11 years agoMerge branch 'tr/maint-1.6.3-add-p-modeonly-fix' into maint-1.6.3
Junio C Hamano [Wed, 26 Aug 2009 18:22:00 +0000 (11:22 -0700)]
Merge branch 'tr/maint-1.6.3-add-p-modeonly-fix' into maint-1.6.3

* tr/maint-1.6.3-add-p-modeonly-fix:
  add -p: do not attempt to coalesce mode changes
  git add -p: demonstrate failure when staging both mode and hunk

11 years agoadd -p: do not attempt to coalesce mode changes
Thomas Rast [Sat, 15 Aug 2009 13:56:39 +0000 (15:56 +0200)]
add -p: do not attempt to coalesce mode changes

In 0392513 (add-interactive: refactor mode hunk handling, 2009-04-16),
we merged the interaction loops for mode changes and hunk staging.
This was fine at the time, because 0beee4c (git-add--interactive:
remove hunk coalescing, 2008-07-02) removed hunk coalescing.

However, in 7a26e65 (Revert "git-add--interactive: remove hunk
coalescing", 2009-05-16), we resurrected it.  Since then, the code
would attempt in vain to merge mode changes with diff hunks,
corrupting both in the process.

We add a check to the coalescing loop to ensure it only looks at diff
hunks, thus skipping mode changes.

Noticed-by: Kirill Smelkov <>
Signed-off-by: Thomas Rast <>
Signed-off-by: Junio C Hamano <>
11 years agogit add -p: demonstrate failure when staging both mode and hunk
Kirill Smelkov [Sat, 15 Aug 2009 12:26:49 +0000 (16:26 +0400)]
git add -p: demonstrate failure when staging both mode and hunk

When trying to stage changes to file which has also pending `chmod +x`,
`git add -p` produces lots of 'Use of uninitialized value ...' warnings
and fails to do the job:

    $ echo content >> file
    $ chmod +x file
    $ git add -p
    diff --git a/file b/file
    index e69de29..d95f3ad
    --- a/file
    +++ b/file
    old mode 100644
    new mode 100755
    Stage mode change [y,n,q,a,d,/,j,J,g,?]? y
    @@ -0,0 +1 @@
    Stage this hunk [y,n,q,a,d,/,K,g,e,?]? y
    Use of uninitialized value $o_ofs in addition (+) at .../git-add--interactive line 776.
    Use of uninitialized value $ofs in numeric le (<=) at .../git-add--interactive line 806.
    Use of uninitialized value $o0_ofs in concatenation (.) or string at .../git-add--interactive line 830.
    Use of uninitialized value $n0_ofs in concatenation (.) or string at .../git-add--interactive line 830.
    Use of uninitialized value $o_ofs in addition (+) at .../git-add--interactive line 776.
    fatal: corrupt patch at line 5
    diff --git a/file b/file
    index e69de29..d95f3ad
    --- a/file
    +++ b/file
    @@ -,0 + @@

Signed-off-by: Kirill Smelkov <>
Signed-off-by: Junio C Hamano <>
11 years agoChange mentions of "git programs" to "git commands"
Ori Avtalion [Fri, 7 Aug 2009 14:24:21 +0000 (17:24 +0300)]
Change mentions of "git programs" to "git commands"

Most of the docs and printouts refer to "commands" when discussing what
the end users call via the "git" top-level program. We should refer them
as "git programs" when we discuss the fact that the commands are
implemented as separate programs, but in other contexts, it is better to
use the term "git commands" consistently.

Signed-off-by: Ori Avtalion <>
Signed-off-by: Nanako Shiraishi <>
Signed-off-by: Junio C Hamano <>
11 years agoDocumentation: merge: one <remote> is required
Paul Bolle [Tue, 11 Aug 2009 13:03:58 +0000 (15:03 +0200)]
Documentation: merge: one <remote> is required

merge only requires one <remote>, so "<remote>..." should be used in the
synopsis (and not "<remote> <remote>...").

Signed-off-by: Paul Bolle <>
Signed-off-by: Junio C Hamano <>
11 years agohelp.c: give correct structure's size to memset()
Johan Herland [Tue, 11 Aug 2009 10:10:21 +0000 (12:10 +0200)]
help.c: give correct structure's size to memset()

These two structures are of the same type, but we'd better be consistent.

