git/git.git
15 years agoMerge branch 'pb/regex'
Junio C Hamano [Wed, 5 Apr 2006 21:06:26 +0000 (14:06 -0700)]
Merge branch 'pb/regex'

* pb/regex:
  On some platforms, certain headers need to be included before regex.h
  Support for pickaxe matching regular expressions

15 years agoAvoid a crash if realloc returns a different pointer.
Mike McCormack [Wed, 5 Apr 2006 14:22:19 +0000 (23:22 +0900)]
Avoid a crash if realloc returns a different pointer.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoAvoid a divide by zero if there's no messages to send.
Mike McCormack [Wed, 5 Apr 2006 14:22:52 +0000 (23:22 +0900)]
Avoid a divide by zero if there's no messages to send.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge git://git.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Wed, 5 Apr 2006 19:45:17 +0000 (12:45 -0700)]
Merge git://git.kernel.org/pub/scm/gitk/gitk

* git://git.kernel.org/pub/scm/gitk/gitk:
  [PATCH] Provide configurable UI font for gitk
  [PATCH] gitk: Use git wrapper to run git-ls-remote.
  [PATCH] gitk: add key bindings for selecting first and last commit
  gitk: Add a help menu item to display key bindings
  [PATCH] gitk: allow goto heads
  gitk: replace parent and children arrays with lists

15 years agoAdd git-clean command
Pavel Roskin [Wed, 5 Apr 2006 06:00:48 +0000 (02:00 -0400)]
Add git-clean command

This command removes untracked files from the working tree.  This
implementation is based on cg-clean with some simplifications.  The
documentation is included.

[jc: with trivial documentation fix, noticed by Jakub Narebski]

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch 'fix'
Junio C Hamano [Wed, 5 Apr 2006 09:50:54 +0000 (02:50 -0700)]
Merge branch 'fix'

* fix:
  diff_flush(): leakfix.
  parse_date(): fix parsing 03/10/2006

15 years agodiff_flush(): leakfix.
Junio C Hamano [Wed, 5 Apr 2006 09:06:49 +0000 (02:06 -0700)]
diff_flush(): leakfix.

We were leaking filepairs when output-format was set to
NO_OUTPUT.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoparse_date(): fix parsing 03/10/2006
Junio C Hamano [Wed, 5 Apr 2006 05:57:15 +0000 (22:57 -0700)]
parse_date(): fix parsing 03/10/2006

The comment associated with the date parsing code for three
numbers separated with slashes or dashes implied we wanted to
interpret using this order:

yyyy-mm-dd
yyyy-dd-mm
mm-dd-yy
dd-mm-yy

However, the actual code had the last two wrong, and making it
prefer dd-mm-yy format over mm-dd-yy.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Provide configurable UI font for gitk
Keith Packard [Tue, 4 Apr 2006 07:19:45 +0000 (00:19 -0700)]
[PATCH] Provide configurable UI font for gitk

This makes the font used in the UI elements of gitk configurable in the
same way the other fonts are. The default fonts used in the Xft build of
tk8.5 are particularily horrific, making this change more important
there.

Signed-off-by: Keith Packard <keithp@neko.keithp.com>
Acked-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years ago[PATCH] gitk: Use git wrapper to run git-ls-remote.
Mark Wooding [Tue, 4 Apr 2006 06:24:55 +0000 (23:24 -0700)]
[PATCH] gitk: Use git wrapper to run git-ls-remote.

For some reason, the Cygwin Tcl's `exec' command has trouble running
scripts.  Fix this by using the C `git' wrapper.  Other GIT programs run
by gitk are written in C already, so we don't need to incur a
performance hit of going via the wrapper (which I'll bet isn't pretty
under Cygwin).

Signed-off-by: Mark Wooding <mdw@distorted.org.uk>
Acked-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years ago[PATCH] gitk: add key bindings for selecting first and last commit
Rutger Nijlunsing [Wed, 5 Apr 2006 00:24:03 +0000 (10:24 +1000)]
[PATCH] gitk: add key bindings for selecting first and last commit

For a keyboard addict like me some keys are still missing from
gitk. Especially a key to select a commit when no commit is selected,
like just after startup. While we're at it, complete the bindings for
moving the view seperately from the selected line. Currently, the up
and down keys act on the selected line while pageup and pagedown act
on the commits viewed.

The idea is to have to normal keys change the selected line:
  - Home selects first commit
  - End selects last commit
  - Up selects previous commit
  - Down selects next commit
  - PageUp moves selected line one page up
  - PageDown moves selected line one page down
...and together with the Control key, it moves the commits view:
  - Control-Home views first page of commits
  - Control-End views last page of commits
  - Control-Up moves commit view one line up
  - Control-Down moves commit view one line down
  - Control-PageUp moves commit view one page up
  - Control-PageDown moves commit view one page down

Signed-off-By: Rutger Nijlunsing <gitk@tux.tmfweb.nl>
and with some cleanups and simplifications...
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agoOn some platforms, certain headers need to be included before regex.h
Johannes Schindelin [Tue, 4 Apr 2006 23:01:03 +0000 (01:01 +0200)]
On some platforms, certain headers need to be included before regex.h

Happily, these are already included in cache.h, which is included anyway...
so: change the order of includes.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agocvsimport: use git-update-ref when updating
Johannes Schindelin [Thu, 30 Mar 2006 12:06:15 +0000 (14:06 +0200)]
cvsimport: use git-update-ref when updating

This simplifies code, and also fixes a subtle bug: when importing in a
shared repository, where another user last imported from CVS, cvsimport
used to complain that it could not open <branch> for update.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogitk: Add a help menu item to display key bindings
Paul Mackerras [Tue, 4 Apr 2006 23:39:51 +0000 (09:39 +1000)]
gitk: Add a help menu item to display key bindings

Suggested by Paul Schulz.  I made it a separate entry under the Help
menu rather than putting it in the About box, though.

