Code

git.git
14 years agosmart-http: Don't deadlock on server failure
Shawn O. Pearce [Fri, 6 Aug 2010 21:19:44 +0000 (14:19 -0700)]
smart-http: Don't deadlock on server failure

If the remote HTTP server fails (e.g. returns 404 or 500) when we
posted the RPC to it, we won't have sent anything to the background
Git process that is supposed to handle the stream.  Because we
didn't send anything, its waiting for input from remote-curl, and
remote-curl cannot read its response payload because doing so would
lead to a deadlock.

Send the background task EOF on its input before we try to read
its response back, that way it will break out of its read loop
and terminate.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoAllow "check-ref-format --branch" from subdirectory
Jonathan Nieder [Fri, 6 Aug 2010 03:39:22 +0000 (22:39 -0500)]
Allow "check-ref-format --branch" from subdirectory

check-ref-format --branch requires access to the repository
to resolve refs like @{-1}.

Noticed by Nguyễn Thái Ngọc Duy.

Cc: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agocheck-ref-format: handle subcommands in separate functions
Jonathan Nieder [Fri, 6 Aug 2010 03:36:07 +0000 (22:36 -0500)]
check-ref-format: handle subcommands in separate functions

The code for each subcommand should be easier to read and manipulate
this way.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agomerge-recursive --renormalize
Jonathan Nieder [Thu, 5 Aug 2010 11:32:41 +0000 (06:32 -0500)]
merge-recursive --renormalize

Teach "git merge-recursive" a --renormalize option to enable the
merge.renormalize configuration.  The --no-renormalize option can
be used to override it in the negative.

So in the future, you might be able to, e.g.:

git checkout -m -Xrenormalize otherbranch

or

git revert -Xrenormalize otherpatch

or

git pull --rebase -Xrenormalize

The bad part: merge.renormalize is still not honored for most
commands.  And it reveals lots of places that -X has not been plumbed
in (so we get "git merge -Xrenormalize" but not much else).

NEEDSWORK: tests

Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agorerere: never renormalize
Jonathan Nieder [Thu, 5 Aug 2010 11:30:26 +0000 (06:30 -0500)]
rerere: never renormalize

plain rerere performs three tasks; let us consider how the new
merge.renormalize option should apply to each.

After an unsuccessful merge, rerere records conflict hunks from the
work tree under .git/rr-cache.  If the merge was performed with
merge.renormalize enabled, both sides of the conflict hunk use the
current work tree’s end-of-line and smudge rules; there is not really
much of a choice.

After a successful manual resolution, rerere records the postimage.
Here, also, the file will be in the current work tree’s canonical
format and there is not much to do about it.

When encountering that conflict again, merge looks up the preimage
and postimage using the conflict hunk as a key and runs a three-way
merge to apply that resolution to the work tree.  Since the conflict
hunk used the current work tree’s canonical format, chances are the
version in the work tree, the preimage, and the postimage will, too.
In fact using the merge.renormalize machinery is exactly the wrong
thing to do, since its result has been run through convert_to_git
and therefore is not suitable for writing to the work tree.

The only affected caller is "git merge".

NEEDSWORK: lacks test

Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agorerere: migrate to parse-options API
Jonathan Nieder [Thu, 5 Aug 2010 11:28:37 +0000 (06:28 -0500)]
rerere: migrate to parse-options API

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot4200 (rerere): modernize style
Jonathan Nieder [Thu, 5 Aug 2010 11:25:34 +0000 (06:25 -0500)]
t4200 (rerere): modernize style

Guard all test code with test_expect_success to make the
script easier to follow.  While at it, pick some other nits:

 - use test_tick (more than we have to, to be realistic);

 - 'single quotes' and \escaped HERE documents where possible
   simplify review for escaping problems;

 - omit whitespace after >redirection operators for
   consistency with other tests;

 - use "update-index --refresh" instead of testing that
   "ls-files -u" output is empty, since the former produces
   nicer output on failure;

 - compare to expected nonempty "ls-files -u" output instead
   of counting lines when it is expected to be nonempty.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoll-merge: let caller decide whether to renormalize
Jonathan Nieder [Thu, 5 Aug 2010 11:24:58 +0000 (06:24 -0500)]
ll-merge: let caller decide whether to renormalize

Add a “renormalize” bit to the ll-merge options word so callers can
decide on a case-by-case basis whether the merge is likely to have
overlapped with a change in smudge/clean rules.

This reveals a few commands that have not been taking that situation
into account, though it does not fix them.

No functional change intended.

Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Improved-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoll-merge: make flag easier to populate
Jonathan Nieder [Thu, 5 Aug 2010 11:17:38 +0000 (06:17 -0500)]
ll-merge: make flag easier to populate

ll_merge() takes its options in a flag word, which has a few
advantages:

 - options flags can be cheaply passed around in registers, while
   an option struct passed by pointer cannot;

 - callers can easily pass 0 without trouble for no options,
   while an option struct passed by value would not allow that.

The downside is that code to populate and access the flag word can be
somewhat opaque.  Mitigate that with a few macros.

Cc: Avery Pennarun <apenwarr@gmail.com>
Cc: Bert Wesarg <bert.wesarg@googlemail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocumentation/technical: document ll_merge
Jonathan Nieder [Thu, 5 Aug 2010 11:16:51 +0000 (06:16 -0500)]
Documentation/technical: document ll_merge

Cc: Junio C Hamano <gitster@pobox.com>
Cc: Avery Pennarun <apenwarr@gmail.com>
Cc: Bert Wesarg <bert.wesarg@googlemail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agomerge-trees: let caller decide whether to renormalize
Jonathan Nieder [Thu, 5 Aug 2010 11:15:32 +0000 (06:15 -0500)]
merge-trees: let caller decide whether to renormalize

Add a "renormalize" option to struct merge_options so callers can
decide on a case-by-case basis whether the merge is likely to have
overlapped with a change in smudge/clean rules.  The option defaults
to the global merge_renormalize setting for now.

No change in behavior intended.

Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agomerge-trees: push choice to renormalize away from low level
Jonathan Nieder [Thu, 5 Aug 2010 11:13:49 +0000 (06:13 -0500)]
merge-trees: push choice to renormalize away from low level

The merge machinery decides whether to resmudge and clean relevant
entries based on the global merge_renormalize setting, which is set by
"git merge" based on its configuration (and left alone by other
commands).

