git.git
17 years agoFix "--pretty=format:" encoding item
Jeff King [Wed, 28 Mar 2007 21:09:05 +0000 (17:09 -0400)]
Fix "--pretty=format:" encoding item

It printed the header "encoding " instead of just showing
the encoding, as all other items do.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix "--pretty=format:" for parent related items.
Junio C Hamano [Wed, 28 Mar 2007 20:33:37 +0000 (13:33 -0700)]
Fix "--pretty=format:" for parent related items.

There are two breakages in the %P/%p interpolation.  It appended
an excess SP at the end of the list, and it gave uninitialized
contents of a buffer on the stack for root commits.

This fixes it, while updating the t6006 test which expected the
wrong output.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agohttp-fetch: remove path_len from struct alt_base, it was computed but never used
Gerrit Pape [Wed, 28 Mar 2007 09:47:35 +0000 (09:47 +0000)]
http-fetch: remove path_len from struct alt_base, it was computed but never used

Signed-off-by: Gerrit Pape <pape@smarden.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agohttp-fetch: don't use double-slash as directory separator in URLs
Gerrit Pape [Wed, 28 Mar 2007 09:46:15 +0000 (09:46 +0000)]
http-fetch: don't use double-slash as directory separator in URLs

Please see http://bugs.debian.org/409887

http-fetch expected the URL given at the command line to have a trailing
slash anyway, and then added '/objects...' when requesting objects files
from the http server.

Now it doesn't require the trailing slash in <url> anymore, and strips
trailing slashes if given nonetheless.

Signed-off-by: Gerrit Pape <pape@smarden.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAdd some basic tests of rev-list --pretty=format
Jeff King [Wed, 28 Mar 2007 00:08:28 +0000 (20:08 -0400)]
Add some basic tests of rev-list --pretty=format

These could stand to be a little more complex, but it should
at least catch obvious problems (like the recently fixed %ct
bug).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-upload-pack: make sure we close unused pipe ends
H. Peter Anvin [Tue, 27 Mar 2007 23:45:06 +0000 (16:45 -0700)]
git-upload-pack: make sure we close unused pipe ends

Right now, we don't close the read end of the pipe when git-upload-pack
runs git-pack-object, so we hang forever (why don't we get SIGALRM?)
instead of dying with SIGPIPE if the latter dies, which seems to be the
norm if the client disconnects.

Thanks to Johannes Schindelin <Johannes.Schindelin@gmx.de> for
pointing out where this close() needed to go.

This patch has been tested on kernel.org for several weeks and appear
to resolve the problem of git-upload-pack processes hanging around
forever.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years ago--pretty=format: fix broken %ct and %at interpolation
Jeff King [Tue, 27 Mar 2007 23:26:28 +0000 (19:26 -0400)]
--pretty=format: fix broken %ct and %at interpolation

A pointer arithmetic error in fill_person caused random data
from the commit object to be included with the timestamp,
which looked something like:

    $ git-rev-list --pretty=format:%ct origin/next | head
    commit 98453bdb3db10db26099749bc4f2dc029bed9aa9
    1174977948 -0700

    Merge branch 'master' into next

    * master:
      Bisect: Use
    commit c0ce981f5ebfd02463ff697b2fca52c7a54b0625
    1174889646 -0700

Signed-off-by: Jeff King <peff@peff.net>
Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agouse xrealloc in help.c
James Bowes [Tue, 27 Mar 2007 22:30:08 +0000 (18:30 -0400)]
use xrealloc in help.c

Signed-off-by: James Bowes <jbowes@dangerouslyinc.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoread-tree: use xcalloc
James Bowes [Tue, 27 Mar 2007 22:30:19 +0000 (18:30 -0400)]
read-tree: use xcalloc

Signed-off-by: James Bowes <jbowes@dangerouslyinc.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix "getaddrinfo()" buglet
Linus Torvalds [Tue, 27 Mar 2007 16:50:20 +0000 (09:50 -0700)]
Fix "getaddrinfo()" buglet

At least in Linux glibc, "getaddrinfo()" has a very irritating feature (or
bug, who knows..).

Namely if you pass it in an empty string for the service name, it will
happily and quietly consider it identical to a NULL port pointer, and
return port number zero and no errors. Which obviously will not work.

Maybe that's what it's really expected to do, although the man-page for
getaddrinfo() certainly implies that it's a bug.

So when somebody passes me a "please pull" request pointing to something
like the following

git://git.kernel.org:/pub/scm/linux/kernel/git/mchehab/v4l-dvb.git

(note the extraneous colon at the end of the host name), git would happily
try to connect to port 0, which would generally just cause the remote to
not even answer, and the "connect()" will take a long time to time out.

So to work around the glibc feature/bug, just notice this empty port case
automatically. Also, add the port information to the error information
when it fails to look up (maybe it's the host-name that fails, maybe it's
the port-name - we should print out both).

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMakefile: remove test-chmtime program in target clean.
Gerrit Pape [Tue, 27 Mar 2007 10:13:42 +0000 (10:13 +0000)]
Makefile: remove test-chmtime program in target clean.

While running 'make test', the test-chmtime program is created, and should
be cleaned up on 'make clean'.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Cleanup and uniquify die_error calls
Jakub Narebski [Tue, 27 Mar 2007 00:07:11 +0000 (02:07 +0200)]
gitweb: Cleanup and uniquify die_error calls

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agosha1_file.c (write_sha1_file): Detect close failure
Junio C Hamano [Tue, 27 Mar 2007 19:55:44 +0000 (12:55 -0700)]
sha1_file.c (write_sha1_file): Detect close failure

This is in the same spirit as earlier fix to write_sha1_from_fd().

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit.el: Display some information about the HEAD commit.
Alexandre Julliard [Mon, 26 Mar 2007 10:16:16 +0000 (12:16 +0200)]
git.el: Display some information about the HEAD commit.