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years ago[PATCH] gitk: allow goto heads
Stephen Rothwell [Thu, 30 Mar 2006 05:13:12 +0000 (16:13 +1100)]
[PATCH] gitk: allow goto heads

This patch allows you to enter a head name in the SHA1 id: field.

It also removes some unnecessary global declarations.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agoGIT 1.3.0-rc2 v1.3.0-rc2
Junio C Hamano [Tue, 4 Apr 2006 21:52:53 +0000 (14:52 -0700)]
GIT 1.3.0-rc2

Bunch of cleanups with a few notable enhancements since
1.3.0-rc1:

 - revision traversal infrastructure is updated so that
   existence of paths limiters and/or --max-age does not cause
   it to call limit_list().  This helps the latency working with
   the command quite a bit.

 - comes with updated gitk.

One notable fix is to make sure that the IO is restarted upon
signal even on platforms whose default signal semantics is not
to do so.  This is the fix for the notorious "clone is broken
since 1.2.2 on Solaris" problem.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge in xdiff cleanup pieces
Junio C Hamano [Tue, 4 Apr 2006 21:43:57 +0000 (14:43 -0700)]
Merge in xdiff cleanup pieces

15 years agoSet HTTP user agent to git/GIT_VERSION
Nick Hengeveld [Tue, 4 Apr 2006 17:11:29 +0000 (10:11 -0700)]
Set HTTP user agent to git/GIT_VERSION

Useful for diagnostics/troubleshooting to know which client versions are
hitting your server.

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogit-ls-remote: send no-cache header when fetching info/refs
Nick Hengeveld [Tue, 4 Apr 2006 18:01:30 +0000 (11:01 -0700)]
git-ls-remote: send no-cache header when fetching info/refs

Proxies should not cache this file as it can cause a client to end up with
a stale version, as reported here:

http://marc.theaimsgroup.com/?l=git&m=114407944125389

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoSupport for pickaxe matching regular expressions
Petr Baudis [Wed, 29 Mar 2006 00:16:33 +0000 (02:16 +0200)]
Support for pickaxe matching regular expressions

git-diff-* --pickaxe-regex will change the -S pickaxe to match
POSIX extended regular expressions instead of fixed strings.

The regex.h library is a rather stupid interface and I like pcre too, but
with any luck it will be everywhere we will want to run Git on, it being
POSIX.2 and all. I'm not sure if we can expect platforms like AIX to
conform to POSIX.2 or if win32 has regex.h. We might add a flag to
Makefile if there is a portability trouble potential.

Signed-off-by: Petr Baudis <pasky@suse.cz>
15 years agoMerge branch 'pe/cleanup'
Junio C Hamano [Tue, 4 Apr 2006 20:43:00 +0000 (13:43 -0700)]
Merge branch 'pe/cleanup'

* pe/cleanup:
  Replace xmalloc+memset(0) with xcalloc.
  Use blob_, commit_, tag_, and tree_type throughout.

15 years agoMerge branch 'lt/fix-sol-pack'
Junio C Hamano [Tue, 4 Apr 2006 20:42:02 +0000 (13:42 -0700)]
Merge branch 'lt/fix-sol-pack'

* lt/fix-sol-pack:
  Use sigaction and SA_RESTART in read-tree.c; add option in Makefile.
  safe_fgets() - even more anal fgets()
  pack-objects: be incredibly anal about stdio semantics
  Fix Solaris stdio signal handling stupidities

15 years agoReplace xmalloc+memset(0) with xcalloc.
Peter Eriksen [Mon, 3 Apr 2006 18:30:46 +0000 (19:30 +0100)]
Replace xmalloc+memset(0) with xcalloc.

Signed-off-by: Peter Eriksen <s022018@student.dtu.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoUse blob_, commit_, tag_, and tree_type throughout.
Peter Eriksen [Sun, 2 Apr 2006 12:44:09 +0000 (14:44 +0200)]
Use blob_, commit_, tag_, and tree_type throughout.

This replaces occurences of "blob", "commit", "tag", and "tree",
where they're really used as type specifiers, which we already
have defined global constants for.

Signed-off-by: Peter Eriksen <s022018@student.dtu.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoClean-up trivially redundant diff.
Davide Libenzi [Tue, 4 Apr 2006 01:47:55 +0000 (18:47 -0700)]
Clean-up trivially redundant diff.

Also corrects the line numbers in unified output when using
zero lines context.

15 years agocontrib/git-svn: handle array values correctly
Eric Wong [Tue, 4 Apr 2006 00:41:44 +0000 (17:41 -0700)]
contrib/git-svn: handle array values correctly

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agocontrib/git-svn: make sure our git-svn is up-to-date for test
Eric Wong [Mon, 3 Apr 2006 22:18:49 +0000 (15:18 -0700)]
contrib/git-svn: make sure our git-svn is up-to-date for test

Bugs like the last one could've been avoided if it weren't for
this...

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agocontrib/git-svn: ensure repo-config returns a value before using it
Eric Wong [Mon, 3 Apr 2006 22:18:48 +0000 (15:18 -0700)]
contrib/git-svn: ensure repo-config returns a value before using it

fetching from repos without an authors-file defined was broken.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoUse sigaction and SA_RESTART in read-tree.c; add option in Makefile. v1.2.5
Jason Riedy [Sun, 2 Apr 2006 22:29:34 +0000 (15:29 -0700)]
Use sigaction and SA_RESTART in read-tree.c; add option in Makefile.

Might as well ape the sigaction change in read-tree.c to avoid
the same potential problems.  The fprintf status output will
be overwritten in a second, so don't bother guarding it.  Do
move the fputc after disabling SIGALRM to ensure we go to the
next line, though.

Also add a NO_SA_RESTART option in the Makefile in case someone
doesn't have SA_RESTART but does restart (maybe older HP/UX?).
We want the builder to chose this specifically in case the
system both lacks SA_RESTART and does not restart stdio calls;
a compat #define in git-compat-utils.h would silently allow
broken systems.

Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agosafe_fgets() - even more anal fgets()
Junio C Hamano [Tue, 4 Apr 2006 06:41:09 +0000 (23:41 -0700)]
safe_fgets() - even more anal fgets()

This is from Linus -- the previous round forgot to clear error
after EINTR case.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agofix repacking with lots of tags
Jim Radford [Mon, 3 Apr 2006 03:50:17 +0000 (20:50 -0700)]
fix repacking with lots of tags

Use git-rev-list's --all instead of git-rev-parse's to keep from
hitting the shell's argument list length limits when repacking
with lots of tags.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoDocumentation: revise top of git man page
J. Bruce Fields [Sun, 2 Apr 2006 21:54:34 +0000 (17:54 -0400)]
Documentation: revise top of git man page

I'm afraid I'll be accused of trying to suck all the jokes and the
personality out of the git documentation.  I'm not!  Really!

That said, "man git" is one of the first things a new user is likely try,
and it seems a little cruel to start off with a somewhat obscure joke
about the architecture of git.

So instead I'm trying for a relatively straightforward description of what
git does, and what features distinguish it from other systems, together
with immediate links to introductory documentation.

I also did some minor reorganization in an attempt to clarify the
classification of commands.  And revised a bit for conciseness (as is
obvious from the diffstat--hopefully I didn't cut anything important).

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agopack-objects: be incredibly anal about stdio semantics
Linus Torvalds [Sun, 2 Apr 2006 20:31:54 +0000 (13:31 -0700)]
pack-objects: be incredibly anal about stdio semantics

This is the "letter of the law" version of using fgets() properly in the
face of incredibly broken stdio implementations.  We can work around the
Solaris breakage with SA_RESTART, but in case anybody else is ever that
stupid, here's the "safe" (read: "insanely anal") way to use fgets.

It probably goes without saying that I'm not terribly impressed by
Solaris libc.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFix Solaris stdio signal handling stupidities
Linus Torvalds [Sun, 2 Apr 2006 20:28:27 +0000 (13:28 -0700)]
Fix Solaris stdio signal handling stupidities

This uses sigaction() to install the SIGALRM handler with SA_RESTART, so
that Solaris stdio doesn't break completely when a signal interrupts a
read.

Thanks to Jason Riedy for confirming the silly Solaris signal behaviour.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFix sparse warnings about non-ANSI function prototypes
Rene Scharfe [Sun, 2 Apr 2006 11:13:10 +0000 (13:13 +0200)]
Fix sparse warnings about non-ANSI function prototypes

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFix sparse warnings about usage of 0 instead of NULL
Rene Scharfe [Sun, 2 Apr 2006 11:13:01 +0000 (13:13 +0200)]
Fix sparse warnings about usage of 0 instead of NULL

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoRemove useless pointer update
Rene Scharfe [Sun, 2 Apr 2006 11:12:47 +0000 (13:12 +0200)]
Remove useless pointer update

buf is not used afterwards.  The compiler optimized the dead store out
anyway, but let's clean the source, too.

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogitk: replace parent and children arrays with lists
Paul Mackerras [Sun, 2 Apr 2006 10:47:40 +0000 (20:47 +1000)]
gitk: replace parent and children arrays with lists

This will make it easier to switch between views efficiently, and
turns out to be slightly faster as well.

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agocontrib/git-svn: documentation updates
Eric Wong [Sun, 2 Apr 2006 02:25:03 +0000 (18:25 -0800)]
contrib/git-svn: documentation updates

contrib/git-svn/git-svn.txt:
added git-repo-config key names for options
fixed quoting of "git-svn-HEAD" in the manpage
use preformatted text for examples

contrib/git-svn/Makefile:
add target to generate HTML:
http://git-svn.yhbt.net/git-svn.html

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agocontrib/git-svn: accept configuration via repo-config
Eric Wong [Sun, 2 Apr 2006 02:25:02 +0000 (18:25 -0800)]
contrib/git-svn: accept configuration via repo-config

repo-config keys are any of the long option names minus the '-'
characters

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorevision: --max-age alone does not need limit_list() anymore.
Junio C Hamano [Sun, 2 Apr 2006 03:13:22 +0000 (19:13 -0800)]
revision: --max-age alone does not need limit_list() anymore.

This makes git log --since=7.days to be streamable.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorevision: simplify argument parsing.
Junio C Hamano [Sun, 2 Apr 2006 02:38:25 +0000 (18:38 -0800)]
revision: simplify argument parsing.

This just moves code around to consolidate the part that sets
revs->limited to one place based on various flags.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorevision: --topo-order and --unpacked
Junio C Hamano [Sun, 2 Apr 2006 02:52:58 +0000 (18:52 -0800)]
revision: --topo-order and --unpacked

Now, using --unpacked without limit_list() does not make much
sense, but this is parallel to the earlier --max-age fix.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorevision: Fix --topo-order and --max-age with reachability limiting.
Linus Torvalds [Sun, 2 Apr 2006 00:35:06 +0000 (16:35 -0800)]
revision: Fix --topo-order and --max-age with reachability limiting.

What ends up not working very well at all is the combination of
"--topo-order" and the output filter in get_revision. It will
return NULL when we see the first commit out of date-order, even
if we have other commits coming.

So we really should do the "past the date order" thing in
get_revision() only if we have _not_ done it already in
limit_list().

Something like this.

The easiest way to test this is with just

gitk --since=3.days.ago

on the kernel tree. Without this patch, it tends to be pretty obviously
broken.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMake path-limiting be incremental when possible.
Linus Torvalds [Fri, 31 Mar 2006 01:05:25 +0000 (17:05 -0800)]
Make path-limiting be incremental when possible.

This makes git-rev-list able to do path-limiting without having to parse
all of history before it starts showing the results.

This makes things like "git log -- pathname" much more pleasant to use.

This is actually a pretty small patch, and the biggest part of it is
purely cleanups (turning the "goto next" statements into "continue"), but
it's conceptually a lot bigger than it looks.

