git
18 years agocleanups: Fix potential bugs in connect.c
Serge E. Hallyn [Mon, 17 Apr 2006 15:14:47 +0000 (10:14 -0500)] 
cleanups: Fix potential bugs in connect.c

The strncmp for ACK was ACK does not include the final space.
Presumably either we should either remove the trailing space,
or compare 4 chars (as this patch does).

'path' is sometimes strdup'ed, but never freed.

Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/boundary'
Junio C Hamano [Mon, 17 Apr 2006 22:03:11 +0000 (15:03 -0700)] 
Merge branch 'jc/boundary'

* jc/boundary:
  rev-list --boundary: show boundary commits even when limited otherwise.

18 years agoMerge branch 'jc/bottomless'
Junio C Hamano [Mon, 17 Apr 2006 22:03:10 +0000 (15:03 -0700)] 
Merge branch 'jc/bottomless'

* jc/bottomless:
  rev-list --bisect: limit list before bisecting.

18 years agoAllow empty lines in info/grafts
Yann Dirson [Mon, 17 Apr 2006 11:41:49 +0000 (13:41 +0200)] 
Allow empty lines in info/grafts

In addition to the existing comment support, that just allows the user
to use a convention that works pretty much everywhere else.

Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorev-list --boundary: show boundary commits even when limited otherwise.
Junio C Hamano [Mon, 17 Apr 2006 01:12:49 +0000 (18:12 -0700)] 
rev-list --boundary: show boundary commits even when limited otherwise.

The boundary commits are shown for UI like gitk to draw them as
soon as topo-order sorting allows, and should not be omitted by
get_revision() filtering logic.  As long as their immediate
child commits are shown, we should not filter them out.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMakefile fixups.
A Large Angry SCM [Mon, 17 Apr 2006 03:17:38 +0000 (20:17 -0700)] 
Makefile fixups.

Signed-off-by: A Large Angry SCM <gitzilla@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge git://git.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Mon, 17 Apr 2006 01:59:30 +0000 (18:59 -0700)] 
Merge git://git.kernel.org/pub/scm/gitk/gitk

* git://git.kernel.org/pub/scm/gitk/gitk:
  gitk: Fix bug caused by missing commitlisted elements

18 years agogitk: Fix bug caused by missing commitlisted elements
Paul Mackerras [Mon, 17 Apr 2006 00:27:59 +0000 (10:27 +1000)] 
gitk: Fix bug caused by missing commitlisted elements

This bug was reported by Yann Dirson, and results in an 'Error:
expected boolean value but got ""' dialog when scrolling to the bottom
of the graph under some circumstances.  The issue is that git-rev-list
isn't outputting all the boundary commits when it is asked for commits
affecting only certain files.  We already cope with that by adding the
missing boundary commits in addextraid, but there we weren't adding a
0 to the end of the commitlisted list when we added the extra id to
the end of the displayorder list.

This fixes it by appending 0 to commitlisted in addextraid, thus keeping
commitlisted and displayorder in sync.

Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agoreading $GIT_DIR/info/graft - skip comments correctly.
Junio C Hamano [Sun, 16 Apr 2006 21:24:56 +0000 (14:24 -0700)] 
reading $GIT_DIR/info/graft - skip comments correctly.

Noticed by Yann Dirson.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agopager: do not fork a pager if PAGER is set to empty.
Johannes Schindelin [Sun, 16 Apr 2006 02:44:25 +0000 (04:44 +0200)] 
pager: do not fork a pager if PAGER is set to empty.

This skips an extra pipe, and helps debugging tremendously.

[jc: PAGER=cat is a questionable hack and should be done as a separate
patch. ]

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff-options: add --patch-with-stat
Johannes Schindelin [Sat, 15 Apr 2006 11:41:18 +0000 (13:41 +0200)] 
diff-options: add --patch-with-stat

With this option, git prepends a diffstat in front of the patch.

Since I really, really do not know what a diffstat of a combined diff
("merge diff") should look like, the diffstat is not generated for these.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff-files --stat: do not dump core with unmerged index.
Junio C Hamano [Sun, 16 Apr 2006 01:38:32 +0000 (18:38 -0700)] 
diff-files --stat: do not dump core with unmerged index.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSupport "git cmd --help" syntax
Linus Torvalds [Sat, 15 Apr 2006 18:13:49 +0000 (11:13 -0700)] 
Support "git cmd --help" syntax

The "--help" argument is special, in that it is (along with "--version")
in that is taken by the "git" program itself rather than the sub-command,
and thus we've had the syntax "git --help cmd".

However, as anybody who has ever used CVS or some similar devil-spawn
program, it's confusing as h*ll when options before the sub-command act
differently from options after the sub-command, so this quick hack just
makes it acceptable to do "git cmd --help" instead, and get the exact same
result.