Use git-log --pretty=oneline to print a short description of the
current HEAD (and merge heads if any) in the buffer header.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocument git-log --first-parent
Junio C Hamano [Tue, 27 Mar 2007 19:51:13 +0000 (12:51 -0700)]
Document git-log --first-parent

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoBisect: add checks at the beginning of "git bisect run".
Christian Couder [Tue, 27 Mar 2007 04:49:57 +0000 (06:49 +0200)]
Bisect: add checks at the beginning of "git bisect run".

We may be able to "run" with only one good revision given
and then verify that the result of the first run is bad.
And perhaps also the other way around.

But for now let's check that we have at least one bad and
one good revision before we start to run.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agosha1_file.c (write_sha1_from_fd): Detect close failure.
Jim Meyering [Mon, 26 Mar 2007 13:57:23 +0000 (15:57 +0200)]
sha1_file.c (write_sha1_from_fd): Detect close failure.

I stumbled across this in the context of the fchmod 0444 patch.
At first, I was going to unlink and call error like the two subsequent
tests do, but a failed write (above) provokes a "die", so I made
this do the same.  This is testing for a write failure, after all.

Signed-off-by: Jim Meyering <jim@meyering.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-rm: don't remove newly added file without -f
Jeff King [Mon, 26 Mar 2007 18:55:39 +0000 (14:55 -0400)]
git-rm: don't remove newly added file without -f

Given this set of commands:

  $ echo "newly added file" >new
  $ git add new
  $ git rm new

the file "new" was previously removed from the working
directory and the index. Because it was not in HEAD, it is
available only by searching for unreachable objects.

Instead, we now err on the safe side and refuse to remove
a file which is not referenced by HEAD.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoBisect: Use "git-show-ref --verify" when reseting.
Christian Couder [Mon, 26 Mar 2007 04:14:40 +0000 (06:14 +0200)]
Bisect: Use "git-show-ref --verify" when reseting.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Add example of config file and how to generate projects list to gitweb/INSTALL
Jakub Narebski [Mon, 26 Mar 2007 01:34:41 +0000 (02:34 +0100)]
gitweb: Add example of config file and how to generate projects list to gitweb/INSTALL

Add simple example of config file (turning on and allowing override of
a few %features). Also example config file and script to generate list
of projects in a format that can be used as GITWEB_LIST / $projects_list.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoGIT 1.5.1-rc2 v1.5.1-rc2
Junio C Hamano [Mon, 26 Mar 2007 01:00:45 +0000 (18:00 -0700)]
GIT 1.5.1-rc2

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: fix rel_path() when not connected to the repository root
Eric Wong [Sun, 25 Mar 2007 23:35:31 +0000 (16:35 -0700)]
git-svn: fix rel_path() when not connected to the repository root

This should fix fetching for people who did not use
"git svn --minimize" or cannot connect to the repository root
due to the lack of permissions.

I'm not sure what I was on when I made the change to the
rel_path() function in 4e9f6cc78e5d955bd0faffe76ae9aea6590189f1
that made it die() when we weren't connected to the repository
root :x

Thanks to Sven Verdoolaege for reporting this bug.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agouse xmalloc in git.c and help.c
James Bowes [Mon, 26 Mar 2007 00:39:36 +0000 (20:39 -0400)]
use xmalloc in git.c and help.c

Signed-off-by: James Bowes <jbowes@dangerouslyinc.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/fpl'
Junio C Hamano [Mon, 26 Mar 2007 00:47:07 +0000 (17:47 -0700)]
Merge branch 'jc/fpl'

* jc/fpl:
  git-log --first-parent: show only the first parent log

17 years agoUpdate README to point at a few key periodical messages to the list
Junio C Hamano [Mon, 26 Mar 2007 00:42:32 +0000 (17:42 -0700)]
Update README to point at a few key periodical messages to the list

They give a good starting point to new people who want to get
involved.  This owes suggestions by Martin Langhoff and Steven
Grimm.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint'
Junio C Hamano [Sun, 25 Mar 2007 22:08:11 +0000 (15:08 -0700)]
Merge branch 'maint'

* maint:
  user-manual: introduce "branch" and "branch head" differently
  glossary: clean up cross-references
  glossary: stop generating automatically
  user-manual: Use def_ instead of ref_ for glossary references.
  user-manual.txt: fix a tiny typo.
  user-manual: run xsltproc without --nonet option

17 years agoMerge branch 'maint' of git://linux-nfs.org/~bfields/git into maint
Junio C Hamano [Sun, 25 Mar 2007 22:07:27 +0000 (15:07 -0700)]
Merge branch 'maint' of git://linux-nfs.org/~bfields/git into maint

* 'maint' of git://linux-nfs.org/~bfields/git:
  user-manual: introduce "branch" and "branch head" differently
  glossary: clean up cross-references
  glossary: stop generating automatically
  user-manual: Use def_ instead of ref_ for glossary references.
  user-manual.txt: fix a tiny typo.
  user-manual: run xsltproc without --nonet option

17 years agoMerge branch 'js/remote-show-push'
Junio C Hamano [Sun, 25 Mar 2007 08:45:06 +0000 (01:45 -0700)]
Merge branch 'js/remote-show-push'

* js/remote-show-push:
  Teach git-remote to list pushed branches.

17 years agoMerge branch 'maint'
Junio C Hamano [Sun, 25 Mar 2007 07:21:40 +0000 (00:21 -0700)]
Merge branch 'maint'

* maint:
  gitweb: Add some installation notes in gitweb/INSTALL
  gitweb: Fix not marking signoff lines in "log" view
  gitweb: Don't escape attributes in CGI.pm HTML methods
  gitweb: Change to use explicitly function call cgi->escapHTML()

17 years agoUse diff* with --exit-code in git-am, git-rebase and git-merge-ours
Alex Riesen [Sun, 25 Mar 2007 00:56:13 +0000 (01:56 +0100)]
Use diff* with --exit-code in git-am, git-rebase and git-merge-ours

This simplifies the shell code, reduces its memory footprint, and
speeds things up. The performance improvements should be noticable
when git-rebase works on big commits.