What it does is that if you do a path-limited revision list, and you do
_not_ ask for pseudo-parenthood information, it won't do all the
path-limiting up-front, but instead do it incrementally in
"get_revision()".

This is an absolutely huge deal for anything like "git log -- <pathname>",
but also for some things that we don't do yet - like the "find where
things changed" logic I've described elsewhere, where we want to find the
previous revision that changed a file.

The reason I put "RFC" in the subject line is that while I've validated it
various ways, like doing

git-rev-list HEAD -- drivers/char/ | md5sum

before-and-after on the kernel archive, it's "git-rev-list" after all. In
other words, it's that really really subtle and complex central piece of
software. So while I think this is important and should go in asap, I also
think it should get lots of testing and eyeballs looking at the code.

Btw, don't even bother testing this with the git archive. git itself is so
small that parsing the whole revision history for it takes about a second
even with path limiting. The thing that _really_ shows this off is doing

git log drivers/

on the kernel archive, or even better, on the _historic_ kernel archive.

With this change, the response is instantaneous (although seeking to the
end of the result will obviously take as long as it ever did). Before this
change, the command would think about the result for tens of seconds - or
even minutes, in the case of the bigger old kernel archive - before
starting to output the results.

NOTE NOTE NOTE! Using path limiting with things like "gitk", which uses
the "--parents" flag to actually generate a pseudo-history of the
resulting commits won't actually see the improvement in interactivity,
since that forces git-rev-list to do the whole-history thing after all.

MAYBE we can fix that too at some point, but I won't promise anything.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMove "--parent" parsing into generic revision.c library code
Linus Torvalds [Fri, 31 Mar 2006 00:52:42 +0000 (16:52 -0800)]
Move "--parent" parsing into generic revision.c library code

Not only do we do it in both rev-list.c and git.c, the revision walking
code will soon want to know whether we should rewrite parenthood
information or not.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMakefile: many programs now depend on xdiff/lib.a having been built.
Junio C Hamano [Sat, 1 Apr 2006 00:23:46 +0000 (16:23 -0800)]
Makefile: many programs now depend on xdiff/lib.a having been built.

The dependency was not properly updated when we added this
library, breaking parallel build with $(MAKE) -j.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorev-list --boundary: fix re-injecting boundary commits.
Junio C Hamano [Fri, 31 Mar 2006 07:59:19 +0000 (23:59 -0800)]
rev-list --boundary: fix re-injecting boundary commits.

Marco reported that

$ git rev-list --boundary --topo-order --parents 5aa44d5..ab57c8d

misses these two boundary commits.

        c649657501bada28794a30102d9c13cc28ca0e5e
        eb38cc689e84a8fd01c1856e889fe8d3b4f1bfb4

Indeed, we can see that gitk shows these two commits at the
bottom, because the --boundary code failed to output them.

The code did not check to avoid pushing the same uninteresting
commit twice to the result list.  I am not sure why this fixes
the reported problem, but this seems to fix it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge git://git.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Fri, 31 Mar 2006 00:27:03 +0000 (16:27 -0800)]
Merge git://git.kernel.org/pub/scm/gitk/gitk

* git://git.kernel.org/pub/scm/gitk/gitk:
  gitk: Better workaround for arrows on diagonal line segments
  gitk: Allow top panes to scroll horizontally with mouse button 2
  gitk: Prevent parent link from overwriting commit headline
  gitk: Show diffs for boundary commits
  gitk: Use the new --boundary flag to git-rev-list

15 years agogitk: Better workaround for arrows on diagonal line segments
Paul Mackerras [Thu, 30 Mar 2006 23:45:14 +0000 (10:45 +1100)]
gitk: Better workaround for arrows on diagonal line segments

Instead of adding extra padding to create a vertical line segment at
the lower end of a line that has an arrow, this now just draws a very
short vertical line segment at the lower end.  This alternative
workaround for the Tk8.4 behaviour (not drawing arrows on diagonal
line segments) doesn't have the problem of making the graph very wide
when people do a lot of merges in a row (hi Junio :).

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agocontrib/git-svn: force GIT_DIR to an absolute path
Eric Wong [Thu, 30 Mar 2006 06:37:18 +0000 (22:37 -0800)]
contrib/git-svn: force GIT_DIR to an absolute path

We chdir internally, so we need a consistent GIT_DIR variable.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogit-clone: exit early if repo isn't specified
Yasushi SHOJI [Thu, 30 Mar 2006 17:01:23 +0000 (17:01 +0000)]
git-clone: exit early if repo isn't specified

git-clone without a repo isn't useful at all.  print message and get
out asap.

This patch also move the variable 'local' to where other variables are
initialized.

Signed-off-by: Yasushi SHOJI <yashi@atmark-techno.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMake git-clone to take long double-dashed origin option (--origin)
Yasushi SHOJI [Thu, 30 Mar 2006 17:00:43 +0000 (17:00 +0000)]
Make git-clone to take long double-dashed origin option (--origin)

git-clone currently take option '-o' to specify origin.  this patch
makes git-clone to take double-dashed option '--origin' and other
abbreviations in addtion to the current single-dashed option.

[jc: with minor fixups]

Signed-off-by: Yasushi SHOJI <yashi@atmark-techno.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogitk: Allow top panes to scroll horizontally with mouse button 2
Paul Mackerras [Thu, 30 Mar 2006 22:55:11 +0000 (09:55 +1100)]
gitk: Allow top panes to scroll horizontally with mouse button 2

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agogitk: Prevent parent link from overwriting commit headline
Paul Mackerras [Thu, 30 Mar 2006 22:54:24 +0000 (09:54 +1100)]
gitk: Prevent parent link from overwriting commit headline

When I made drawlineseg responsible for drawing the link to the first
child rather than drawparentlinks, that meant that the right-most X
value computed by drawparentlinks didn't include those first-child
links, and thus the first-child link could go over the top of the
commit headline.  This fixes it.