It may be hacky, but it's simple and does the trick.

Of course, this does not help if you use one of the non-builtin commands
without using the "git" helper. Ie you won't be getting a man-page just
because you do "git-rev-list --help". Don't expect us to be quite _that_
helpful.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff --stat: do not do its own three-dashes.
Junio C Hamano [Sat, 15 Apr 2006 21:02:10 +0000 (14:02 -0700)] 
diff --stat: do not do its own three-dashes.

I missed that "git-diff-* --stat" spits out three-dash separator
on its own without being asked.  Remove it.

When we output commit log followed by diff, perhaps --patch-with-stat,
for downstream consumer, we _would_ want the three-dash between
the message and the diff material, but that logic belongs to the
caller, not diff generator.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff-tree: typefix.
Junio C Hamano [Sat, 15 Apr 2006 10:22:00 +0000 (03:22 -0700)] 
diff-tree: typefix.

Recent diff_tree_setup_paths() update made it take a second
argument of type "struct diff_options", but we passed another
struct that happenes to have that type at the beginning by
mistake.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoGIT v1.3.0-rc4 v1.3.0-rc4
Junio C Hamano [Sat, 15 Apr 2006 06:21:34 +0000 (23:21 -0700)] 
GIT v1.3.0-rc4

I've merged everything I think is ready for 1.3.0, so this is
the final round -- hopefully I can release this with minimum
last-minute fixup as v1.3.0 early next week.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'dl/xdiff'
Junio C Hamano [Sat, 15 Apr 2006 05:58:17 +0000 (22:58 -0700)] 
Merge branch 'dl/xdiff'

* dl/xdiff:
  xdiff: post-process hunks to make them consistent.

18 years agoMerge branch 'js/diffstat'
Junio C Hamano [Sat, 15 Apr 2006 04:55:23 +0000 (21:55 -0700)] 
Merge branch 'js/diffstat'

* js/diffstat:
  diff --stat: no need to ask funcnames nor context.
  diff-options: add --stat (take 2)
  diff-options: add --stat (take 2)

18 years agoMerge branch 'jc/fix5500'
Junio C Hamano [Sat, 15 Apr 2006 04:55:22 +0000 (21:55 -0700)] 
Merge branch 'jc/fix5500'

* jc/fix5500:
  t5500: test fix

18 years agorev-list --bisect: limit list before bisecting.
Junio C Hamano [Fri, 14 Apr 2006 22:57:32 +0000 (15:57 -0700)] 
rev-list --bisect: limit list before bisecting.

I noticed bisect does not work well without both good and bad.
Running this script in git.git repository would give you quite
different results:

#!/bin/sh
        initial=e83c5163316f89bfbde7d9ab23ca2e25604af290

        mid0=`git rev-list --bisect ^$initial --all`

        git rev-list $mid0 | wc -l
        git rev-list ^$mid0 --all | wc -l

        mid1=`git rev-list --bisect --all`

        git rev-list $mid1 | wc -l
        git rev-list ^$mid1 --all | wc -l

The $initial commit is the very first commit you made.  The
first midpoint bisects things evenly as designed, but the latter
does not.

The reason I got interested in this was because I was wondering
if something like the following would help people converting a
huge repository from foreign SCM, or preparing a repository to
be fetched over plain dumb HTTP only:

        #!/bin/sh

        N=4
        P=.git/objects/pack
        bottom=

        while test 0 \< $N
        do
                N=$((N-1))
                if test -z "$bottom"
                then
                        newbottom=`git rev-list --bisect --all`
                else
                        newbottom=`git rev-list --bisect ^$bottom --all`
                fi
                if test -z "$bottom"
                then
                        rev_list="$newbottom"
                elif test 0 = $N
                then
                        rev_list="^$bottom --all"
                else
                        rev_list="^$bottom $newbottom"
                fi
                p=$(git rev-list --unpacked --objects $rev_list |
                    git pack-objects $P/pack)
                git show-index <$P/pack-$p.idx | wc -l
                bottom=$newbottom
        done

The idea is to pack older half of the history to one pack, then
older half of the remaining history to another, to continue a
few times, using finer granularity as we get closer to the tip.

This may not matter, since for a truly huge history, running
bisect number of times could be quite time consuming, and we
might be better off running "git rev-list --all" once into a
temporary file, and manually pick cut-off points from the
resulting list of commits.  After all we are talking about
"approximately half" for such an usage, and older history does
not matter much.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoClean up trailing whitespace when pretty-printing commits
Linus Torvalds [Sat, 15 Apr 2006 04:20:51 +0000 (21:20 -0700)] 
Clean up trailing whitespace when pretty-printing commits

Partly because we've messed up and now have some commits with trailing
whitespace, but partly because this also just simplifies the code, let's
remove trailing whitespace from the end when pretty-printing commits.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago"git cmd -h" for shell scripts.
Junio C Hamano [Fri, 14 Apr 2006 22:54:51 +0000 (15:54 -0700)] 
"git cmd -h" for shell scripts.