Signed-off-by: Alex Riesen <raa.lkml@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocument --quiet option to git-diff
Alex Riesen [Sun, 25 Mar 2007 00:55:43 +0000 (01:55 +0100)]
Document --quiet option to git-diff

Signed-off-by: Alex Riesen <raa.lkml@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agowrite_sha1_from_fd() should make new objects read-only
Nicolas Pitre [Sat, 24 Mar 2007 16:02:27 +0000 (12:02 -0400)]
write_sha1_from_fd() should make new objects read-only

... like it is done everywhere else.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agomake it more obvious that temporary files are temporary files
Nicolas Pitre [Sat, 24 Mar 2007 15:58:22 +0000 (11:58 -0400)]
make it more obvious that temporary files are temporary files

When some operations are interrupted (or "die()'d" or crashed) then the
partial object/pack/index file may remain around.  Make it more obvious
in their name that those files are temporary stuff and can be cleaned up
if no operation is in progress.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoupdate-hook: remove e-mail sending hook.
Andy Parkins [Fri, 23 Mar 2007 10:21:59 +0000 (10:21 +0000)]
update-hook: remove e-mail sending hook.

The update hook's only job is to decide is a particular update
is allowed or not.  It was not the right place to send out
update notification e-mails from to begin with, as the final
stage of updating refs can fail after this hook runs.

Signed-off-by: Andy Parkins <andyparkins@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Add some installation notes in gitweb/INSTALL
Jakub Narebski [Sat, 24 Mar 2007 19:59:53 +0000 (20:59 +0100)]
gitweb: Add some installation notes in gitweb/INSTALL

Add some installation and configuration notes for gitweb in
gitweb/INSTALL. Make use of filling gitweb configuration by
Makefile.

It does not cover (yet?) all the configuration variables and
options.

Some of contents duplicates information in gitweb/README file
(it is referred from gitweb/INSTALL).

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Fix not marking signoff lines in "log" view
Jakub Narebski [Sat, 24 Mar 2007 19:59:13 +0000 (20:59 +0100)]
gitweb: Fix not marking signoff lines in "log" view

The CSS selector for signoff lines style was too strict: in the "log"
view the commit message is not encompassed in container "page_body"
div.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Don't escape attributes in CGI.pm HTML methods
Jakub Narebski [Wed, 7 Mar 2007 01:21:25 +0000 (02:21 +0100)]
gitweb: Don't escape attributes in CGI.pm HTML methods

There is no need to escape HTML tag's attributes in CGI.pm
HTML methods (like CGI::a()), because CGI.pm does attribute
escaping automatically.

  $cgi->a({ ... -attribute => atribute_value }, tag_contents)

is translated to

  <a ... attribute="attribute_value">tag_contents</a>

The rules for escaping attribute values (which are string contents) are
different. For example you have to take care about escaping embedded '"'
and "'" characters; CGI::a() does that for us automatically.

CGI::a() does not HTML escape tag_contents; we would need to write

  <a href="URL">some <b>bold</b> text</a>

for example. So we use esc_html (or esc_path) to escape tag_contents
as needed.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Change to use explicitly function call cgi->escapHTML()
Li Yang [Tue, 6 Mar 2007 03:58:56 +0000 (11:58 +0800)]
gitweb: Change to use explicitly function call cgi->escapHTML()

Change to use explicitly function call cgi->escapHTML().
This fix the problem on some systems that escapeHTML() is not
functioning, as default CGI is not setting 'escape' parameter.

Signed-off-by: Li Yang <leoli@freescale.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-am documentation: describe what is taken from where.
Junio C Hamano [Sat, 24 Mar 2007 10:08:54 +0000 (03:08 -0700)]
git-am documentation: describe what is taken from where.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-revert: Revert revert message to old behaviour
Johannes Schindelin [Fri, 23 Mar 2007 16:06:11 +0000 (17:06 +0100)]
git-revert: Revert revert message to old behaviour

When converting from the shell script, based on a misreading of the
sed invocation, the builtin included the abbreviated commit name,
and did _not_ include the quotes around the oneline message.

This fixes it.

[jc: with a fix for the typo/thinko spotted by Linus, and also
 removing the unwanted abbrev at the beginning.]

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint'
Junio C Hamano [Sat, 24 Mar 2007 06:29:37 +0000 (23:29 -0700)]
Merge branch 'maint'

* maint:
  gitweb: Fix "next" link in commit view

17 years agoDocumentation: bisect: make a comment fit better in the man page.
Christian Couder [Sat, 24 Mar 2007 05:32:05 +0000 (06:32 +0100)]
Documentation: bisect: make a comment fit better in the man page.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: bisect: add some titles to some paragraphs.
Christian Couder [Sat, 24 Mar 2007 05:30:33 +0000 (06:30 +0100)]
Documentation: bisect: add some titles to some paragraphs.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: bisect: reformat more paragraphs.
Christian Couder [Sat, 24 Mar 2007 05:31:49 +0000 (06:31 +0100)]
Documentation: bisect: reformat more paragraphs.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: bisect: reword one paragraph.
Christian Couder [Sat, 24 Mar 2007 05:31:16 +0000 (06:31 +0100)]
Documentation: bisect: reword one paragraph.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: bisect: reformat some paragraphs.
Christian Couder [Sat, 24 Mar 2007 05:29:58 +0000 (06:29 +0100)]
Documentation: bisect: reformat some paragraphs.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix path-limited "rev-list --bisect" termination condition.
Junio C Hamano [Fri, 23 Mar 2007 20:34:49 +0000 (13:34 -0700)]
Fix path-limited "rev-list --bisect" termination condition.