A nicer interface would make that decision a parameter to merge_trees
so callers would pass in a choice made on a call-by-call basis.
Start by making blob_unchanged stop examining the merge_renormalize
global.

In other words, this change is a trivial no-op, but it brings us
closer to something good.

Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot6038 (merge.renormalize): check that it can be turned off
Jonathan Nieder [Thu, 5 Aug 2010 11:13:04 +0000 (06:13 -0500)]
t6038 (merge.renormalize): check that it can be turned off

An unusual sort of person (not me) may even enjoy the conflicts
from line-ending changes.  But more importantly, it is useful to
document that behavior so we can more easily notice if it changes
in an uncontrolled way while no one is watching.

Cc: Eyvind Bernhardsen <eyvind.bernhardsen@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot6038 (merge.renormalize): try checkout -m and cherry-pick
Jonathan Nieder [Thu, 5 Aug 2010 11:11:12 +0000 (06:11 -0500)]
t6038 (merge.renormalize): try checkout -m and cherry-pick

checkout -m and cherry-pick have not been wired up to respect
merge.renormalize, but a naïve user would not know that.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot6038 (merge.renormalize): style nitpicks
Jonathan Nieder [Thu, 5 Aug 2010 11:09:33 +0000 (06:09 -0500)]
t6038 (merge.renormalize): style nitpicks

Some tweaks to simplify adding and running tests.

 - Use test_tick for predictable, sort of realistic commit dates;

 - Use test_cmp as "test_cmp expected actual" --- some crazy
   content that was not expected should cause the test to fail;

 - Remove and re-add all files at the start of each test so the
   worktree is easier to think about;

 - Avoid using cat where not necessary for clarity.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agolog: parse separate option for --glob
Matthieu Moy [Thu, 5 Aug 2010 08:22:56 +0000 (10:22 +0200)]
log: parse separate option for --glob

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agolog: parse separate options like git log --grep foo
Matthieu Moy [Thu, 5 Aug 2010 08:22:55 +0000 (10:22 +0200)]
log: parse separate options like git log --grep foo

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agodiff: parse separate options --stat-width n, --stat-name-width n
Matthieu Moy [Thu, 5 Aug 2010 08:22:54 +0000 (10:22 +0200)]
diff: parse separate options --stat-width n, --stat-name-width n

Part of a campaign for unstuck forms of options.

[jn: with some refactoring]

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agodiff: split off a function for --stat-* option parsing
Jonathan Nieder [Thu, 5 Aug 2010 08:22:53 +0000 (10:22 +0200)]
diff: split off a function for --stat-* option parsing

As an optimization, the diff_opt_parse() switchboard has
a single case for all the --stat-* options.  Split it
off into a separate function so we can enhance it
without bringing code dangerously close to the right
margin.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agodiff: parse separate options like -S foo
Matthieu Moy [Thu, 5 Aug 2010 08:22:52 +0000 (10:22 +0200)]
diff: parse separate options like -S foo

Change the option parsing logic in revision.c to accept separate forms
like `-S foo' in addition to `-Sfoo'. The rest of git already accepted
this form, but revision.c still used its own option parsing.

Short options affected are -S<string>, -l<num> and -O<orderfile>, for
which an empty string wouldn't make sense, hence -<option> <arg> isn't
ambiguous.

This patch does not handle --stat-name-width and --stat-width, which are
special-cases where diff_long_opt do not apply. They are handled in a
separate patch to ease review.

Original patch by Matthieu Moy, plus refactoring by Jonathan Nieder.

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMerge git://git.bogomips.org/git-svn
Junio C Hamano [Fri, 6 Aug 2010 16:06:20 +0000 (09:06 -0700)]
Merge git://git.bogomips.org/git-svn

* git://git.bogomips.org/git-svn:
  git svn: URL-decode left-hand side of svn refspec

14 years agopretty-options.txt: match --format's documentation with implementation.
Matthieu Moy [Thu, 5 Aug 2010 13:08:05 +0000 (15:08 +0200)]
pretty-options.txt: match --format's documentation with implementation.

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoinstaweb: add access+error logging for WEBrick
Eric Wong [Thu, 5 Aug 2010 08:46:04 +0000 (08:46 +0000)]
instaweb: add access+error logging for WEBrick

This allows WEBrick to support all the logging functionality
in a manner consistent with the other web servers.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
14 years agoinstaweb: minimize moving parts for WEBrick
Eric Wong [Thu, 5 Aug 2010 08:35:45 +0000 (08:35 +0000)]
instaweb: minimize moving parts for WEBrick

Since there are WEBrick configuration settings (including the
upcoming AccessLog support) that cannot be represented in YAML
and require Ruby anyways, the YAML config file is an unnecessary
layer of complexity.

Additionally, the shell script wrapper to start WEBrick is
unecessary since our generated Ruby script can be made
executable in the same manner with /usr/bin/env.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
14 years agoinstaweb: fix WEBrick server support
Eric Wong [Wed, 4 Aug 2010 09:51:25 +0000 (09:51 +0000)]
instaweb: fix WEBrick server support

This has been broken since commit be5347b ("httpd logs in a
"$httpd_only" subdirectory").

Since WEBrick has no other way of preserving environment
variables needed for gitweb, we create a shell script wrapper
that sets the environment variables as our CGI interpreter
to run gitweb.cgi.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
14 years agogit svn: URL-decode left-hand side of svn refspec
Steven Walter [Tue, 3 Aug 2010 23:21:25 +0000 (19:21 -0400)]
git svn: URL-decode left-hand side of svn refspec

This change allows git-svn to handle an URL with colons in the path

[ew: rewritten to use uri_decode() function]

Signed-off-by: Eric Wong <normalperson@yhbt.net>
14 years agoconfig: add --local option
Sverre Rabbelier [Wed, 4 Aug 2010 01:59:23 +0000 (20:59 -0500)]
config: add --local option

This is a shorthand similar to --system but instead uses
the config file of the current repository.