Wrappers that use sh-setup took --help but not -h.  Noticed by
Sébastien Pierre.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-log <diff-options> <paths> documentation
Junio C Hamano [Fri, 14 Apr 2006 19:59:09 +0000 (12:59 -0700)] 
git-log <diff-options> <paths> documentation

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRetire git-log.sh (take #4)
Junio C Hamano [Fri, 14 Apr 2006 19:47:10 +0000 (12:47 -0700)] 
Retire git-log.sh (take #4)

Noticed by Johannes.  We do not install it anymore, but still have
been shipping the source, which was crazy.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agostripspace: incomplete line fix (take #2)
Junio C Hamano [Fri, 14 Apr 2006 19:41:51 +0000 (12:41 -0700)] 
stripspace: incomplete line fix (take #2)

This fixes f4ee3eb68906f079dea45de4f1bbb03d68189eb3 breakage, which
added an extra trailing blank line after stripping trailing blank lines
by mistake.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot5500: test fix
Junio C Hamano [Fri, 14 Apr 2006 05:28:06 +0000 (22:28 -0700)] 
t5500: test fix

Relying on eye-candy progress bar was fragile to begin with.
Run fetch-pack with -k option, and count the objects that are in
the pack that were transferred from the other end.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff --stat: no need to ask funcnames nor context.
Junio C Hamano [Fri, 14 Apr 2006 04:35:54 +0000 (21:35 -0700)] 
diff --stat: no need to ask funcnames nor context.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix-up previous expr changes.
Junio C Hamano [Fri, 14 Apr 2006 02:05:38 +0000 (19:05 -0700)] 
Fix-up previous expr changes.

The regexp on the right hand side of expr : operator somehow was
broken.

expr 'z+pu:refs/tags/ko-pu' : 'z\+\(.*\)'

does not strip '+'; write 'z+\(.*\)' instead.

We probably should switch to shell based substring post 1.3.0;
that's not bashism but just POSIX anyway.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff-options: add --stat (take 2)
Johannes Schindelin [Thu, 13 Apr 2006 23:09:48 +0000 (01:09 +0200)] 
diff-options: add --stat (take 2)

... and a fix for an invalid free():

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoxdiff: post-process hunks to make them consistent.
Davide Libenzi [Thu, 13 Apr 2006 23:45:13 +0000 (16:45 -0700)] 
xdiff: post-process hunks to make them consistent.

18 years agodiff-options: add --stat (take 2)
Johannes Schindelin [Thu, 13 Apr 2006 22:15:30 +0000 (00:15 +0200)] 
diff-options: add --stat (take 2)

Now, you can say "git diff --stat" (to get an idea how many changes are
uncommitted), or "git log --stat".

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoShell utilities: Guard against expr' magic tokens.
Mark Wooding [Thu, 13 Apr 2006 22:01:24 +0000 (22:01 +0000)] 
Shell utilities: Guard against expr' magic tokens.

Some words, e.g., `match', are special to expr(1), and cause strange
parsing effects.  Track down all uses of expr and mangle the arguments
so that this isn't a problem.

Signed-off-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot3600-rm: skip failed-remove test when we cannot make an unremovable file.
Junio C Hamano [Thu, 13 Apr 2006 18:57:57 +0000 (11:57 -0700)] 
t3600-rm: skip failed-remove test when we cannot make an unremovable file.

When running t3600-rm test under fakeroot (or as root), we
cannot make a file unremovable with "chmod a-w .".  Detect this
case early and skip that test.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUse less memory in "git log"
Linus Torvalds [Thu, 13 Apr 2006 17:01:02 +0000 (10:01 -0700)] 
Use less memory in "git log"

This trivially avoids keeping the commit message data around after we
don't need it any more, avoiding a continually growing "git log" memory
footprint.

It's not a huge deal, but it's somewhat noticeable. For the current kernel
tree, doing a full "git log" I got

 - before: /usr/bin/time git log > /dev/null
0.81user 0.02system 0:00.84elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+8851minor)pagefaults 0swaps

 - after: /usr/bin/time git log > /dev/null
0.79user 0.03system 0:00.83elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+5039minor)pagefaults 0swaps

ie the touched pages dropped from 8851 to 5039. For the historic kernel
archive, the numbers are 18357->11037 minor page faults.

We could/should in theory free the commits themselves, but that's really a
lot harder, since during revision traversal we may hit the same commit
twice through different children having it as a parent, even after we've
shown it once (when that happens, we'll silently ignore it next time, but
we still need the "struct commit" to know).