In a path-limited bisection, when the $bad commit is not
changing the limited path, and the number of suspects is 1, the
code miscounted and returned $bad from find_bisection(), which
is not marked with TREECHANGE.  This is of course filtered by
the output routine, resulting in an empty output, in turn
causing git-bisect driver to say "$bad was both good and bad".

Illustration.  Suppose you have these four commits, and only C
changes path P.  You know D is bad and A is good.

A---B---C*--D

git-bisect driver runs this to find a bisection point:

$ git rev-list --bisect A..D -- P

which calls find_bisection() with B, C and D.  The set of
commits that is given to this function is the same set of
commits as rev-list without --bisect option and pathspec
returns.  Among them, only C is marked with TREECHANGE.  Let's
call the set of commits given to find_bisection() that are
marked with TREECHANGE (or all of them if no path limiter is in
effect) "the bisect set".  In the above example, the size of the
bisect set is 1 (contains only "C").

For each commit in its input, find_bisection() computes the
number of commits it can reach in the bisect set.  For a commit
in the bisect set, this number includes itself, so the number is
1 or more.  This number is called "depth", and computed by
count_distance() function.

When you have a bisect set of N commits, and a commit has depth
D, how good is your bisection if you returned that commit?  How
good this bisection is can be measured by how many commits are
effectively tested "together" by testing one commit.

Currently you have (N-1) untested commits (the tip of the bisect
set, although it is included in the bisect set, is already known
to be bad).  If the commit with depth D turns out to be bad,
then your next bisect set will have D commits and you will have
(D-1) untested commits left, which means you tested (N-1)-(D-1)
= (N-D) commits with this bisection.  If it turns out to be good, then
your next bisect set will have (N-D) commits, and you will have
(N-D-1) untested commits left, which means you tested
(N-1)-(N-D-1) = D commits with this bisection.

Therefore, the goodness of this bisection is is min(N-D, D), and
find_bisection() function tries to find a commit that maximizes
this, by initializing "closest" variable to 0 and whenever a
commit with the goodness that is larger than the current
"closest" is found, that commit and its goodness are remembered
by updating "closest" variable.  The "the commit with the best
goodness so far" is kept in "best" variable, and is initialized
to a commit that happens to be at the beginning of the list of
commits given to this function (which may or may not be in the
bisect set when path-limit is in use).

However, when N is 1, then the sole tree-changing commit has
depth of 1, and min(N-D, D) evaluates to 0.  This is not larger
than the initial value of "closest", and the "so far the best
one" commit is never replaced in the loop.

When path-limit is not in use, this is not a problem, as any
commit in the input set is tree-changing.  But when path-limit
is in use, and when the starting "bad" commit does not change
the specified path, it is not correct to return it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Fix "next" link in commit view
Jakub Narebski [Fri, 23 Mar 2007 20:04:31 +0000 (21:04 +0100)]
gitweb: Fix "next" link in commit view

Fix copy'n'paste error in commit c9d193df which caused that "next"
link for merge commits in "commit" view
  (merge: _commit_ _commit_ ...)
was to "commitdiff" view instead of being to "commit" view.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-bisect.sh: properly dq $GIT_DIR
Junio C Hamano [Fri, 23 Mar 2007 20:24:22 +0000 (13:24 -0700)]
git-bisect.sh: properly dq $GIT_DIR

Otherwise you would be in trouble if your GIT_DIR has IFS letters in it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-bisect: typofix
Junio C Hamano [Fri, 23 Mar 2007 20:15:21 +0000 (13:15 -0700)]
git-bisect: typofix

The branch you are on while bisecting is always "bisect", and
checking for "refs/heads/bisect*" is wrong.  Only check if it is
exactly "refs/heads/bisect".

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocheckout: report where the new HEAD is upon detaching HEAD
Junio C Hamano [Fri, 23 Mar 2007 09:37:19 +0000 (02:37 -0700)]
checkout: report where the new HEAD is upon detaching HEAD

After "git reset" moves the HEAD around, it reports which commit
you are on, which gives the user a warm fuzzy feeling of
assurance.  Give the same assurance from git-checkout when
moving the detached HEAD around.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoBisect: implement "git bisect run <cmd>..." to automatically bisect.
Christian Couder [Fri, 23 Mar 2007 07:49:59 +0000 (08:49 +0100)]
Bisect: implement "git bisect run <cmd>..." to automatically bisect.

This idea was suggested by Bill Lear
(Message-ID: <17920.38942.364466.642979@lisa.zopyra.com>)
and I think it is a very good one.

This patch adds a new test file for "git bisect run", but there
is currently only one basic test.

Signed-off-by: Christian Couder <chriscool@tuxfamily.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoBisect: convert revs given to good and bad to commits
Uwe Kleine-König [Thu, 22 Mar 2007 16:18:08 +0000 (17:18 +0100)]
Bisect: convert revs given to good and bad to commits

Without this the rev could be (e.g.) a tag and then the condition to end the
bisect might fail and you have to check the already known to be bad revision
once more.

Signed-off-by: Uwe Kleine-König <ukleinek@informatik.uni-freiburg.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot4118: be nice to non-GNU sed
Johannes Schindelin [Fri, 23 Mar 2007 01:05:28 +0000 (21:05 -0400)]
t4118: be nice to non-GNU sed

Elias Pipping:
> I'm on a mac, hence /usr/bin/sed is not gnu sed, which makes
> t4118 fail.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Ack'd-by: Elias Pipping <pipping@macports.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-apply: Do not free the wrong buffer when we convert the data for writeout
Junio C Hamano [Fri, 23 Mar 2007 00:32:51 +0000 (17:32 -0700)]
git-apply: Do not free the wrong buffer when we convert the data for writeout

When we write out the result of patch application, we sometimes
need to munge the data (e.g. under core.autocrlf).  After doing
so, what we should free is the temporary buffer that holds the
converted data returned from convert_to_working_tree(), not the
original one.