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agogitk: Show diffs for boundary commits
Paul Mackerras [Thu, 30 Mar 2006 09:50:40 +0000 (20:50 +1100)]
gitk: Show diffs for boundary commits

With this we run git-diff-tree on a commit even if we think it has
no parents, either because it really has no parents or because it
is a boundary commit.  This means that gitk shows the diff for a
boundary commit when it is selected.

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agotree/diff header cleanup.
Junio C Hamano [Thu, 30 Mar 2006 06:55:43 +0000 (22:55 -0800)]
tree/diff header cleanup.

Introduce tree-walk.[ch] and move "struct tree_desc" and
associated functions from various places.

Rename DIFF_FILE_CANON_MODE(mode) macro to canon_mode(mode) and
move it to cache.h.  This macro returns the canonicalized
st_mode value in the host byte order for files, symlinks and
directories -- to be compared with a tree_desc entry.
create_ce_mode(mode) in cache.h is similar but is intended to be
used for index entries (so it does not work for directories) and
returns the value in the network byte order.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoassume unchanged git: diff-index fix.
Junio C Hamano [Thu, 30 Mar 2006 07:00:42 +0000 (23:00 -0800)]
assume unchanged git: diff-index fix.

When the executable bit is untrustworthy and when we are
comparing the tree with the working tree, we tried to reuse the
mode bits recorded in the index incorrectly (the computation was
bogus on little endian architectures).  Just use mode from index
when it is a regular file.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogitk: Use the new --boundary flag to git-rev-list
Paul Mackerras [Thu, 30 Mar 2006 07:43:51 +0000 (18:43 +1100)]
gitk: Use the new --boundary flag to git-rev-list

With this, we can show the boundary (open-circle) commits immediately
after their last child, which looks much better than putting all the
boundary commits at the bottom of the graph.

Signed-off-by: Paul Mackerras <paulus@samba.org>
15 years agorevision.c "..B" syntax: constness fix
Junio C Hamano [Thu, 30 Mar 2006 07:30:52 +0000 (23:30 -0800)]
revision.c "..B" syntax: constness fix

The earlier change to make "..B" to mean "HEAD..B" (aka ^HEAD B)
has constness gotcha GCC complains.  Fix it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorevision arguments: ..B means HEAD..B, just like A.. means A..HEAD
Junio C Hamano [Thu, 30 Mar 2006 03:41:37 +0000 (19:41 -0800)]
revision arguments: ..B means HEAD..B, just like A.. means A..HEAD

For consistency reasons, we should probably allow that to be written as
just "..branch", the same way we can write "branch.." to mean "everything
in HEAD but not in "branch".

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorev-list --boundary
Junio C Hamano [Tue, 28 Mar 2006 07:58:34 +0000 (23:58 -0800)]
rev-list --boundary

With the new --boundary flag, the output from rev-list includes
the UNINTERESING commits at the boundary, which are usually not
shown.  Their object names are prefixed with '-'.

For example, with this graph:

              C side
             /
A---B---D master

You would get something like this:

$ git rev-list --boundary --header --parents side..master
D B
        tree D^{tree}
        parent B
        ... log message for commit D here ...
        \0-B A
        tree B^{tree}
        parent A
        ... log message for commit B here ...
        \0

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agorev-list: memory usage reduction.
Junio C Hamano [Wed, 29 Mar 2006 01:28:04 +0000 (17:28 -0800)]
rev-list: memory usage reduction.

We do not need to track object refs, neither we need to save commit
unless we are doing verbose header.  A lot of traversal happens
inside prepare_revision_walk() these days so setting things up before
calling that function is necessary.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Acked-by: Linus Torvalds <torvalds@osdl.org>
15 years agorev-list --no-merges: argument parsing fix.
Junio C Hamano [Tue, 28 Mar 2006 08:04:50 +0000 (00:04 -0800)]
rev-list --no-merges: argument parsing fix.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoxdiff: Show function names in hunk headers.
Mark Wooding [Tue, 28 Mar 2006 02:23:31 +0000 (03:23 +0100)]
xdiff: Show function names in hunk headers.

The speed of the built-in diff generator is nice; but the function names
shown by `diff -p' are /really/ nice.  And I hate having to choose.  So,
we hack xdiff to find the function names and print them.

xdiff has grown a flag to say whether to dig up the function names.  The
builtin_diff function passes this flag unconditionally.  I suppose it
could parse GIT_DIFF_OPTS, but it doesn't at the moment.  I've also
reintroduced the `function name' into the test suite, from which it was
removed in commit 3ce8f089.

The function names are parsed by a particularly stupid algorithm at the
moment: it just tries to find a line in the `old' file, from before the
start of the hunk, whose first character looks plausible.  Still, it's
most definitely a start.

Signed-off-by: Mark Wooding <mdw@distorted.org.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoAdd ALL_LDFLAGS to the git target.
Jason Riedy [Tue, 28 Mar 2006 01:15:24 +0000 (17:15 -0800)]
Add ALL_LDFLAGS to the git target.

For some reason, I need ALL_LDFLAGS in the git target only on
AIX.  Once it builds, only one test "fails" on AIX 5.1 with
1.3.0.rc1, t5500-fetch-pack.sh, but it looks like it's some
odd tool problem in the tester + my setup and not a real bug.

Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoGIT 1.3.0 rc1 v1.3.0-rc1
Junio C Hamano [Tue, 28 Mar 2006 00:08:29 +0000 (16:08 -0800)]
GIT 1.3.0 rc1

All of the things that were not in the "master" branch were
either cooked long enough in "next" without causing problems
(e.g. insanely fast rename detector or true built-in diff) or
isolated in a specific subsystem (e.g. tar-tree and svnimport).