Signed-off-by: Sverre Rabbelier <srabbelier@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agosetup: split off a function to handle ordinary .git directories
Nguyễn Thái Ngọc Duy [Sat, 24 Jul 2010 12:11:58 +0000 (07:11 -0500)]
setup: split off a function to handle ordinary .git directories

Finish the clean-up of setup_git_directory_gently() by splitting the
last case of validation+setup (global variables, prefix, check_format,
set_git_dir) into its own function.  Now setup_git_git_directory_gently
itself takes care of discovery only and the functions that pick up
from there are nearby in the source file so they can be easily
compared.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoRevert "rehabilitate 'git index-pack' inside the object store"
Nguyễn Thái Ngọc Duy [Sat, 24 Jul 2010 11:30:49 +0000 (06:30 -0500)]
Revert "rehabilitate 'git index-pack' inside the object store"

Now setup_git_directory_gently behaves sanely even from subdirs of
.git, so simplify index-pack by no longer protecting against that.

This reverts commit a672ea6ac5a1b876bc7adfe6534b16fa2a32c94b
(excluding tests).

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agosetup: do not forget working dir from subdir of gitdir
Nguyễn Thái Ngọc Duy [Sat, 24 Jul 2010 11:29:41 +0000 (06:29 -0500)]
setup: do not forget working dir from subdir of gitdir

v1.6.1.3~4^2 (Fix gitdir detection when in subdir of gitdir,
2009-01-16) did not go far enough: when a git directory is
an ancestor of the original working directory, not only
should GIT_DIR be set to point to the .git directory, but
the original working directory should be restored before
carrying out the relevant command.

This way, the effect of running a git command from a subdir
of .git will be the same whether or not GIT_DIR is explicitly
set.

Noticed while investigating v1.6.0.3~1 (rehabilitate 'git
index-pack' inside the object store, 2008-10-20).

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot4111 (apply): refresh index before applying patches to it
Jonathan Nieder [Fri, 30 Jul 2010 18:31:22 +0000 (13:31 -0500)]
t4111 (apply): refresh index before applying patches to it

"git apply", like most plumbing, does not automatically refresh the
index file even if it is only stat-dirty.  So unless the two "cp"
commands in reset_preimage() for a given file happen to have the same
time stamp, there will be a spurious

error: sub/dir/file: does not match index

Refresh the index to eliminate this timing dependency.  Noticed by
running the test with --valgrind (which slows things down a lot).

Reported-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMerge branch 'maint'
Junio C Hamano [Tue, 3 Aug 2010 22:20:01 +0000 (15:20 -0700)]
Merge branch 'maint'

* maint:
  contrib/svn-fe: Add the svn-fe target to .gitignore
  contrib/svn-fe: Fix IncludePath
  Fix DIFF_QUEUE_CLEAR refactoring
  git-gui: fix size and position of window panes on startup
  git-gui: mc cannot be used before msgcat has been loaded
  git-gui: use textconv filter for diff and blame
  git-gui: Avoid using the <<Copy>> binding as a menu accelerator on win32
  git-gui: fix shortcut creation on cygwin
  git-gui: fix PATH environment for mingw development environment
  git-gui: fix usage of _gitworktree when creating shortcut for windows
  git-gui: fix "Explore Working Copy" for Windows again
  git-gui: fix usage of themed widgets variable
  git-gui: Handle failure of core.worktree to identify the working directory.
  git-gui: check whether systems nice command works or disable it

14 years agoMerge branch 'gp/pack-refs-remove-empty-dirs'
Junio C Hamano [Tue, 3 Aug 2010 22:19:02 +0000 (15:19 -0700)]
Merge branch 'gp/pack-refs-remove-empty-dirs'

* gp/pack-refs-remove-empty-dirs:
  pack-refs: remove newly empty directories

14 years agoMerge branch 'sv/maint-diff-q-clear-fix' into maint
Junio C Hamano [Tue, 3 Aug 2010 22:17:34 +0000 (15:17 -0700)]
Merge branch 'sv/maint-diff-q-clear-fix' into maint

* sv/maint-diff-q-clear-fix:
  Fix DIFF_QUEUE_CLEAR refactoring

14 years agoMerge branch 'rr/svn-fe' into maint
Junio C Hamano [Tue, 3 Aug 2010 22:17:11 +0000 (15:17 -0700)]
Merge branch 'rr/svn-fe' into maint

* rr/svn-fe:
  contrib/svn-fe: Add the svn-fe target to .gitignore
  contrib/svn-fe: Fix IncludePath

14 years agoMerge branch 'pt/git-gui' into maint
Junio C Hamano [Tue, 3 Aug 2010 22:16:38 +0000 (15:16 -0700)]
Merge branch 'pt/git-gui' into maint

* pt/git-gui:
  git-gui: fix size and position of window panes on startup
  git-gui: mc cannot be used before msgcat has been loaded
  git-gui: use textconv filter for diff and blame
  git-gui: Avoid using the <<Copy>> binding as a menu accelerator on win32
  git-gui: fix shortcut creation on cygwin
  git-gui: fix PATH environment for mingw development environment
  git-gui: fix usage of _gitworktree when creating shortcut for windows
  git-gui: fix "Explore Working Copy" for Windows again
  git-gui: fix usage of themed widgets variable
  git-gui: Handle failure of core.worktree to identify the working directory.
  git-gui: check whether systems nice command works or disable it

14 years agorebase: support -X to pass through strategy options
Mike Lundy [Wed, 28 Jul 2010 22:04:29 +0000 (00:04 +0200)]
rebase: support -X to pass through strategy options

git-rebase calls out to merge strategies, but did not support merge
strategy options so far.  Add this, in the same style used in
git-merge.

Sadly we have to do the full quoting/eval dance here, since
merge-recursive supports the --subtree=<path> option which potentially
contains whitespace.

This patch does not cover git rebase -i, which does not call any merge
strategy directly except in --preserve-merges, and even then only for
merges.

[jc: with a trivial fix-up for 'expr']

Signed-off-by: Mike Lundy <mike@fluffypenguin.org>
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agocontrib/svn-fe: Add the svn-fe target to .gitignore
Ramkumar Ramachandra [Thu, 29 Jul 2010 05:36:20 +0000 (11:06 +0530)]
contrib/svn-fe: Add the svn-fe target to .gitignore

Add the svn-fe target to .gitignroe.

Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agocontrib/svn-fe: Fix IncludePath
Ramkumar Ramachandra [Thu, 29 Jul 2010 05:36:19 +0000 (11:06 +0530)]
contrib/svn-fe: Fix IncludePath

Include the path "../../vcs-svn" while compiling it in the Makefile
and change svn-fe.c to include svndump.h.

Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocumentation: flesh out “git pull” description
Jonathan Nieder [Mon, 2 Aug 2010 21:39:30 +0000 (16:39 -0500)]
Documentation: flesh out “git pull” description

The current description in the pull man page does not say much more
than that “git pull” is fetch + merge.  Though that is all a person
needs to know in the end, it would be useful to summarize a bit about
what those commands do for new readers.

Most of this description is taken from the “git merge” docs.

Now that we explain how to back out of a failed merge (reset --merge),
we can tone down the warning against that a bit.

Except, as Thomas noticed, there’s a risk with that because people
might read this version of the manpage online and then conclude that
it is safe to try a merge with uncommitted changes, only to find that
their “git reset” doesn't support --merge yet.  Or worse, verify that
their git-reset has --merge by a quick test (1b5b465 is in 1.6.2) but
then find that it does not help with backing out of a merge (e11d7b5
is only in 1.7.0!).  So keep the warning.

With clarifications from Ævar, Thomas, and Junio.

Noticed-by: Geoff Russell <geoffrey.russell@gmail.com>
Cc: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Cc: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMerge git://git.bogomips.org/git-svn
Junio C Hamano [Mon, 2 Aug 2010 22:38:55 +0000 (15:38 -0700)]
Merge git://git.bogomips.org/git-svn

* git://git.bogomips.org/git-svn:
  git svn: fix dcommit to work with touched files
  git svn: add an option to recode pathnames

14 years agoupload-pack: Improve error message when bad ref requested
Elijah Newren [Sat, 31 Jul 2010 20:11:46 +0000 (14:11 -0600)]
upload-pack: Improve error message when bad ref requested

When printing an error message saying a ref was requested that we do not
have, only print that ref, rather than the ref and everything sent to us
on the same packet line (e.g. protocol support specifications).

Signed-off-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocument git-instaweb start/stop/restart
Jared Hance [Mon, 2 Aug 2010 18:20:38 +0000 (14:20 -0400)]
Document git-instaweb start/stop/restart

The flags --start, --stop, and --restart can be used without the "--".
Document this feature.

Signed-off-by: Jared Hance <jaredhance@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agogitweb: Fix typo in run() subroutine
Jakub Narebski [Mon, 2 Aug 2010 20:21:47 +0000 (22:21 +0200)]
gitweb: Fix typo in run() subroutine

Run $post_dispatch_hook->() not $pre_dispatch_hook->() after each
request.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agosha1_name.c: fix parsing of ":/token" syntax
Junio C Hamano [Mon, 2 Aug 2010 21:37:06 +0000 (14:37 -0700)]
sha1_name.c: fix parsing of ":/token" syntax

The parser tried to clean up the object flags it used while finding
commits with matching string, but was not doing a very good job at it.

This caused "checkout -b new ':/token'", which internally tries to parse
':/token' twice as an object name, to fail when the commit in question
was reachable from only one ref.

The mask bits given to pop_most_recent_commit(&list, MASK) means "I have
already been on the list to be processed, so please do not place me again
even if I am found to be a parent of some other commit on the list."  So
mark them when we add them to the list at the beginning.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agogit svn: fix dcommit to work with touched files
David D. Kilzer [Mon, 2 Aug 2010 19:58:19 +0000 (12:58 -0700)]
git svn: fix dcommit to work with touched files

The dcommit command fails if an otherwise unmodified file has
been touched in the working directory:

    Cannot dcommit with a dirty index.  Commit your changes
    first, or stash them with `git stash'.

This happens because "git diff-index" reports a difference
between the index and the filesystem:

    :100644 100644 d00491...... 000000...... M      file

The fix is to run "git update-index --refresh" before
"git diff-index" as is done in git-rebase and
git-rebase--interactive before "git diff-files".

This changes dcommit to display a list of modified files before
exiting.

Also add a similar test case for "git svn rebase".

[ew: rearranged commit message subject]

Signed-off-by: David D. Kilzer <ddkilzer@kilzer.net>
Acked-by: Eric Wong <normalperson@yhbt.net>
14 years agols-files: learn a debugging dump format
Thomas Rast [Fri, 30 Jul 2010 22:35:59 +0000 (00:35 +0200)]
ls-files: learn a debugging dump format

Teach git-ls-files a new option --debug that just tacks all available
data from the cache onto each file's line.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMerge branch 'maint'
Junio C Hamano [Mon, 2 Aug 2010 19:02:16 +0000 (12:02 -0700)]
Merge branch 'maint'

* maint:
  test-lib: Remove 3 year old no-op --no-python option
  test-lib: Ignore --quiet under a TAP harness
  Documentation/rev-parse: quoting is required with --parseopt
  Documentation: reporting bugs
  Fix git rebase --continue to work with touched files
  Document ls-files -t as semi-obsolete.

14 years agoMerge branch 'ab/tap' into maint
Junio C Hamano [Mon, 2 Aug 2010 18:53:58 +0000 (11:53 -0700)]
Merge branch 'ab/tap' into maint

* ab/tap:
  test-lib: Remove 3 year old no-op --no-python option
  test-lib: Ignore --quiet under a TAP harness

14 years agotest-lib: Remove 3 year old no-op --no-python option
Ævar Arnfjörð Bjarmason [Sat, 31 Jul 2010 16:49:53 +0000 (16:49 +0000)]
test-lib: Remove 3 year old no-op --no-python option

The --no-python option was added to test-lib.sh by Johannes Schindelin
in early 2006 in abb7c7b3. It was later turned into a no-op by Junio C
Hamano in 7cdbff14 the same year.

Over three years is long enough before removing this old wart which
was retained for backwards compatibility. Our tests have been using
NO_PYTHON and "test_have_prereq PYTHON" for a long time now.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agotest-lib: Ignore --quiet under a TAP harness
Ævar Arnfjörð Bjarmason [Sat, 31 Jul 2010 16:40:05 +0000 (16:40 +0000)]
test-lib: Ignore --quiet under a TAP harness

Running the tests with --quiet under a TAP harness will always fail,
since a TAP harness always needs actual test output to go along with
the plan that's being emitted.