This patch also moves the call to open() up in the function, as
the caller expects us to fail cheaply if leading directories
need to be created (and then the caller creates them and calls
us again).  For that calling pattern, attempting conversion
before opening the file adds unnecessary overhead.

Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge git://git2.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Thu, 22 Mar 2007 10:05:34 +0000 (03:05 -0700)]
Merge git://git2.kernel.org/pub/scm/gitk/gitk

* git://git2.kernel.org/pub/scm/gitk/gitk:
  [PATCH] prefer "git COMMAND" over "git-COMMAND" in gitk

17 years agoMerge branch 'maint'
Junio C Hamano [Thu, 22 Mar 2007 10:05:25 +0000 (03:05 -0700)]
Merge branch 'maint'

* maint:
  Documentation/pack-format.txt: Clear up description of types.
  fix typo in git-am manpage

17 years agoDocumentation/pack-format.txt: Clear up description of types.
Peter Eriksen [Wed, 21 Mar 2007 18:43:37 +0000 (19:43 +0100)]
Documentation/pack-format.txt: Clear up description of types.

Signed-off-by: Peter Eriksen <s022018@student.dtu.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoupdate HEAD reflog when branch pointed to by HEAD is directly modified
Nicolas Pitre [Wed, 21 Mar 2007 21:11:44 +0000 (17:11 -0400)]
update HEAD reflog when branch pointed to by HEAD is directly modified

The HEAD reflog is updated as well as the reflog for the branch pointed
to by HEAD whenever it is referenced with "HEAD".

There are some cases where a specific branch may be modified directly.
In those cases, the HEAD reflog should be updated as well if it is a
symref to that branch in order to be consistent.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoupdate-hook: abort early if the project description is unset
Andy Parkins [Tue, 20 Mar 2007 10:58:32 +0000 (10:58 +0000)]
update-hook: abort early if the project description is unset

It was annoying to always have the first email from a project be from
the "Unnamed repository; edit this file to name it for gitweb project";
just because it's so easy to forget to set it.

This patch checks to see if the description file is still default (or
empty) and aborts if so - allowing you to fix the problem before sending
out silly looking emails to every developer.

Signed-off-by: Andy Parkins <andyparkins@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-merge: Put FETCH_HEAD data in merge commit message
Michael S. Tsirkin [Thu, 22 Mar 2007 09:07:30 +0000 (11:07 +0200)]
git-merge: Put FETCH_HEAD data in merge commit message

This makes git-fetch <URL> && git-merge FETCH_HEAD produce the
same merge message as git-pull <URL>.

Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-rebase: make 'rebase HEAD branch' work as expected.
Junio C Hamano [Thu, 22 Mar 2007 09:54:59 +0000 (02:54 -0700)]
git-rebase: make 'rebase HEAD branch' work as expected.

When you want to amend the commit message of 3 commits before
the tip of the current branch, say 'master',

A--B--C--D--E(master)

it is sometimes handy to make your head detached at that commit
with:

$ git checkout HEAD~3 ;# check out B
$ git commit --amend ;# without modifying contents...

to create:

          .B'(HEAD)
         /
A--B--C--D--E(master)

and then rebase 'master' branch onto HEAD with this:

$ git rebase HEAD master

to result in:

          .B'-C'-D'-E(master=HEAD)
         /
A--B--C--D--E

However, the current code interprets HEAD after it switches to
the branch 'master', which means the rebase will not do
anything.  You have to say something unwieldly like this
instead:

$ git rebase $(git rev-parse HEAD) master

This fixes it by expanding the $onto commit name before
switching to the target branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agotree_entry_interesting(): allow it to say "everything is interesting"
Junio C Hamano [Thu, 22 Mar 2007 00:00:27 +0000 (17:00 -0700)]
tree_entry_interesting(): allow it to say "everything is interesting"

In addition to optimizing pathspecs that would never match,
which was done earlier, this optimizes pathspecs that would
always match (e.g. "arch/" while the traversal is already in
"arch/i386/" hierarchy).

This patch makes the worst case slightly more palatable, while
improving average case.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agotree-diff: avoid strncmp()
Junio C Hamano [Wed, 21 Mar 2007 19:34:46 +0000 (12:34 -0700)]
tree-diff: avoid strncmp()

If we already know that some of the pathspecs can match later
entries in the tree we are looking at, we do not have to do more
expensive strncmp() upfront before comparing the length of the
match pattern and the path, as a path longer than the match
pattern will not match it, and a path shorter than the match
pattern will match only if the path is a directory-component
wise prefix of the match pattern.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTeach tree_entry_interesting() that the tree entries are sorted.
Junio C Hamano [Wed, 21 Mar 2007 16:51:47 +0000 (09:51 -0700)]
Teach tree_entry_interesting() that the tree entries are sorted.

When we are looking at a tree entry with pathspecs, if all the
pathspecs sort strictly earlier than the entry we are currently
looking at, there is no way later entries in the same tree would
match our pathspecs, because the entries are sorted.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoSwitch over tree descriptors to contain a pre-parsed entry
Linus Torvalds [Wed, 21 Mar 2007 17:09:56 +0000 (10:09 -0700)]
Switch over tree descriptors to contain a pre-parsed entry

This makes the tree descriptor contain a "struct name_entry" as part of
it, and it gets filled in so that it always contains a valid entry. On
some benchmarks, it improves performance by up to 15%.

That makes tree entry "extract" trivial, and means that we only actually
need to decode each tree entry just once: we decode the first one when
we initialize the tree descriptor, and each subsequent one when doing
"update_tree_entry()".  In particular, this means that we don't need to
do strlen() both at extract time _and_ at update time.

Finally, it also allows more sharing of code (entry_extract(), that
wanted a "struct name_entry", just got totally trivial, along with the
"tree_entry()" function).