So I am clearing the deck to prepare for a 1.3.0.  Remaining
wrinkles, if any, will be ironed in the "master" branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch ak/svn
Junio C Hamano [Tue, 28 Mar 2006 00:03:36 +0000 (16:03 -0800)]
Merge branch ak/svn

15 years agoMerge branch 'lt/diffgen' into next
Junio C Hamano [Mon, 27 Mar 2006 07:44:28 +0000 (23:44 -0800)]
Merge branch 'lt/diffgen' into next

* lt/diffgen:
  add clean and ignore rules for xdiff/
  Remove dependency on a file named "-lz"

15 years agoMerge branch 'master' into next
Junio C Hamano [Mon, 27 Mar 2006 07:44:14 +0000 (23:44 -0800)]
Merge branch 'master' into next

* master:
  Optionally do not list empty directories in git-ls-files --others
  Document git-rebase behavior on conflicts.
  Fix error handling for nonexistent names

15 years agoadd clean and ignore rules for xdiff/
Junio C Hamano [Mon, 27 Mar 2006 07:41:22 +0000 (23:41 -0800)]
add clean and ignore rules for xdiff/

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoOptionally do not list empty directories in git-ls-files --others
Petr Baudis [Sun, 26 Mar 2006 14:59:52 +0000 (16:59 +0200)]
Optionally do not list empty directories in git-ls-files --others

Without the --directory flag, git-ls-files wouldn't ever list directories,
producing no output for empty directories, which is good since they cannot
be added and they bear no content, even untracked one (if Git ever starts
tracking directories on their own, this should obviously change since the
content notion will change).

With the --directory flag however, git-ls-files would list even empty
directories. This may be good in some situations but sometimes you want to
prevent that. This patch adds a --no-empty-directory option which makes
git-ls-files omit empty directories.

Signed-off-by: Petr Baudis <pasky@suse.cz>
15 years agoDocument git-rebase behavior on conflicts.
J. Bruce Fields [Sun, 26 Mar 2006 20:29:28 +0000 (15:29 -0500)]
Document git-rebase behavior on conflicts.

15 years agoRemove dependency on a file named "-lz"
Johannes Schindelin [Sun, 26 Mar 2006 23:14:52 +0000 (01:14 +0200)]
Remove dependency on a file named "-lz"

By changing the dependency "$(LIB_H)" to "$(LIBS)", at least one version
of make thought that a file named "-lz" would be needed.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFix error handling for nonexistent names
Linus Torvalds [Mon, 27 Mar 2006 00:28:20 +0000 (16:28 -0800)]
Fix error handling for nonexistent names

When passing in a pathname pattern without the "--" separator on the
command line, we verify that the pathnames in question exist. However,
there were two bugs in that verification:

 - git-rev-parse would only check the first pathname, and silently allow
   any invalid subsequent pathname, whether it existed or not (which
   defeats the purpose of the check, and is also inconsistent with what
   git-rev-list actually does)

 - git-rev-list (and "git log" etc) would check each filename, but if the
   check failed, it would print the error using the first one, i.e.:

[torvalds@g5 git]$ git log Makefile bad-file
fatal: 'Makefile': No such file or directory

   instead of saying that it's 'bad-file' that doesn't exist.

This fixes both bugs.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch 'jc/thin' into next
Junio C Hamano [Sun, 26 Mar 2006 08:24:03 +0000 (00:24 -0800)]
Merge branch 'jc/thin' into next

* jc/thin:
  git-push: make --thin pack transfer the default.
  gitk: Fix two bugs reported by users
  gitk: Improve appearance of first child links
  gitk: Make downward-pointing arrows end in vertical line segment
  gitk: Don't change cursor at end of layout if find in progress
  gitk: Make commitdata an array rather than a list
  gitk: Fix display of diff lines beginning with --- or +++
  [PATCH] gitk: Make error_popup react to Return
  gitk: Fix a bug in drawing the selected line as a thick line
  gitk: Further speedups
  gitk: Various speed improvements
  gitk: Fix Update menu item
  gitk: Fix clicks on arrows on line ends
  gitk: New improved gitk
  contrib/git-svn: stabilize memory usage for big fetches

15 years agogit-push: make --thin pack transfer the default.
Junio C Hamano [Sun, 26 Mar 2006 02:27:15 +0000 (18:27 -0800)]
git-push: make --thin pack transfer the default.

Just in case it has problems, you can say "git push --no-thin".

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branches 'jc/clone' and 'jc/name'
Junio C Hamano [Sun, 26 Mar 2006 08:22:53 +0000 (00:22 -0800)]
Merge branches 'jc/clone' and 'jc/name'

* jc/clone:
  git-clone: typofix.
  clone: record the remote primary branch with remotes/$origin/HEAD
  revamp git-clone (take #2).
  revamp git-clone.
  fetch,parse-remote,fmt-merge-msg: refs/remotes/* support

* jc/name:
  sha1_name: make core.warnambiguousrefs the default.
  sha1_name: warning ambiguous refs.
  get_sha1_basic(): try refs/... and finally refs/remotes/$foo/HEAD
  core.warnambiguousrefs: warns when "name" is used and both "name" branch and tag exists.

15 years agoMerge branch 'jc/merge'
Junio C Hamano [Sun, 26 Mar 2006 08:22:48 +0000 (00:22 -0800)]
Merge branch 'jc/merge'

* jc/merge:
  git-merge knows some strategies want to skip trivial merges

15 years agoMerge branch 'lt/diffgen' into next
Junio C Hamano [Sun, 26 Mar 2006 08:15:44 +0000 (00:15 -0800)]
Merge branch 'lt/diffgen' into next

* lt/diffgen:
  true built-in diff: run everything in-core.

15 years agogit-svnimport: if a limit is specified, respect it
Anand Kumria [Sat, 25 Mar 2006 22:43:46 +0000 (09:43 +1100)]
git-svnimport: if a limit is specified, respect it

git-svnimport will import the same revision over and over again if a
limit (-l <rev>) has been specified. Instead if that revision has already
been processed, exit with an up-to-date message.