And as the commit message data is clearly the biggest part of the commit,
this is the really easy 60% solution.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-log: do not output excess blank line between commits
Junio C Hamano [Thu, 13 Apr 2006 07:26:21 +0000 (00:26 -0700)] 
git-log: do not output excess blank line between commits

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMakefile: $(MAKE) check-docs
Junio C Hamano [Thu, 13 Apr 2006 07:17:19 +0000 (00:17 -0700)] 
Makefile: $(MAKE) check-docs

This target lists undocumented commands, and/or whose document
is not referenced from the main git documentation.

For now, there are some exceptions I added primarily because I
lack the energy to document them myself:

 - merge backends (we should really document them)
 - ssh-push/ssh-pull (does anybody still use them?)
 - annotate and blame (maybe after one of them eats the other ;-)

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation: add a couple of missing docs.
Junio C Hamano [Thu, 13 Apr 2006 07:09:54 +0000 (00:09 -0700)] 
Documentation: add a couple of missing docs.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/combine' into next
Junio C Hamano [Wed, 12 Apr 2006 20:24:48 +0000 (13:24 -0700)] 
Merge branch 'jc/combine' into next

* jc/combine:
  stripspace: make sure not to leave an incomplete line.
  git-commit: do not muck with commit message when no_edit is set.
  When showing a commit message, do not lose an incomplete line.
  Retire t5501-old-fetch-and-upload test.
  combine-diff: type fix.

18 years agoMerge branch 'master' into jc/combine
Junio C Hamano [Wed, 12 Apr 2006 20:24:04 +0000 (13:24 -0700)] 
Merge branch 'master' into jc/combine

* master:
  stripspace: make sure not to leave an incomplete line.
  git-commit: do not muck with commit message when no_edit is set.
  When showing a commit message, do not lose an incomplete line.
  Retire t5501-old-fetch-and-upload test.

18 years agocombine-diff: type fix.
Junio C Hamano [Wed, 12 Apr 2006 20:23:50 +0000 (13:23 -0700)] 
combine-diff: type fix.

The variable hunk_end points at a line number, which is
represented as unsigned long by all the other variables.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agostripspace: make sure not to leave an incomplete line.
Junio C Hamano [Wed, 12 Apr 2006 20:10:27 +0000 (13:10 -0700)] 
stripspace: make sure not to leave an incomplete line.

When dealing with a commit log message for human consumption, it
never makes sense to keep a log that ends with an incomplete
line, so make it a part of the clean-up process done by
git-stripspace.

Acked-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit: do not muck with commit message when no_edit is set.
Junio C Hamano [Wed, 12 Apr 2006 18:45:18 +0000 (11:45 -0700)] 
git-commit: do not muck with commit message when no_edit is set.

Spotted by Linus and Darrin Thompson.  When we took a commit
message from -F <file> with an incomplete line, we appended "git
status" output, which ended up attaching a lone "#" at the end.

We still need the "do we have anything to commit?" check by
running "status" (which has to know what to do in different
cases with -i/-o/-a), but there is no point appending its output
to the proposed commit message given by the user.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoWhen showing a commit message, do not lose an incomplete line.
Linus Torvalds [Wed, 12 Apr 2006 18:31:23 +0000 (11:31 -0700)] 
When showing a commit message, do not lose an incomplete line.

18 years agoRetire t5501-old-fetch-and-upload test.
Junio C Hamano [Tue, 11 Apr 2006 23:35:39 +0000 (16:35 -0700)] 
Retire t5501-old-fetch-and-upload test.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/combine' into next
Junio C Hamano [Tue, 11 Apr 2006 21:34:59 +0000 (14:34 -0700)] 
Merge branch 'jc/combine' into next

* jc/combine:
  combine-diff: fix hunks at the end (take #2).
  combine-diff: do not lose hunks with only deletion at end.

18 years agoMerge branch 'jc/diff' into next
Junio C Hamano [Tue, 11 Apr 2006 21:34:53 +0000 (14:34 -0700)] 
Merge branch 'jc/diff' into next

* jc/diff:
  blame and friends: adjust to multiple pathspec change.
  git log --full-diff
  tree-diff: do not assume we use only one pathspec

18 years agocombine-diff: fix hunks at the end (take #2).
Junio C Hamano [Tue, 11 Apr 2006 21:31:31 +0000 (14:31 -0700)] 
combine-diff: fix hunks at the end (take #2).

The previous round showed the delete-only hunks at the end, but
forgot to mark them interesting when they were.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocombine-diff: do not lose hunks with only deletion at end.
Junio C Hamano [Tue, 11 Apr 2006 10:13:29 +0000 (03:13 -0700)] 
combine-diff: do not lose hunks with only deletion at end.