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoInitialize tree descriptors with a helper function rather than by hand.
Linus Torvalds [Wed, 21 Mar 2007 17:08:25 +0000 (10:08 -0700)]
Initialize tree descriptors with a helper function rather than by hand.

This removes slightly more lines than it adds, but the real reason for
doing this is that future optimizations will require more setup of the
tree descriptor, and so we want to do it in one place.

Also renamed the "desc.buf" field to "desc.buffer" just to trigger
compiler errors for old-style manual initializations, making sure I
didn't miss anything.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRemove "pathlen" from "struct name_entry"
Linus Torvalds [Wed, 21 Mar 2007 17:07:46 +0000 (10:07 -0700)]
Remove "pathlen" from "struct name_entry"

Since we have the "tree_entry_len()" helper function these days, and
don't need to do a full strlen(), there's no point in saving the path
length - it's just redundant information.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years ago[PATCH] prefer "git COMMAND" over "git-COMMAND" in gitk
Brandon Casey [Mon, 19 Mar 2007 23:00:37 +0000 (18:00 -0500)]
[PATCH] prefer "git COMMAND" over "git-COMMAND" in gitk

Preferring git _space_ COMMAND over git _dash_ COMMAND allows the
user to have only git and gitk in their path. e.g. when git and gitk
are symbolic links in a personal bin directory to the real git and gitk.

Signed-off-by: Paul Mackerras <paulus@samba.org>
17 years agofix typo in git-am manpage
Michael S. Tsirkin [Wed, 21 Mar 2007 08:37:36 +0000 (10:37 +0200)]
fix typo in git-am manpage

Fix typo in git-am manpage

Signed-off-by: Michael S. Tsirkin <mst@dev.mellanox.co.il>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoblame: cmp_suspect is not "cmp" anymore.
Junio C Hamano [Wed, 21 Mar 2007 06:37:51 +0000 (23:37 -0700)]
blame: cmp_suspect is not "cmp" anymore.

The earlier round makes the function return "is it different"
and it does not return a value suitable for sorting anymore.  Reverse
the logic to return "are they the same suspect" instead, and rename
it to "same_suspect()".

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agominor git-prune optimization
Nicolas Pitre [Wed, 21 Mar 2007 03:32:13 +0000 (23:32 -0400)]
minor git-prune optimization

Don't try to remove the containing directory for every pruned object but
try only once after the directory has been scanned instead.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoimprove checkout message when asking for same branch
Nicolas Pitre [Wed, 21 Mar 2007 02:26:47 +0000 (22:26 -0400)]
improve checkout message when asking for same branch

Change the feedback message if doing 'git checkout foo' when already on
branch "foo".

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoBe more careful about zlib return values
Linus Torvalds [Tue, 20 Mar 2007 18:38:34 +0000 (11:38 -0700)]
Be more careful about zlib return values

When creating a new object, we use "deflate(stream, Z_FINISH)" in a loop
until it no longer returns Z_OK, and then we do "deflateEnd()" to finish
up business.

That should all work, but the fact is, it's not how you're _supposed_ to
use the zlib return values properly:

 - deflate() should never return Z_OK in the first place, except if we
   need to increase the output buffer size (which we're not doing, and
   should never need to do, since we pre-allocated a buffer that is
   supposed to be able to hold the output in full). So the "while()" loop
   was incorrect: Z_OK doesn't actually mean "ok, continue", it means "ok,
   allocate more memory for me and continue"!

 - if we got an error return, we would consider it to be end-of-stream,
   but it could be some internal zlib error.  In short, we should check
   for Z_STREAM_END explicitly, since that's the only valid return value
   anyway for the Z_FINISH case.

 - we never checked deflateEnd() return codes at all.

Now, admittedly, none of these issues should ever happen, unless there is
some internal bug in zlib. So this patch should make zero difference, but
it seems to be the right thing to do.

We should probablybe anal and check the return value of "deflateInit()"
too!

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDon't ever return corrupt objects from "parse_object()"
Linus Torvalds [Tue, 20 Mar 2007 17:05:20 +0000 (10:05 -0700)]
Don't ever return corrupt objects from "parse_object()"

Looking at the SHA1 validation code due to the corruption that Alexander
Litvinov is seeing under Cygwin, I notice that one of the most central
places where we read objects, we actually do end up verifying the SHA1 of
the result, but then we happily parse it anyway.

And using "printf" to write the error message means that it not only can
get lost, but will actually mess up stdout, and cause other strange and
hard-to-debug failures downstream.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoindex-pack: more validation checks and cleanups
Nicolas Pitre [Tue, 20 Mar 2007 21:07:48 +0000 (17:07 -0400)]
index-pack: more validation checks and cleanups

When appending objects to a pack, make sure the appended data is really
what we expect instead of simply loading potentially corrupted objects
and legitimating them by computing a SHA1 of that corrupt data.

With this the sha1_object() can lose its test_for_collision parameter
which is now redundent.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoindex-pack: use hash_sha1_file()
Nicolas Pitre [Tue, 20 Mar 2007 20:02:09 +0000 (16:02 -0400)]
index-pack: use hash_sha1_file()

Use hash_sha1_file() instead of duplicating code to compute object SHA1.
While at it make it accept a const pointer.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agodon't ever allow SHA1 collisions to exist by fetching a pack
Nicolas Pitre [Tue, 20 Mar 2007 19:32:35 +0000 (15:32 -0400)]
don't ever allow SHA1 collisions to exist by fetching a pack

Waaaaaaay back Git was considered to be secure as it never overwrote
an object it already had.  This was ensured by always unpacking the
packfile received over the network (both in fetch and receive-pack)
and our already existing logic to not create a loose object for an
object we already have.

Lately however we keep "large-ish" packfiles on both fetch and push
by running them through index-pack instead of unpack-objects.  This
would let an attacker perform a birthday attack.