Signed-off-by: Anand Kumria <wildfire@progsoc.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge git://git.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Sun, 26 Mar 2006 08:13:25 +0000 (00:13 -0800)]
Merge git://git.kernel.org/pub/scm/gitk/gitk

* git://git.kernel.org/pub/scm/gitk/gitk:
  gitk: Fix two bugs reported by users
  gitk: Improve appearance of first child links
  gitk: Make downward-pointing arrows end in vertical line segment
  gitk: Don't change cursor at end of layout if find in progress
  gitk: Make commitdata an array rather than a list
  gitk: Fix display of diff lines beginning with --- or +++
  [PATCH] gitk: Make error_popup react to Return
  gitk: Fix a bug in drawing the selected line as a thick line
  gitk: Further speedups
  gitk: Various speed improvements
  gitk: Fix Update menu item
  gitk: Fix clicks on arrows on line ends
  gitk: New improved gitk

15 years agotrue built-in diff: run everything in-core.
Junio C Hamano [Sun, 26 Mar 2006 07:12:17 +0000 (23:12 -0800)]
true built-in diff: run everything in-core.

This stops using temporary files when we are using the built-in
diff (including the complete rewrite).

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agocontrib/git-svn: stabilize memory usage for big fetches
Eric Wong [Sun, 26 Mar 2006 02:52:31 +0000 (18:52 -0800)]
contrib/git-svn: stabilize memory usage for big fetches

We should be safely able to import histories with thousands
of revisions without hogging up lots of memory.

With this, we lose the ability to autocorrect mistakes when
people specify revisions in reverse, but it's probably no longer
a problem since we only have one method of log parsing nowadays.

I've added an extra check to ensure that revision numbers do
increment.

Also, increment the version number to 0.11.0.  I really should
just call it 1.0 soon...

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch 'ew/email' into next
Junio C Hamano [Sun, 26 Mar 2006 01:44:09 +0000 (17:44 -0800)]
Merge branch 'ew/email' into next

* ew/email:
  send-email: lazy-load Email::Valid and make it optional
  send-email: try to order messages in email clients more correctly
  send-email: Change from Mail::Sendmail to Net::SMTP
  send-email: use built-in time() instead of /bin/date '+%s'

15 years agoMerge branch 'lt/diffgen' into next
Junio C Hamano [Sun, 26 Mar 2006 01:44:01 +0000 (17:44 -0800)]
Merge branch 'lt/diffgen' into next

* lt/diffgen:
  built-in diff: minimum tweaks
  builtin-diff: \No newline at end of file.
  Use a *real* built-in diff generator

15 years agoMerge branch 'rs/tar-tree' into next
Junio C Hamano [Sun, 26 Mar 2006 01:43:22 +0000 (17:43 -0800)]
Merge branch 'rs/tar-tree' into next

* rs/tar-tree:
  tar-tree: Use the prefix field of a tar header
  tar-tree: Remove obsolete code
  tar-tree: Use write_entry() to write the archive contents
  tar-tree: Introduce write_entry()
  tar-tree: Use SHA1 of root tree for the basedir
  git-apply: safety fixes
  Removed bogus "<snap>" identifier.
  Clarify and expand some hook documentation.
  commit-tree: check return value from write_sha1_file()
  send-email: Identify author at the top when sending e-mail
  Format tweaks for asciidoc.

15 years agosend-email: lazy-load Email::Valid and make it optional
Eric Wong [Sun, 26 Mar 2006 00:47:12 +0000 (16:47 -0800)]
send-email: lazy-load Email::Valid and make it optional

It's not installed on enough machines, and is overkill most of
the time.  We'll fallback to a very basic regexp just in case,
but nothing like the monster regexp Email::Valid has to offer :)

Small cleanup from Merlyn.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agosend-email: try to order messages in email clients more correctly
Eric Wong [Sat, 25 Mar 2006 11:01:01 +0000 (03:01 -0800)]
send-email: try to order messages in email clients more correctly

If --no-chain-reply-to is set, patches may not always be ordered
correctly in email clients.  This patch makes sure each email
sent from a different second.

I chose to start with a time (slightly) in the past because
those are probably more likely in real-world usage and spam
filters might be more tolerant of them.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agosend-email: Change from Mail::Sendmail to Net::SMTP
Eric Wong [Sun, 26 Mar 2006 01:20:48 +0000 (17:20 -0800)]
send-email: Change from Mail::Sendmail to Net::SMTP

Net::SMTP is in the base Perl distribution, so users are more
likely to have it.  Net::SMTP also allows reusing the SMTP
connection, so sending multiple emails is faster.

[jc: tweaked X-Mailer further while we are at it.]

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agosend-email: use built-in time() instead of /bin/date '+%s'
Eric Wong [Sat, 25 Mar 2006 10:43:31 +0000 (02:43 -0800)]
send-email: use built-in time() instead of /bin/date '+%s'

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agobuilt-in diff: minimum tweaks
Junio C Hamano [Sat, 25 Mar 2006 20:16:17 +0000 (12:16 -0800)]
built-in diff: minimum tweaks

This fixes up a couple of minor issues with the real built-in
diff to be more usable:

 - Omit ---/+++ header unless we emit diff output;

 - Detect and punt binary diff like GNU does;

 - Honor GIT_DIFF_OPTS minimally (only -u<number> and
   --unified=<number> are currently supported);

 - Omit line count of 1 from "@@ -l,k +m,n @@" hunk header
   (i.e. when k == 1 or n == 1)

 - Adjust testsuite for the lack of -p support.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agobuiltin-diff: \No newline at end of file.
Linus Torvalds [Sat, 25 Mar 2006 15:56:03 +0000 (07:56 -0800)]
builtin-diff: \No newline at end of file.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoUse a *real* built-in diff generator
Linus Torvalds [Sat, 25 Mar 2006 04:13:22 +0000 (20:13 -0800)]
Use a *real* built-in diff generator

This uses a simplified libxdiff setup to generate unified diffs _without_
doing  fork/execve of GNU "diff".