Change the test-lib.sh to ignore the --quiet option under
HARNESS_ACTIVE to work around this. Then users that have --quiet in
their GIT_TEST_OPTS can run tests under prove(1) without everything
breaking.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agobuiltin/push.c: remove useless temporary variable
Jared Hance [Sat, 31 Jul 2010 12:54:55 +0000 (08:54 -0400)]
builtin/push.c: remove useless temporary variable

Creating a variable nr here to use throughout the function only to change
refspec_nr to nr at the end, having not used refspec_nr the entire time,
is rather pointless. Instead, simply increment refspec_nr.

While at it, use ALLOC_GROW() instead of xrealloc().

Signed-off-by: Jared Hance <jaredhance@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agogitweb: allow configurations that change with each request
Jonathan Nieder [Sat, 31 Jul 2010 03:01:59 +0000 (22:01 -0500)]
gitweb: allow configurations that change with each request

gitolite's contrib/gitweb/gitweb.conf includes:

$ENV{GL_USER} = $cgi->remote_user || "gitweb";

which is useful for setups where a user has to be authenticated
to access certain repos.  Perhaps other typical configurations
change per session in other ways, too.

v1.7.2-rc2~6 (gitweb: Move evaluate_gitweb_config out of run_request,
2010-07-05) broke such configurations for a speedup, by loading
the configuration once per FastCGI process.

Probably in the end there should be a way to specify in the
configuration whether a particular installation wants the speedup or
the flexibility.  But for now it is easier to just undo the relevant
change.

This partially reverts commit 869d58813b24c74e84c9388041eafcef40cb51e4.

Reported-by: Julio Lajara <julio.lajara@alum.rpi.edu>
Analysis-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot1011 (sparse checkout): style nitpicks
Jonathan Nieder [Sat, 31 Jul 2010 06:14:25 +0000 (13:14 +0700)]
t1011 (sparse checkout): style nitpicks

Tweak the rest of the script to more closely follow the test
style guide.  Guarding setup commands with test_expect_success
makes it easy to see the scope in which some particular data is
used; removal of whitespace after >redirection operators is just
for consistency.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMerge git://repo.or.cz/git-gui into pt/git-gui
Junio C Hamano [Mon, 2 Aug 2010 16:28:30 +0000 (09:28 -0700)]
Merge git://repo.or.cz/git-gui into pt/git-gui

* git://repo.or.cz/git-gui:
  git-gui: fix size and position of window panes on startup
  git-gui: mc cannot be used before msgcat has been loaded
  git-gui: use textconv filter for diff and blame
  git-gui: Avoid using the <<Copy>> binding as a menu accelerator on win32
  git-gui: fix shortcut creation on cygwin
  git-gui: fix PATH environment for mingw development environment
  git-gui: fix usage of _gitworktree when creating shortcut for windows
  git-gui: fix "Explore Working Copy" for Windows again
  git-gui: fix usage of themed widgets variable
  git-gui: Handle failure of core.worktree to identify the working directory.
  git-gui: check whether systems nice command works or disable it

14 years agoDocumentation/rev-parse: quoting is required with --parseopt
Thomas Rast [Fri, 30 Jul 2010 15:01:50 +0000 (17:01 +0200)]
Documentation/rev-parse: quoting is required with --parseopt

When calling rev-parse --parseopt, as in the (now fixed) documented
example

  eval "$(echo "$OPTS_SPEC" | git rev-parse --parseopt -- "$@" || echo exit $?)"

the outermost quoting is required, as otherwise all runs of arbitrary
whitespace inside the resulting 'set -- ...' call would be collapsed
into a single space.

This was exposed as a result of our new use of cat <<\EOF since
47e9cd2 (parseopt: wrap rev-parse --parseopt usage for eval
consumption, 2010-06-12), but has always been a problem when handling
arguments containing e.g. newlines.

Point this out in the documentation, and in particular correct the
example that did not have the quotes.

Noticed-by: Joshua Jensen <jjensen@workspacewhiz.com>
Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocumentation: reporting bugs
Junio C Hamano [Mon, 2 Aug 2010 16:07:39 +0000 (09:07 -0700)]
Documentation: reporting bugs

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoFix DIFF_QUEUE_CLEAR refactoring
Junio C Hamano [Mon, 2 Aug 2010 15:29:42 +0000 (08:29 -0700)]
Fix DIFF_QUEUE_CLEAR refactoring

It introduced a macro to reduce repeated assignments to three fields,
but an unrelated and incorrect change snuck in by mistake, which broke
commands like "git diff-files -p --submodule".

Noticed by Sven Verdoolaege.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agogit-gui: fix size and position of window panes on startup
Pat Thoyts [Mon, 2 Aug 2010 12:42:45 +0000 (13:42 +0100)]
git-gui: fix size and position of window panes on startup

The themed panedwindow needs to have the sash position set after the
widget has been mapped therefore apply this setting in the Map event
binding. To avoid visible redraws as the application is constructed
the main window should be withdrawn until all the widgets have been added

Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: mc cannot be used before msgcat has been loaded
Pat Thoyts [Mon, 2 Aug 2010 11:13:05 +0000 (12:13 +0100)]
git-gui: mc cannot be used before msgcat has been loaded

If someone attempts to use an older version that Tk 8.4 the error was
masked by the lack of a mc command.

Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: use textconv filter for diff and blame
Clément Poulain [Fri, 30 Jul 2010 08:11:02 +0000 (09:11 +0100)]
git-gui: use textconv filter for diff and blame

Create a checkbox "Use Textconv For Diffs and Blame" in git-gui options.
If checked and if the driver for the concerned file exists, git-gui calls diff
and blame with --textconv option

Signed-off-by: Clément Poulain <clement.poulain@ensimag.imag.fr>
Signed-off-by: Diane Gasselin <diane.gasselin@ensimag.imag.fr>
Signed-off-by: Axel Bonnet <axel.bonnet@ensimag.imag.fr>
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: Avoid using the <<Copy>> binding as a menu accelerator on win32
Pat Thoyts [Mon, 8 Mar 2010 12:43:27 +0000 (12:43 +0000)]
git-gui: Avoid using the <<Copy>> binding as a menu accelerator on win32

On Windows the Control-C binding is used to copy and is mapped to the Tk
virtual event <<Copy>>. In the initial git-gui dialog this is also bound
as an accelerator for the Clone menu item. The effect is that both bindings
run, copying the text but resetting the clone page or switching to the clone
page when the user tries to copy text from one of the entry fields.
This patch avoids this by using Control-L instead for Windows only.

Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: fix shortcut creation on cygwin
Heiko Voigt [Sat, 27 Feb 2010 20:48:48 +0000 (21:48 +0100)]
git-gui: fix shortcut creation on cygwin

When the user tried to create a desktop icon with git gui on cygwin
wscript was complaining about an unknown option and displaying the
non-native path as such.

Signed-off-by: Heiko Voigt <hvoigt@hvoigt.net>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: fix PATH environment for mingw development environment
Heiko Voigt [Sat, 27 Feb 2010 20:47:42 +0000 (21:47 +0100)]
git-gui: fix PATH environment for mingw development environment

When creating a desktop shortcut from the gui the shortcut directly
starts wish with the git-gui script. In the msysgit development
environment some dll's reside in the mingw/bin directory which causes
that git can not start because libiconv2.dll is not found.

When using such a link the error is even more cryptic stating:
"child killed: unknown signal"

Signed-off-by: Heiko Voigt <hvoigt@hvoigt.net>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: fix usage of _gitworktree when creating shortcut for windows
Heiko Voigt [Sat, 27 Feb 2010 20:45:51 +0000 (21:45 +0100)]
git-gui: fix usage of _gitworktree when creating shortcut for windows

This fixes msysGit issue 425.

Signed-off-by: Heiko Voigt <heiko.voigt@mahr.de>
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit-gui: fix "Explore Working Copy" for Windows again
Markus Heidelberg [Thu, 25 Feb 2010 00:14:22 +0000 (01:14 +0100)]
git-gui: fix "Explore Working Copy" for Windows again

It has already been fixed in commit 454efb47 (git-gui (Win): make
"Explore Working Copy" more robust, 2009-04-01), but has been broken in
commit 21985a11 (git-gui: handle non-standard worktree locations,
2010-01-23) by accidentally replacing too much with a new variable.

The problem can be reproduced when starting git-gui from within a
subdirectory. The solution is to convert the path name, explorer.exe is
invoked with, to a platform native name.

Signed-off-by: Markus Heidelberg <markus.heidelberg@web.de>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
14 years agogit svn: add an option to recode pathnames
Dmitry Statyvka [Fri, 30 Jul 2010 02:30:13 +0000 (04:30 +0200)]
git svn: add an option to recode pathnames

Introduce a new option 'svn.pathnameencoding' that instructs git svn to
recode pathnames to a given encoding.  It can be used by windows users
and by those who work in non-utf8 locales to avoid corrupted file names
with non-ascii characters.

[rp: renamed the option and added manpage documentation]

Signed-off-by: Dmitry Statyvka <dstatyvka@tmsoft-ltd.kiev.ua>
Signed-off-by: Robert Pollak <robert.pollak@gmail.com>
Acked-by: Eric Wong <normalperson@yhbt.net>
14 years agogitweb: move highlight config out of guess_file_syntax()
Alejandro R. Sedeño [Wed, 28 Jul 2010 18:40:53 +0000 (14:40 -0400)]
gitweb: move highlight config out of guess_file_syntax()

Move highlight config out of guess_file_syntax() so that it can be
extended/overridden by system/user configuration.

Signed-off-by: Alejandro R. Sedeño <asedeno@mit.edu>
Acked-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoxsize_t: check whether we lose bits
Thomas Rast [Wed, 28 Jul 2010 16:36:31 +0000 (18:36 +0200)]
xsize_t: check whether we lose bits

Attempting to mmap (via git-add or similar) a file larger than 4GB on
32-bit Linux systems results in a repository that has only the file
modulo 4GB stored, because of truncation of the off_t file size to a
size_t for mmap.

When xsize_t was introduced to handle this truncation in dc49cd7 (Cast
64 bit off_t to 32 bit size_t, 2007-03-06), Shawn even pointed out
that it should detect when such a cutoff happens.

Make it so.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoFix git rebase --continue to work with touched files
David D. Kilzer [Wed, 28 Jul 2010 08:20:16 +0000 (01:20 -0700)]
Fix git rebase --continue to work with touched files

When performing a non-interactive rebase, sometimes
"git rebase --continue" will fail if an unmodified file is
touched in the working directory:

    You must edit all merge conflicts and then
    mark them as resolved using git add

This is caused by "git diff-files" reporting a difference
between the index and the filesystem:

    :100644 100644 d00491...... 000000...... M file

The fix is to run "git update-index --refresh" before
"git diff-files" as is done in git-rebase--interactive.

Signed-off-by: David D. Kilzer <ddkilzer@kilzer.net>
Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocument ls-files -t as semi-obsolete.
Matthieu Moy [Wed, 28 Jul 2010 07:24:27 +0000 (09:24 +0200)]
Document ls-files -t as semi-obsolete.

The behavior of "git ls-files -t" is very misleading (see
http://thread.gmane.org/gmane.comp.version-control.git/126516 and
http://thread.gmane.org/gmane.comp.version-control.git/144394/focus=144397
for examples of mislead users) and badly documented, hence we point the
users to superior alternatives.

The feature is marked as "semi-obsolete" but not "scheduled for removal"
since it's a plumbing command, scripts might use it, and Git testsuite
already uses it to test the state of the index.

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoSubmittingPatches: Cite the 50 char subject limit
Ævar Arnfjörð Bjarmason [Wed, 28 Jul 2010 14:11:25 +0000 (14:11 +0000)]
SubmittingPatches: Cite the 50 char subject limit

Change the SubmittingPatches recommendations to mention the 50
character soft limit on patch subject lines. 50 characters is the soft
limit mentioned in git-commit(1) and gittutorial(7), it's also the
point at which Gitweb, GitHub and various other Git front ends start
abbreviating the commit message.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoSubmittingPatches: Clarify the Signed-off-by rules
Ævar Arnfjörð Bjarmason [Wed, 28 Jul 2010 14:12:59 +0000 (14:12 +0000)]
SubmittingPatches: Clarify the Signed-off-by rules

The wording of the Signed-off-by rules could be read as stating that
S-O-B should only be added when the submitter considered the patch
ready for inclusion in git.git.