How?  Assume the attacker knows a SHA-1 that has two different
data streams.  He knows the client is likely to have the "good"
one.  So he sends the "evil" variant to the other end as part of
a "large-ish" packfile.  The recipient keeps that packfile, and
indexes it.  Now since this is a birthday attack there is a SHA-1
collision; two objects exist in the repository with the same SHA-1.
They have *very* different data streams.  One of them is "evil".

Currently the poor recipient cannot tell the two objects apart,
short of by examining the timestamp of the packfiles.  But lets
say the recipient repacks before he realizes he's been attacked.
We may wind up packing the "evil" version of the object, and deleting
the "good" one.  This is made *even more likely* by Junio's recent
rearrange_packed_git patch (b867092f).

It is extremely unlikely for a SHA1 collisions to occur, but if it
ever happens with a remote (hence untrusted) object we simply must
not let the fetch succeed.

Normally received packs should not contain objects we already have.
But when they do we must ensure duplicated objects with the same SHA1
actually contain the same data.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTeach git-remote to list pushed branches.
Johannes Sixt [Sun, 18 Mar 2007 20:34:46 +0000 (21:34 +0100)]
Teach git-remote to list pushed branches.

The configured refspecs are printed almost verbatim, i.e. both the local
and the remote branch name separated by a colon are printed; only the
prefix 'refs/heads/' is removed, like this:

  Local branch(es) pushed with 'git push'
    master refs/tags/*:refs/tags/* next:next

Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-fetch: Fix single_force in append_fetch_head
Santi Béjar [Sun, 18 Mar 2007 23:16:23 +0000 (00:16 +0100)]
git-fetch: Fix single_force in append_fetch_head

This fixes the single force (+) when fetched with fetch_per_ref.

Also use $LF as separator because IFS is $LF.

Signed-off-by: Santi Béjar <sbejar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge git://git2.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Tue, 20 Mar 2007 06:47:22 +0000 (23:47 -0700)]
Merge git://git2.kernel.org/pub/scm/gitk/gitk

* git://git2.kernel.org/pub/scm/gitk/gitk:
  [PATCH] gitk: bind <F5> key to Update (reread commits)

17 years agomake git clone -q suppress the noise with http fetch
Chris Wright [Tue, 20 Mar 2007 02:18:18 +0000 (19:18 -0700)]
make git clone -q suppress the noise with http fetch

We already have -q in git clone.  So for those who care to suppress
the noise during an http based clone, make -q actually do a quiet
http fetch.

Signed-off-by: Chris Wright <chrisw@sous-sol.org>
Cc: Fernando Herrera <fherrera@onirica.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix loose object uncompression check.
Linus Torvalds [Tue, 20 Mar 2007 05:49:53 +0000 (22:49 -0700)]
Fix loose object uncompression check.

The thing is, if the output buffer is empty, we should *still* actually
use the zlib routines to *unpack* that empty output buffer.

But we had a test that said "only unpack if we still expect more output".

So we wouldn't use up all the zlib stream, because we felt that we didn't
need it, because we already had all the bytes we wanted. And it was
"true": we did have all the output data. We just needed to also eat all
the input data!

We've had this bug before - thinking that we don't need to inflate()
anything because we already had it all..

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocontrib/continuous: a continuous integration build manager
Shawn O. Pearce [Tue, 20 Mar 2007 04:33:41 +0000 (00:33 -0400)]
contrib/continuous: a continuous integration build manager

This is a simple but powerful continuous integration build system
for Git.  It works by receiving push events from repositories
through the post-receive hook, aggregates them on a per-branch
basis into a first-come-first-serve build queue, and lets a
background build daemon perform builds one at a time.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoProvide some technical documentation for shallow clones
Johannes Schindelin [Tue, 20 Mar 2007 02:41:11 +0000 (03:41 +0100)]
Provide some technical documentation for shallow clones

There has not been any work on the shallow stuff lately, so it is hard
to find out what it does, and how. This document describes the ideas
as well as the current problems, and can serve as a starting point for
shallow people.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAdd a HOWTO for setting up a standalone git daemon
Johannes Schindelin [Tue, 20 Mar 2007 02:29:56 +0000 (03:29 +0100)]
Add a HOWTO for setting up a standalone git daemon

Setting up a git-daemon came up the other day on IRC, and it is slightly
non trivial for the uninitiated.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoxdiff/xutils.c(xdl_hash_record): factor out whitespace handling
Johannes Schindelin [Tue, 20 Mar 2007 03:05:10 +0000 (04:05 +0100)]
xdiff/xutils.c(xdl_hash_record): factor out whitespace handling

Since in at least one use case, xdl_hash_record() takes over 15% of the
CPU time, it makes sense to even micro-optimize it. For many cases, no
whitespace special handling is needed, and in these cases we should not
even bother to check for whitespace in _every_ iteration of the loop.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoblame: micro-optimize cmp_suspect()
Junio C Hamano [Tue, 20 Mar 2007 05:17:10 +0000 (22:17 -0700)]
blame: micro-optimize cmp_suspect()

The commit structures are guaranteed their uniqueness by the object
layer, so we can check their address and see if they are the same
without going down to the object sha1 level.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoReplace remaining instances of strdup with xstrdup.
James Bowes [Mon, 19 Mar 2007 21:42:40 +0000 (17:42 -0400)]
Replace remaining instances of strdup with xstrdup.

Signed-off-by: James Bowes <jbowes@dangerouslyinc.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agouse a LRU eviction policy for the delta base cache
Nicolas Pitre [Mon, 19 Mar 2007 20:31:04 +0000 (16:31 -0400)]
use a LRU eviction policy for the delta base cache

This provides a smoother degradation in performance when the cache
gets trashed due to the delta_base_cache_limit being reached.  Limited
testing with really small delta_base_cache_limit values appears to confirm
this.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoclean up the delta base cache size a bit
Nicolas Pitre [Mon, 19 Mar 2007 20:28:51 +0000 (16:28 -0400)]
clean up the delta base cache size a bit

Currently there are 3 different ways to deal with the cache size.
Let's stick to only one.  The compiler is smart enough to produce the exact
same code in those cases anyway.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoGIT 1.5.1-rc1 v1.5.1-rc1
Junio C Hamano [Mon, 19 Mar 2007 09:28:29 +0000 (02:28 -0700)]
GIT 1.5.1-rc1

I think we can start to slow down, as we now have covered
everything I listed earlier in the short-term release plan.