This has several huge advantages, for example:

Before:

[torvalds@g5 linux]$ time git diff v2.6.16.. > /dev/null

real    0m24.818s
user    0m13.332s
sys     0m8.664s

After:

[torvalds@g5 linux]$ time git diff v2.6.16.. > /dev/null

real    0m4.563s
user    0m2.944s
sys     0m1.580s

and the fact that this should be a lot more portable (ie we can ignore all
the issues with doing fork/execve under Windows).

Perhaps even more importantly, this allows us to do diffs without actually
ever writing out the git file contents to a temporary file (and without
any of the shell quoting issues on filenames etc etc).

NOTE! THIS PATCH DOES NOT DO THAT OPTIMIZATION YET! I was lazy, and the
current "diff-core" code actually will always write the temp-files,
because it used to be something that you simply had to do. So this current
one actually writes a temp-file like before, and then reads it into memory
again just to do the diff. Stupid.

But if this basic infrastructure is accepted, we can start switching over
diff-core to not write temp-files, which should speed things up even
further, especially when doing big tree-to-tree diffs.

Now, in the interest of full disclosure, I should also point out a few
downsides:

 - the libxdiff algorithm is different, and I bet GNU diff has gotten a
   lot more testing. And the thing is, generating a diff is not an exact
   science - you can get two different diffs (and you will), and they can
   both be perfectly valid. So it's not possible to "validate" the
   libxdiff output by just comparing it against GNU diff.

 - GNU diff does some nice eye-candy, like trying to figure out what the
   last function was, and adding that information to the "@@ .." line.
   libxdiff doesn't do that.

 - The libxdiff thing has some known deficiencies. In particular, it gets
   the "\No newline at end of file" case wrong. So this is currently for
   the experimental branch only. I hope Davide will help fix it.

That said, I think the huge performance advantage, and the fact that it
integrates better is definitely worth it. But it should go into a
development branch at least due to the missing newline issue.

Technical note: this is based on libxdiff-0.17, but I did some surgery to
get rid of the extraneous fat - stuff that git doesn't need, and seriously
cutting down on mmfile_t, which had much more capabilities than the diff
algorithm either needed or used. In this version, "mmfile_t" is just a
trivial <pointer,length> tuple.

That said, I tried to keep the differences to simple removals, so that you
can do a diff between this and the libxdiff origin, and you'll basically
see just things getting deleted. Even the mmfile_t simplifications are
left in a state where the diffs should be readable.

Apologies to Davide, whom I'd love to get feedback on this all from (I
wrote my own "fill_mmfile()" for the new simpler mmfile_t format: the old
complex format had a helper function for that, but I did my surgery with
the goal in mind that eventually we _should_ just do

mmfile_t mf;

buf = read_sha1_file(sha1, type, &size);
mf->ptr = buf;
mf->size = size;
.. use "mf" directly ..

which was really a nightmare with the old "helpful" mmfile_t, and really
is that easy with the new cut-down interfaces).

[ Btw, as any hawk-eye can see from the diff, this was actually generated
  with itself, so it is "self-hosting". That's about all the testing it
  has gotten, along with the above kernel diff, which eye-balls correctly,
  but shows the newline issue when you double-check it with "git-apply" ]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agotar-tree: Use the prefix field of a tar header
Rene Scharfe [Sat, 25 Mar 2006 22:21:07 +0000 (23:21 +0100)]
tar-tree: Use the prefix field of a tar header

... to store parts of the path, if possible.  This allows us to avoid
writing extended headers in certain cases (long pathes can only be
split at '/' chars).

Also adds a file to the test repo with a 100 chars long directory name.
Even old versions of tar that don't understand POSIX extended headers
should be able to handle this testcase.

Btw.: The longest path in the kernel tree currently has 70 chars.
Together with a 30 chars long prefix this would already cross the
field limit of 100 chars.

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agotar-tree: Remove obsolete code
Rene Scharfe [Sat, 25 Mar 2006 22:21:05 +0000 (23:21 +0100)]
tar-tree: Remove obsolete code

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agotar-tree: Use write_entry() to write the archive contents
Rene Scharfe [Sat, 25 Mar 2006 22:21:04 +0000 (23:21 +0100)]
tar-tree: Use write_entry() to write the archive contents

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agotar-tree: Introduce write_entry()
Rene Scharfe [Sat, 25 Mar 2006 22:21:03 +0000 (23:21 +0100)]
tar-tree: Introduce write_entry()

... and use it initially to write global extended header records.
Improvements compared to the old write_header():

  - Uses a struct ustar_header instead of hardcoded offsets.
  - Takes one struct strbuf as path argument instead of a (basedir,
    prefix, name) tuple.
  - Not only writes the tar header, but also the contents of the
    file, if any.
  - Does not write directly into the ring buffer.  This allows the
    code to be layed out more naturally, because there is no more
    ordering constraint.  Before we had to first finish writing the
    extended header, now we can construct the extended and normal
    headers in parallel.
  - The typeflag parameter has been replaced by (reasonable) magic
    values.  path == NULL indicates an extended header, additionally
    sha1 == NULL means it is a global extended header.

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agotar-tree: Use SHA1 of root tree for the basedir
Rene Scharfe [Sat, 25 Mar 2006 22:21:02 +0000 (23:21 +0100)]
tar-tree: Use SHA1 of root tree for the basedir

... instead of the made-up "0".

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogit-apply: safety fixes
Linus Torvalds [Sat, 25 Mar 2006 21:28:28 +0000 (13:28 -0800)]
git-apply: safety fixes

This was triggered by me testing the "@@" numbering shorthand by GNU
patch, which not only showed that git-apply thought it meant the number
was duplicated (when it means that the second number is 1), but my tests
showed than when git-apply mis-understood the number, it would then not
raise an alarm about it if the patch ended early.

Now, this doesn't actually _matter_, since with a three-line context, the
only case that "x,1" will be shorthanded to "x" is when x itself is 1 (in
which case git-apply got it right), but the fact that git-apply would also
silently accept truncated patches was a missed opportunity for additional
sanity-checking.

So make git-apply refuse to look at a patch fragment that ends early.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>