We used to lose hunks that appear at the end and have only
deletion.  This makes sure that the record beyond the end of
file (which holds such deletions) is examined.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'ds/index' into next
Junio C Hamano [Tue, 11 Apr 2006 18:52:36 +0000 (11:52 -0700)] 
Merge branch 'ds/index' into next

* ds/index:
  Replace index() with strchr().
  Solaris 9 also wants our own unsetenv/setenv.
  Retire git-log.sh (take #3)

18 years agoMerge branch 'jc/withraw' into next
Junio C Hamano [Tue, 11 Apr 2006 18:52:01 +0000 (11:52 -0700)] 
Merge branch 'jc/withraw' into next

* jc/withraw:
  Separate the raw diff and patch with a newline
  Document --patch-with-raw

18 years agoReplace index() with strchr().
Dennis Stosberg [Tue, 11 Apr 2006 16:37:58 +0000 (18:37 +0200)] 
Replace index() with strchr().

strchr() is more portable than index() and is used everywhere in
git already.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSolaris 9 also wants our own unsetenv/setenv.
Dennis Stosberg [Tue, 11 Apr 2006 18:42:26 +0000 (11:42 -0700)] 
Solaris 9 also wants our own unsetenv/setenv.

[jc: the original had "index() is evil" but that should be a
separate patch.]

18 years agoRetire git-log.sh (take #3)
Junio C Hamano [Tue, 11 Apr 2006 18:29:36 +0000 (11:29 -0700)] 
Retire git-log.sh (take #3)

Do not install built-in commands as separate files -- use
hardlinks instead.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSeparate the raw diff and patch with a newline
Petr Baudis [Tue, 11 Apr 2006 11:30:46 +0000 (13:30 +0200)] 
Separate the raw diff and patch with a newline

More friendly for human reading I believe, and possibly friendlier to some
parsers (although only by an epsilon).

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocument --patch-with-raw
Petr Baudis [Tue, 11 Apr 2006 11:22:17 +0000 (13:22 +0200)] 
Document --patch-with-raw

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'eb/apply' into next
Junio C Hamano [Tue, 11 Apr 2006 02:45:30 +0000 (19:45 -0700)] 
Merge branch 'eb/apply' into next

* eb/apply:
  Implement limited context matching in git-apply.

18 years agoMerge branch 'jc/withraw' into next
Junio C Hamano [Tue, 11 Apr 2006 02:44:35 +0000 (19:44 -0700)] 
Merge branch 'jc/withraw' into next

* jc/withraw:
  diff-* --patch-with-raw
  Retire git-log.sh (take#2)
  Retire git-log.sh
  Retire diffcore-pathspec.
  Improve the git-diff-tree -c/-cc documentation

18 years agodiff-* --patch-with-raw
Junio C Hamano [Tue, 11 Apr 2006 00:36:53 +0000 (17:36 -0700)] 
diff-* --patch-with-raw

This new flag outputs the diff-raw output and diff-patch output
at the same time.  Requested by Cogito.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoImplement limited context matching in git-apply.
Eric W. Biederman [Mon, 10 Apr 2006 09:33:06 +0000 (03:33 -0600)] 
Implement limited context matching in git-apply.

Ok this really should be the good version.  The option
handling has been reworked to be automation safe.

Currently to import the -mm tree I have to work around
git-apply by using patch.  Because some of Andrews
patches in quilt will only apply with fuzz.

I started out implementing a --fuzz option and then I realized
fuzz is not a very safe concept for an automated system.  What
you really want is a minimum number of context lines that must
match.  This allows policy to be set without knowing how many
lines of context a patch actually provides.   By default
the policy remains to match all provided lines of context.

Allowng git-apply to match a restricted set of context makes
it much easier to import the -mm tree into git.  I am still only
processing  1.5 to 1.6 patches a second for the 692 patches in
2.6.17-rc1-mm2 is still painful but it does help.

If I just loop through all of Andrews patches in order
and run git-apply --index -C1 I process the entire patchset
in 1m53s or about 6 patches per second.  So running
git-mailinfo, git-write-tree, git-commit-tree, and
git-update-ref everytime has a measurable impact,
and shows things can be speeded up even more.

All of these timings were taking on my poor 700Mhz Athlon
with 512MB of ram.  So people with fast machiens should
see much better performance.

When a match is found after the number of context are reduced a
warning is generated.  Since this is a rare event and possibly
dangerous this seems to make sense.  Unless you are patching
a single file the error message is a little bit terse at
the moment, but it should be easy to go back and fix.