The last release 1.5.0 took painfully too long.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix merge-index
Junio C Hamano [Mon, 19 Mar 2007 09:48:37 +0000 (02:48 -0700)]
Fix merge-index

An earlier conversion to run_command() from execlp() forgot that
run_command() takes an array that is terminated with NULL.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoSet up for better tree diff optimizations
Linus Torvalds [Sun, 18 Mar 2007 22:18:30 +0000 (15:18 -0700)]
Set up for better tree diff optimizations

This is mainly just a cleanup patch, and sets up for later changes where
the tree-diff.c "interesting()" function can return more than just a
yes/no value.

In particular, it should be quite possible to say "no subsequent entries
in this tree can possibly be interesting any more", and thus allow the
callers to short-circuit the tree entirely.

In fact, changing the callers to do so is trivial, and is really all this
patch really does, because changing "interesting()" itself to say that
nothing further is going to be interesting is definitely more complicated,
considering that we may have arbitrary pathspecs.

But in cleaning up the callers, this actually fixes a potential small
performance issue in diff_tree(): if the second tree has a lot of
uninterestign crud in it, we would keep on doing the "is it interesting?"
check on the first tree for each uninteresting entry in the second one.

The answer is obviously not going to change, so that was just not helping.
The new code is clearer and simpler and avoids this issue entirely.

I also renamed "interesting()" to "tree_entry_interesting()", because I
got frustrated by the fact that

 - we actually had *another* function called "interesting()" in another
   file, and I couldn't tell from the profiles which one was the one that
   mattered more.

 - when rewriting it to return a ternary value, you can't just do

if (interesting(...))
...

   any more, but want to assign the return value to a local variable. The
   name of choice for that variable would normally be "interesting", so
   I just wanted to make the function name be more specific, and avoid
   that whole issue (even though I then didn't choose that name for either
   of the users, just to avoid confusion in the patch itself ;)

In other words, this doesn't really change anything, but I think it's a
good thing to do, and if somebody comes along and writes the logic for
"yeah, none of the pathspecs you have are interesting", we now support
that trivially.

It could easily be a meaningful optimization for things like "blame",
where there's just one pathspec, and stopping when you've seen it would
allow you to avoid about 50% of the tree traversals on average.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTrivial cleanup of track_tree_refs()
Linus Torvalds [Sun, 18 Mar 2007 20:38:19 +0000 (13:38 -0700)]
Trivial cleanup of track_tree_refs()

This makes "track_tree_refs()" use the same "tree_entry()" function for
counting the entries as it does for actually traversing them a few lines
later.

Not a biggie, but the reason I care was that this was the only user of
"update_tree_entry()" that didn't actually *extract* the tree entry first.
It doesn't matter as things stand now, but it meant that a separate
test-patch I had that avoided a few more "strlen()" calls by just saving
the entry length in the entry descriptor and using it directly when
updating wouldn't work without this patch.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit.el: Add support for commit hooks.
Alexandre Julliard [Sat, 17 Mar 2007 19:40:12 +0000 (20:40 +0100)]
git.el: Add support for commit hooks.

Run the pre-commit and post-commit hooks at appropriate places, and
display their output if any.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jb/gc'
Junio C Hamano [Mon, 19 Mar 2007 05:46:30 +0000 (22:46 -0700)]
Merge branch 'jb/gc'

* jb/gc:
  Make gc a builtin.

17 years agoMerge branch 'fl/cvsserver'
Junio C Hamano [Mon, 19 Mar 2007 05:44:25 +0000 (22:44 -0700)]
Merge branch 'fl/cvsserver'

* fl/cvsserver:
  cvsserver: further improve messages on commit and status
  cvsserver: Be more chatty

17 years agoLimit the size of the new delta_base_cache
Shawn O. Pearce [Mon, 19 Mar 2007 05:14:37 +0000 (01:14 -0400)]
Limit the size of the new delta_base_cache

The new configuration variable core.deltaBaseCacheLimit allows the
user to control how much memory they are willing to give to Git for
caching base objects of deltas.  This is not normally meant to be
a user tweakable knob; the "out of the box" settings are meant to
be suitable for almost all workloads.

We default to 16 MiB under the assumption that the cache is not
meant to consume all of the user's available memory, and that the
cache's main purpose was to cache trees, for faster path limiters
during revision traversal.  Since trees tend to be relatively small
objects, this relatively small limit should still allow a large
number of objects.

On the other hand we don't want the cache to start storing 200
different versions of a 200 MiB blob, as this could easily blow
the entire address space of a 32 bit process.

We evict OBJ_BLOB from the cache first (credit goes to Junio) as
we want to favor OBJ_TREE within the cache.  These are the objects
that have the highest inflate() startup penalty, as they tend to
be small and thus don't have that much of a chance to ammortize
that penalty over the entire data.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'sp/run-command'
Junio C Hamano [Mon, 19 Mar 2007 05:21:06 +0000 (22:21 -0700)]
Merge branch 'sp/run-command'

* sp/run-command:
  Use run_command within send-pack
  Use run_command within receive-pack to invoke index-pack
  Use run_command within merge-index
  Use run_command for proxy connections
  Use RUN_GIT_CMD to run push backends
  Correct new compiler warnings in builtin-revert
  Replace fork_with_pipe in bundle with run_command
  Teach run-command to redirect stdout to /dev/null
  Teach run-command about stdout redirection