22 Jun 2006 09:15:45 +0000 (02:15 -0700)]
Merge branch 'pb/config'

* pb/config:
  git_config: access() returns 0 on success, not > 0
  repo-config: Fix late-night bug
  Read configuration also from $HOME/.gitconfig
  Fix setting config variables with an alternative GIT_CONFIG
  Support for extracting configuration from different files

17 years agoMerge branch 'lt/objlist'
Junio C Hamano [Thu, 22 Jun 2006 09:15:22 +0000 (02:15 -0700)]
Merge branch 'lt/objlist'

* lt/objlist:
  Add "named object array" concept

17 years agogit-svn: fix commit --edit flag when using SVN:: libraries
Eric Wong [Thu, 22 Jun 2006 08:22:46 +0000 (01:22 -0700)]
git-svn: fix commit --edit flag when using SVN:: libraries

Trying to open an interactive editor in the console while stdout is
being piped to the parent process doesn't work out very well.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMakefile: do not force unneeded recompilation upon GIT_VERSION changes
Junio C Hamano [Thu, 22 Jun 2006 09:04:27 +0000 (02:04 -0700)]
Makefile: do not force unneeded recompilation upon GIT_VERSION changes

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorebase --merge: fix for rebasing more than 7 commits.
Junio C Hamano [Thu, 22 Jun 2006 08:44:54 +0000 (01:44 -0700)]
rebase --merge: fix for rebasing more than 7 commits.

Instead of using 4-digit numbers to name commits being rebased,
just use "cmt.$msgnum" string, with $msgnum as a decimal number
without leading zero padding.  This makes it possible to rebase
more than 9999 commits, but of more practical importance is that
the earlier code used "printf" to format already formatted
$msgnum and barfed when it counted up to 0008.  In other words,
the old code was incapable of rebasing more than 7 commits, and
this fixes that problem.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoCheck and document the options to prevent mistakes.
Eric W. Biederman [Wed, 21 Jun 2006 13:17:31 +0000 (07:17 -0600)]
Check and document the options to prevent mistakes.

When multiple recipients are given to git-send-email on the same
--cc line the code does not properly handle it.

Full and proper parsing of the email addresses so I can detect
which commas mean a new email address is more than I care to implement.

In particular this email address: "bibo,mao" <bibo.mao@intel.com>
must not be treated as two email addresses.

So this patch simply treats all commas in recipient lists as
an error and fails if one is given.

At the same time it documents that git-send-email wants multiple
instances of --cc specified on the command line if you want to
cc multiple recipients.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoPass -DDEFAULT_GIT_TEMPLATE_DIR only where actually used.
Yakov Lerner [Thu, 22 Jun 2006 01:47:00 +0000 (04:47 +0300)]
Pass -DDEFAULT_GIT_TEMPLATE_DIR only where actually used.

Before this patch, -DDEFAULT_GIT_TEMPLATE_DIR was passed on compilation
command line to all and every .c file compiled. In fact the macro
is used by only one .c file, and unused by all other .c files.
Remove -DDEFAULT_GIT_TEMPLATE_DIR where unused. Follow the example of
exec_cmd.o. Pass -DDEFAULT_GIT_TEMPLATE_DIR only where actually used.

Signed-off-by: Yakov Lerner <iler.ml@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'master' into next
Junio C Hamano [Thu, 22 Jun 2006 01:37:31 +0000 (18:37 -0700)]
Merge branch 'master' into next

* master:
  Fix grow_refs_hash()

17 years agoMerge branch 'jc/upload-corrupt' into next
Junio C Hamano [Wed, 21 Jun 2006 23:38:11 +0000 (16:38 -0700)]
Merge branch 'jc/upload-corrupt' into next

* jc/upload-corrupt:
  daemon: send stderr to /dev/null instead of closing.

17 years agoMerge branch 'jn/web' into next
Junio C Hamano [Wed, 21 Jun 2006 23:38:05 +0000 (16:38 -0700)]
Merge branch 'jn/web' into next

* jn/web:
  gitweb: Make use of $PATH_INFO for project parameter

17 years agodaemon: send stderr to /dev/null instead of closing.
Junio C Hamano [Wed, 21 Jun 2006 23:37:48 +0000 (16:37 -0700)]
daemon: send stderr to /dev/null instead of closing.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Make use of $PATH_INFO for project parameter
Jakub Narebski [Wed, 21 Jun 2006 15:06:39 +0000 (17:06 +0200)]
gitweb: Make use of $PATH_INFO for project parameter

Allow to have project name in the path part of URL, just after the name of
script. For example instead of gitweb.cgi?p=git.git you can write
gitweb.cgi/git.git or gitweb.cgi/git.git/

Not used in URLs inside gitweb; it means that the above alternate syntax
must be generated by hand, at least for now.

Side effect: project name parameter is now stripped of leading and
trailing slash before validation.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix grow_refs_hash()
Linus Torvalds [Wed, 21 Jun 2006 18:01:12 +0000 (11:01 -0700)]
Fix grow_refs_hash()

Earlier commit 3e4339e6f96e8c4f38a9c6607b98d3e96a2ed783 had a
thinko that did not check for collisions while repopulating the
objects in the new hash table.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'ew/rebase' into next
Junio C Hamano [Wed, 21 Jun 2006 10:56:41 +0000 (03:56 -0700)]
Merge branch 'ew/rebase' into next

* ew/rebase:
  rebase: error out for NO_PYTHON if they use recursive merge
  Add renaming-rebase test.
  rebase: Allow merge strategies to be used when rebasing
  object-refs: avoid division by zero

17 years agorebase: error out for NO_PYTHON if they use recursive merge
Eric Wong [Wed, 21 Jun 2006 10:04:42 +0000 (03:04 -0700)]
rebase: error out for NO_PYTHON if they use recursive merge

recursive merge relies on Python, and we can't perform
rename-aware merges without the recursive merge.  So bail out
before trying it.

The test won't work w/o recursive merge, either, so skip that,
too.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>