I have also updated the documentation for git-apply to reflect
the new -C option that sets the minimum number of context
lines that must match.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRetire git-log.sh (take#2)
Junio C Hamano [Tue, 11 Apr 2006 00:37:58 +0000 (17:37 -0700)] 
Retire git-log.sh (take#2)

... and install a replacement properly.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoblame and friends: adjust to multiple pathspec change.
Junio C Hamano [Tue, 11 Apr 2006 01:14:54 +0000 (18:14 -0700)] 
blame and friends: adjust to multiple pathspec change.

This makes things that include revision.h build again.

Blame is also built, but I am not sure how well it works (or how
well it worked to begin with) -- it was relying on tree-diff to
be using whatever pathspec was used the last time, which smells
a bit suspicious.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRetire git-log.sh
Junio C Hamano [Tue, 11 Apr 2006 00:37:58 +0000 (17:37 -0700)] 
Retire git-log.sh

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit log --full-diff
Junio C Hamano [Mon, 10 Apr 2006 23:40:40 +0000 (16:40 -0700)] 
git log --full-diff

Without this flag, "git log -p paths..." shows commits that
touch the specified paths, and diffs about the same specified
paths.  With this, the full diff is shown for commits that touch
the specified paths.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotree-diff: do not assume we use only one pathspec
Junio C Hamano [Mon, 10 Apr 2006 23:39:11 +0000 (16:39 -0700)] 
tree-diff: do not assume we use only one pathspec

The way tree-diff was set up assumed we would use only one set
of pathspec during the entire life of the program.  Move the
pathspec related static variables out to diff_options structure
so that we can filter commits with one set of paths while show
the actual diffs using different set of paths.

I suspect this breaks blame.c, and makes "git log paths..." to
default to the --full-diff, the latter of which is dealt with
the next commit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into jc/diff
Junio C Hamano [Mon, 10 Apr 2006 23:44:59 +0000 (16:44 -0700)] 
Merge branch 'master' into jc/diff

* master:
  Make "--parents" logs also be incremental
  Retire diffcore-pathspec.
  Improve the git-diff-tree -c/-cc documentation

18 years agoMerge branch 'lt/rev'
Junio C Hamano [Mon, 10 Apr 2006 22:58:41 +0000 (15:58 -0700)] 
Merge branch 'lt/rev'

* lt/rev:
  Make "--parents" logs also be incremental

18 years agoRetire diffcore-pathspec.
Junio C Hamano [Mon, 10 Apr 2006 22:57:24 +0000 (15:57 -0700)] 
Retire diffcore-pathspec.

Nobody except diff-stages used it -- the callers instead filtered
the input to diffcore themselves.  Make diff-stages do that as
well and retire diffcore-pathspec.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoImprove the git-diff-tree -c/-cc documentation
Petr Baudis [Sun, 9 Apr 2006 13:43:17 +0000 (15:43 +0200)] 
Improve the git-diff-tree -c/-cc documentation

This tries to clarify the -c/-cc documentation and clean up the style and
grammar.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/diff' into next
Junio C Hamano [Sun, 9 Apr 2006 09:07:33 +0000 (02:07 -0700)] 
Merge branch 'jc/diff' into next

* jc/diff:
  git log [diff-tree options]...
  log-tree: separate major part of diff-tree.

18 years agogit log [diff-tree options]...
Junio C Hamano [Sun, 9 Apr 2006 08:59:03 +0000 (01:59 -0700)] 
git log [diff-tree options]...

And this makes "git log" to take common diff-tree options, so
that it can be used as "git whatchanged".

The recent revision walker updates by Linus to make path
limiting low-latency helps this quite a bit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agolog-tree: separate major part of diff-tree.
Junio C Hamano [Sun, 9 Apr 2006 08:11:11 +0000 (01:11 -0700)] 
log-tree: separate major part of diff-tree.

This separates out the part that deals with one-commit diff-tree
(and --stdin form) into a separate log-tree module.

There are two goals with this.  The more important one is to be
able to make this part available to "git log --diff", so that we
can have a native "git whatchanged" command.  Another is to
simplify the commit log generation part simpler.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake "--parents" logs also be incremental
Linus Torvalds [Sun, 9 Apr 2006 00:05:58 +0000 (17:05 -0700)] 
Make "--parents" logs also be incremental

The parent rewriting feature caused us to create the whole history in one
go, and then simplify it later, because of how rewrite_parents() had been
written. However, with a little tweaking, it's perfectly possible to do
even that one incrementally.

Right now, this doesn't really much matter, because every user of
"--parents" will probably generally _also_ use "--topo-order", which will
cause the old non-incremental behaviour anyway. However, I'm hopeful that
we could make even the topological sort incremental, or at least
_partially_ so (for example, make it incremental up to the first merge).