We also want Signed-off-by to be used for e.g. RFC patches, in case
someone wants to dig an old patch out of the archive and improve
it. Change the wording to recommend a Signed-off-by for all submitted
patches.

The problem with the wording came up in the "[PATCH/RFC] Hacky version
of a glob() driven config include" thread[1]. Bert Wesarg suggested[2]
that it be removed to avoid confusion, which this change implements.

  1. <1273180440-8641-1-git-send-email-avarab@gmail.com>
  2. <AANLkTimziTKL13VKIOcaS1TX1F_xvTVjH8Q398Yx36Us@mail.gmail.com>

Suggested-by: Bert Wesarg <bert.wesarg@googlemail.com>
Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoupload-pack: remove unused "create_full_pack" code in do_rev_list
Nguyễn Thái Ngọc Duy [Wed, 28 Jul 2010 09:39:10 +0000 (16:39 +0700)]
upload-pack: remove unused "create_full_pack" code in do_rev_list

A bit of history in chronological order, the newest at bottom:

80ccaa7 (upload-pack: Move the revision walker into a separate function.)
   do_rev_list was introduced with create_full_pack argument

21edd3f (upload-pack: Run rev-list in an asynchronous function.)
   do_rev_list was now called by start_async, create_full_pack was
   passed by rev_list.data

f0cea83 (Shift object enumeration out of upload-pack)
   rev_list.data was now zero permanently. Creating full pack was
   done by passing --all to pack-objects

ae6a560 (run-command: support custom fd-set in async)
   rev_list.data = 0 was found out redudant and got rid of.

Get rid of the code as well, for less headache while reading do_rev_list.

[jc: noticed by Elijah Newren]

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoSync with 1.7.2.1
Junio C Hamano [Tue, 27 Jul 2010 23:54:06 +0000 (16:54 -0700)]
Sync with 1.7.2.1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoGit 1.7.2.1 v1.7.2.1
Junio C Hamano [Tue, 27 Jul 2010 23:42:47 +0000 (16:42 -0700)]
Git 1.7.2.1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoSync with 1.7.1.2
Junio C Hamano [Tue, 27 Jul 2010 23:40:15 +0000 (16:40 -0700)]
Sync with 1.7.1.2

14 years agoGit 1.7.1.2 v1.7.1.2
Junio C Hamano [Tue, 27 Jul 2010 22:04:49 +0000 (15:04 -0700)]
Git 1.7.1.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoSync with 1.7.0 series
Junio C Hamano [Tue, 27 Jul 2010 22:01:36 +0000 (15:01 -0700)]
Sync with 1.7.0 series

14 years agoGit 1.7.0.7 v1.7.0.7
Junio C Hamano [Tue, 27 Jul 2010 21:07:41 +0000 (14:07 -0700)]
Git 1.7.0.7

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoexamples/commit: use --abbrev for commit summary
Jonathan Nieder [Tue, 27 Jul 2010 18:44:51 +0000 (13:44 -0500)]
examples/commit: use --abbrev for commit summary

After v1.7.1.1~17^2~3 (pretty: Respect --abbrev option, 2010-05-03),
plumbing users do not abbreviate %h hashes by default any more.

Noticed while investigating the bug fixed by v1.7.1.1~17^2
(commit::print_summary(): don't use format_commit_message(),
2010-06-12).

Cc: Will Palmer <wmpalmer@gmail.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agocheckout, commit: remove confusing assignments to rev.abbrev
Jonathan Nieder [Tue, 27 Jul 2010 18:37:07 +0000 (13:37 -0500)]
checkout, commit: remove confusing assignments to rev.abbrev

Since they do not precede setup_revisions, these assignments of 0 to
rev.abbrev have no effect.

v1.7.1.1~17^2~3 (2010-05-03) taught the log --format=%h machinery
to respect --abbrev instead of always abbreviating, so we have to pay
attention to the abbrev setting now.

Helped-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoarchive: abbreviate substituted commit ids again
Jonathan Nieder [Tue, 27 Jul 2010 18:32:36 +0000 (13:32 -0500)]
archive: abbreviate substituted commit ids again

Given a file with:

  (define archive-id "$Format:%ct|%h|a$")

and an export-subst attribute, the "%h" results in an full 40-digit
object name instead of the expected 7-digit one.

The export-subst feature requests unabbreviated object names because
that is the low-level default.  The effect was not observable until
v1.7.1.1~17^2~3 (2010-05-03), which taught log --format=%h to respect
the --abbrev option.

Reported-by: Eli Barzilay <eli@barzilay.org>
Tested-by: Eli Barzilay <eli@barzilay.org>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot/t6035-merge-dir-to-symlink.sh: Remove TODO on passing test
Ævar Arnfjörð Bjarmason [Tue, 27 Jul 2010 16:54:58 +0000 (16:54 +0000)]
t/t6035-merge-dir-to-symlink.sh: Remove TODO on passing test

Various merge-recursive cases were fixed in "merge-recursive: Fix D/F
conflicts" by Elijah Newren. Some tests were changed from
test_expect_failure to test_expect_success, but one fell through the
cracks.

Change that test to use test_expect_success.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoconfig --get --path: check for unset $HOME
Jonathan Nieder [Mon, 26 Jul 2010 15:06:51 +0000 (10:06 -0500)]
config --get --path: check for unset $HOME

If $HOME is unset (as in some automated build situations),
currently

git config --path path.home "~"
git config --path --get path.home

segfaults.  Error out with

Failed to expand user dir in: '~/'

instead.

Reported-by: Julien Cristau <jcristau@debian.org>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agocommit: remove full stop from usage help for -u
Stephen Boyd [Tue, 27 Jul 2010 05:40:32 +0000 (22:40 -0700)]
commit: remove full stop from usage help for -u

From api-parse-options.txt:

  `description` is a short string to describe the effect of the option.
  It shall begin with a lower-case letter and a full stop (`.`) shall be
  omitted at the end.

It also makes it less confusing if the argument is 'no.' or 'no'.

Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoClarify help message when no remote is specified in fetch/pull.
Matthieu Moy [Mon, 26 Jul 2010 16:32:09 +0000 (18:32 +0200)]
Clarify help message when no remote is specified in fetch/pull.