In the meantime, this at least moves things in the right direction, and
removes a strange special case.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoxdiff/xdiffi.c: fix warnings about possibly uninitialized variables
Marco Roeland [Sat, 8 Apr 2006 15:27:20 +0000 (17:27 +0200)] 
xdiff/xdiffi.c: fix warnings about possibly uninitialized variables

Compiling this module gave the following warnings (some double dutch!):

xdiff/xdiffi.c: In functie 'xdl_recs_cmp':
xdiff/xdiffi.c:298: let op: 'spl.i1' may be used uninitialized in this function
xdiff/xdiffi.c:298: let op: 'spl.i2' may be used uninitialized in this function
xdiff/xdiffi.c:219: let op: 'fbest1' may be used uninitialized in this function
xdiff/xdiffi.c:219: let op: 'bbest1' may be used uninitialized in this function

A superficial tracking of their usage, without deeper knowledge about the
algorithm, indeed confirms that there are code paths on which these
variables will be used uninitialized. In practice these code paths might never
be reached, but then these fixes will not change the algorithm. If these
code paths are ever reached we now at least have a predictable outcome. And
should the very small performance impact of these initializations be
noticeable, then they should at least be replaced by comments why certain
code paths will never be reached.

Some extra initializations in this patch now fix the warnings.

18 years agodiffcore-rename: fix merging back a broken pair.
Junio C Hamano [Sun, 9 Apr 2006 03:17:46 +0000 (20:17 -0700)] 
diffcore-rename: fix merging back a broken pair.

When a broken pair is matched up by rename detector to be merged
back, we do not want to say it is "dissimilar" with the
similarity index.  The output should just say they were changed,
taking the break score left by the earlier diffcore-break run if
any.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff: fix output of total-rewrite diff.
Junio C Hamano [Sun, 9 Apr 2006 02:45:39 +0000 (19:45 -0700)] 
diff: fix output of total-rewrite diff.

We did not read in the file data before emitting the
total-rewrite diff.  Noticed by Pasky.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-log: match rev-list --abbrev and --abbrev-commit
Junio C Hamano [Sat, 8 Apr 2006 05:34:11 +0000 (22:34 -0700)] 
git-log: match rev-list --abbrev and --abbrev-commit

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Sat, 8 Apr 2006 01:03:07 +0000 (18:03 -0700)] 
Merge branch 'master' into next

* master:
  GIT 1.3.0-rc3

18 years agoGIT 1.3.0-rc3 v1.3.0-rc3
Junio C Hamano [Sat, 8 Apr 2006 01:02:40 +0000 (18:02 -0700)] 
GIT 1.3.0-rc3

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'kh/svn'
Junio C Hamano [Sat, 8 Apr 2006 01:01:55 +0000 (18:01 -0700)] 
Merge branch 'kh/svn'

* kh/svn:
  git-svnimport: Don't assume that copied files haven't changed

18 years agoMerge branch 'jc/thinpack'
Junio C Hamano [Sat, 8 Apr 2006 01:00:16 +0000 (18:00 -0700)] 
Merge branch 'jc/thinpack'

* jc/thinpack:
  Thin pack generation: optimization.

18 years agoMerge branch 'jc/date'
Junio C Hamano [Sat, 8 Apr 2006 01:00:06 +0000 (18:00 -0700)] 
Merge branch 'jc/date'

* jc/date:
  date parsing: be friendlier to our European friends.

18 years agoMerge branch 'nh/http'
Junio C Hamano [Sat, 8 Apr 2006 00:59:36 +0000 (17:59 -0700)] 
Merge branch 'nh/http'

* nh/http:
  Fix compile with expat, but an old curl version
  http-fetch: add optional DAV-based pack list

18 years agoMerge branch 'ew/rev-abbrev'
Junio C Hamano [Sat, 8 Apr 2006 00:59:10 +0000 (17:59 -0700)] 
Merge branch 'ew/rev-abbrev'

* ew/rev-abbrev:
  rev-list --abbrev-commit

18 years agoMerge branch 'jc/blame'
Junio C Hamano [Sat, 8 Apr 2006 00:57:46 +0000 (17:57 -0700)] 
Merge branch 'jc/blame'

* jc/blame:
  blame -S <ancestry-file>
  Match ofs/cnt types in diff interface.
  blame: use built-in xdiff
  combine-diff: move the code to parse hunk-header into common library.
  combine-diff: refactor built-in xdiff interface.
  combine-diff: use built-in xdiff.

18 years agoMerge branch 'master' into next
Junio C Hamano [Fri, 7 Apr 2006 23:53:06 +0000 (16:53 -0700)] 
Merge branch 'master' into next

* master:
  gitk: Fix incorrect invocation of getmergediffline
  [PATCH] gitk: Fix searching for filenames in gitk
  count-delta: match get_delta_hdr_size() changes.
  check patch_delta bounds more carefully

18 years agoMerge git://git.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Fri, 7 Apr 2006 23:52:59 +0000 (16:52 -0700)] 
Merge git://git.kernel.org/pub/scm/gitk/gitk

* git://git.kernel.org/pub/scm/gitk/gitk:
  gitk: Fix incorrect invocation of getmergediffline
  [PATCH] gitk: Fix searching for filenames in gitk

18 years agoMerge branch 'maint'
Junio C Hamano [Fri, 7 Apr 2006 23:51:55 +0000 (16:51 -0700)] 
Merge branch 'maint'

* maint:
  count-delta: match get_delta_hdr_size() changes.
  check patch_delta bounds more carefully

18 years agocount-delta: match get_delta_hdr_size() changes. v1.2.6
Junio C Hamano [Fri, 7 Apr 2006 23:48:09 +0000 (16:48 -0700)] 
count-delta: match get_delta_hdr_size() changes.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocheck patch_delta bounds more carefully
Nicolas Pitre [Fri, 7 Apr 2006 19:26:10 +0000 (15:26 -0400)] 
check patch_delta bounds more carefully

Let's avoid going south with invalid delta data.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/thinpack' into next
Junio C Hamano [Fri, 7 Apr 2006 09:13:20 +0000 (02:13 -0700)] 
Merge branch 'jc/thinpack' into next

* jc/thinpack:
  Thin pack generation: optimization.

18 years agoMerge branch 'ew/rev-abbrev' into next
Junio C Hamano [Fri, 7 Apr 2006 09:12:55 +0000 (02:12 -0700)] 
Merge branch 'ew/rev-abbrev' into next

* ew/rev-abbrev:
  rev-list --abbrev-commit

18 years agoMerge branch 'jc/blame' into next
Junio C Hamano [Fri, 7 Apr 2006 09:12:48 +0000 (02:12 -0700)] 
Merge branch 'jc/blame' into next

* jc/blame:
  blame -S <ancestry-file>
  Match ofs/cnt types in diff interface.

18 years agoMerge branch 'kh/svn' into next
Junio C Hamano [Fri, 7 Apr 2006 09:12:18 +0000 (02:12 -0700)] 
Merge branch 'kh/svn' into next

* kh/svn:
  git-svnimport: Don't assume that copied files haven't changed
  Add Documentation/technical/pack-format.txt
  Added Packing Heursitics IRC writeup.
  Add documentation for git-imap-send.

18 years agogit-svnimport: Don't assume that copied files haven't changed
Karl Hasselström [Fri, 7 Apr 2006 06:06:09 +0000 (08:06 +0200)] 
git-svnimport: Don't assume that copied files haven't changed

Don't assume that a file that SVN claims was copied from somewhere
else is bit-for-bit identical with its parent, since SVN allows
changes to copied files before they are committed.

Without this fix, such copy-modify-commit operations causes the
imported file to lack the "modify" part -- that is, we get subtle data
corruption.

Signed-off-by: Karl Hasselström <kha@treskal.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorev-list --abbrev-commit
Junio C Hamano [Fri, 7 Apr 2006 04:32:36 +0000 (21:32 -0700)] 
rev-list --abbrev-commit

This should make --pretty=oneline a whole lot more readable for
people using 80-column terminals.  Originally from Eric Wong.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoThin pack generation: optimization.
Junio C Hamano [Thu, 6 Apr 2006 06:24:57 +0000 (23:24 -0700)] 
Thin pack generation: optimization.

Jens Axboe noticed that recent "git push" has become very slow
since we made --thin transfer the default.

Thin pack generation to push a handful revisions that touch
relatively small number of paths out of huge tree was stupid; it
registered _everything_ from the excluded revisions.  As a
result, "Counting objects" phase was unnecessarily expensive.

This changes the logic to register the blobs and trees from
excluded revisions only for paths we are actually going to send
to the other end.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd Documentation/technical/pack-format.txt
Junio C Hamano [Fri, 7 Apr 2006 09:07:40 +0000 (02:07 -0700)] 
Add Documentation/technical/pack-format.txt

... along with the previous one, pack-heuristics, by popular
demand.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdded Packing Heursitics IRC writeup.
Jon Loeliger [Fri, 3 Mar 2006 01:19:29 +0000 (19:19 -0600)] 
Added Packing Heursitics IRC writeup.

Signed-off-by: Jon Loeliger <jdl@jdl.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoblame -S <ancestry-file>
Junio C Hamano [Fri, 7 Apr 2006 06:58:51 +0000 (23:58 -0700)] 
blame -S <ancestry-file>

This adds the -S <ancestry-file> option to blame, which is
needed by the CVS server emulation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMatch ofs/cnt types in diff interface.
Junio C Hamano [Fri, 7 Apr 2006 05:29:55 +0000 (22:29 -0700)] 
Match ofs/cnt types in diff interface.

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