The message is especially confusing when "git fetch" is ran from "git
pull", for users not aware of "git fetch". The new message makes it clear
that "fetch" means "fetch new revisions", and gives hint on the solution.

We don't add a advice.* configuration option since this message doesn't
appear in normal use, and shouldn't disturb advanced users.

Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMakefile: add check-docs exception for gitrevisions
Thomas Rast [Mon, 26 Jul 2010 08:40:58 +0000 (10:40 +0200)]
Makefile: add check-docs exception for gitrevisions

The manpage was added in 1ed6f2c (Documentation: gitrevisions,
2010-07-05), but since it does not have a corresponding git command,
it needs an exception for check-docs.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Acked-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoFix 'git' wrapper usage string
Thomas Rast [Mon, 26 Jul 2010 08:36:24 +0000 (10:36 +0200)]
Fix 'git' wrapper usage string

8b1fa77 (Allow passing of configuration parameters in the command
line, 2010-03-26) forgot the closing ']' for the -c option.

While we're there, also rewrap.  Instead of folding the last two lines
together, try to highlight that COMMAND is required by starting a line
with it.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agotests: Ignore the Test::Harness .prove file
Ævar Arnfjörð Bjarmason [Fri, 23 Jul 2010 22:58:44 +0000 (22:58 +0000)]
tests: Ignore the Test::Harness .prove file

We document how to run prove with the --state option in t/README. This
produces a .prove YAML file in the current directory. Change the t/
gitignore to ignore it, and clean it up on `make clean'.

Signed-off-by: Ævar Arnfjörð Bjarmason <avarab@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocumentation/git-push: Explain status output in more detail
Thomas Rast [Sat, 24 Jul 2010 16:03:15 +0000 (18:03 +0200)]
Documentation/git-push: Explain status output in more detail

Mention the effects of the receive.deny* family of options for the
"remote rejected" case.  While there, also split up the explanation
into an easier-to-parse list format.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoDocument receive.denyDeleteCurrent
Thomas Rast [Sat, 24 Jul 2010 16:07:53 +0000 (18:07 +0200)]
Document receive.denyDeleteCurrent

This option was introduced by 747ca24 (receive-pack:
receive.denyDeleteCurrent, 2009-02-08) but never documented.

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoCast execl*() NULL sentinels to (char *)
Thomas Rast [Sat, 24 Jul 2010 15:20:23 +0000 (17:20 +0200)]
Cast execl*() NULL sentinels to (char *)

The NULL sentinel argument to the execl*() family of calls must be
cast to (char *), as otherwise:

- platforms where NULL is just 0 (not (void *)) would pass an int

- (admittedly esoteric) platforms where NULL is (void *)0 and (void *)
  and (char *) have different memory layouts would pass the wrong kind
  of pointer

Signed-off-by: Thomas Rast <trast@student.ethz.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agogit-read-tree.txt: acknowledge the directory matching bug in sparse checkout
Nguyễn Thái Ngọc Duy [Sun, 25 Jul 2010 09:30:32 +0000 (16:30 +0700)]
git-read-tree.txt: acknowledge the directory matching bug in sparse checkout

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoINSTALL: configure /etc/xml/catalog to build docs on Cygwin
Jon Seymour [Sun, 25 Jul 2010 03:57:35 +0000 (13:57 +1000)]
INSTALL: configure /etc/xml/catalog to build docs on Cygwin

Without additional configuration steps, the documentation build on Cygwin
fails because the XML catalog is missing required rewrites for certain
docbook resources.

This patch documents the required configuration.

Signed-off-by: Jon Seymour <jon.seymour@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agot3700-add: fix dependence on stdout and stderr buffering
Johannes Sixt [Thu, 22 Jul 2010 08:13:33 +0000 (10:13 +0200)]
t3700-add: fix dependence on stdout and stderr buffering

One test case checked the stdout and stderr of 'git add' by constructing a
single 'expect' file that contained both streams. But when the command
runs, the order of stdout and stderr output is unpredictable because it
depends on how the streams are buffered. At least on Windows, the buffering
is different from what the test case expected. Hence, check the two output
texts separately.

Signed-off-by: Johannes Sixt <j6t@kdbg.org>
Acked-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agogit-rebase--interactive.sh: use printf instead of echo to print commit message
Brandon Casey [Thu, 22 Jul 2010 19:15:11 +0000 (14:15 -0500)]
git-rebase--interactive.sh: use printf instead of echo to print commit message

On systems with an echo which defaults to the XSI-conformant behavior
(Solaris, or others using Ksh), echo will interpret certain backslashed
characters as control sequences.  This can cause a problem for interactive
rebase when it is used to rebase commits whose commit "subject" (the first
line) contains any of these backslashed sequences.  In this case, echo will
substitute the control sequence for the backslashed characters and either
the rebased commit message will differ from the original, or the rebase
process will fail.  Neither is desirable.

So work around this issue by replacing the echo statements used to print
out portions of the commit message, with printf.

Also, add a test to test for this breakage.

Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil>
Acked-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoAdd a google-chrome option for web--browse
Nathan W. Panike [Thu, 22 Jul 2010 16:58:20 +0000 (11:58 -0500)]
Add a google-chrome option for web--browse

On some systems, the chrome browser is named google-chrome.  We add
support for this case.

Signed-off-by: Nathan W. Panike <nathan.panike@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
14 years agoMerge branch 'maint-1.7.1' into maint
Junio C Hamano [Mon, 26 Jul 2010 06:13:41 +0000 (23:13 -0700)]
Merge branch 'maint-1.7.1' into maint

* maint-1.7.1:
  request-pull.txt: Document -p option
  Check size of path buffer before writing into it

14 years agoMakefile: make hard/symbolic links for non-builtins too
Brandon Casey [Fri, 23 Jul 2010 17:50:45 +0000 (12:50 -0500)]
Makefile: make hard/symbolic links for non-builtins too

To conserve space/improve file caching we try to make hard or symbolic links
from each builtin program to the main git executable rather than having
each be a complete duplicate copy of it.  We weren't doing this for the
non-builtin programs though.  So, just because we can, and because it's
easy, and for completeness sake, let's do it.

Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>