Signed-off-by: Johan Herland <>
Signed-off-by: Junio C Hamano <>
11 years agoFix typos in git-remote.txt and git-symbolic-ref.txt
Štěpán Němec [Tue, 11 Aug 2009 00:52:07 +0000 (02:52 +0200)]
Fix typos in git-remote.txt and git-symbolic-ref.txt

Signed-off-by: Štěpán Němec <>
Signed-off-by: Junio C Hamano <>
11 years agoMerge branch 'maint-1.6.2' into maint-1.6.3
Junio C Hamano [Sat, 8 Aug 2009 03:44:15 +0000 (20:44 -0700)]
Merge branch 'maint-1.6.2' into maint-1.6.3

* maint-1.6.2:
  verify-pack -v: do not report "chain length 0"
  t5510: harden the way verify-pack is used

11 years agoMerge branch 'maint-1.6.1' into maint-1.6.2
Junio C Hamano [Sat, 8 Aug 2009 03:44:09 +0000 (20:44 -0700)]
Merge branch 'maint-1.6.1' into maint-1.6.2

* maint-1.6.1:
  verify-pack -v: do not report "chain length 0"
  t5510: harden the way verify-pack is used

11 years agoMerge branch 'maint-1.6.0' into maint-1.6.1
Junio C Hamano [Sat, 8 Aug 2009 03:44:02 +0000 (20:44 -0700)]
Merge branch 'maint-1.6.0' into maint-1.6.1

* maint-1.6.0:
  verify-pack -v: do not report "chain length 0"
  t5510: harden the way verify-pack is used

11 years agoverify-pack -v: do not report "chain length 0"
Junio C Hamano [Fri, 7 Aug 2009 22:36:31 +0000 (15:36 -0700)]
verify-pack -v: do not report "chain length 0"

When making a histogram of delta chain length in the pack, the program
collects number of objects whose delta depth exceeds the MAX_CHAIN limit
in histogram[0], and showed it as the number of items that exceeds the
limit correctly.  HOWEVER, it also showed the same number labeled as
"chain length = 0".

In fact, we are not showing the number of objects whose chain length is
zero, i.e. the base objects.  Correct this.

Signed-off-by: Junio C Hamano <>
11 years agot5510: harden the way verify-pack is used
Junio C Hamano [Sat, 8 Aug 2009 03:12:13 +0000 (20:12 -0700)]
t5510: harden the way verify-pack is used

The test ignored the exit status from verify pack command, and also relied
on not seeing any delta chain statistics.

Signed-off-by: Junio C Hamano <>
11 years agoBetter usage string for reflog.
Matthieu Moy [Wed, 5 Aug 2009 15:36:28 +0000 (17:36 +0200)]
Better usage string for reflog.

Signed-off-by: Matthieu Moy <>
Signed-off-by: Junio C Hamano <>
11 years agohg-to-git: don't import the unused popen2 module
Miklos Vajna [Mon, 3 Aug 2009 16:41:34 +0000 (18:41 +0200)]
hg-to-git: don't import the unused popen2 module

Importing the popen2 module in Python-2.6 results in the
"DeprecationWarning: The popen2 module is deprecated.  Use the
subprocess module." message. The module itself isn't used in fact, so
just removing it solves the problem.

Signed-off-by: Miklos Vajna <>
Signed-off-by: Junio C Hamano <>
11 years agosend-email: remove debug trace
Erik Faye-Lund [Tue, 4 Aug 2009 21:57:34 +0000 (21:57 +0000)]
send-email: remove debug trace

Signed-off-by: Erik Faye-Lund <>
11 years agoconfig: Keep inner whitespace verbatim
Björn Steinbrink [Thu, 30 Jul 2009 11:41:57 +0000 (13:41 +0200)]
config: Keep inner whitespace verbatim

Configuration values are expected to be quoted when they have leading or
trailing whitespace, but inner whitespace should be kept verbatim even if
the value is not quoted. This is already documented in git-config(1), but
the code caused inner whitespace to be collapsed to a single space,
breaking, for example, clones from a path that has two consecutive spaces
in it, as future fetches would only see a single space.

Reported-by: John te Bokkel <>
Signed-off-by: Björn Steinbrink <>
Signed-off-by: Junio C Hamano <>
11 years agoGIT v1.6.3.4
Junio C Hamano [Wed, 29 Jul 2009 06:52:58 +0000 (23:52 -0700)]

Signed-off-by: Junio C Hamano <>
11 years agoconfig.txt: document add.ignore-errors
Stephen Boyd [Sun, 31 May 2009 05:08:02 +0000 (22:08 -0700)]
config.txt: document add.ignore-errors

Use the description of "--ignore-errors" from git-add.txt as

Signed-off-by: Stephen Boyd <>
Signed-off-by: Junio C Hamano <>
11 years agorequest-pull: allow ls-remote to notice remote.$nickname.uploadpack
Tom Grennan [Wed, 29 Jul 2009 01:30:02 +0000 (18:30 -0700)]
request-pull: allow ls-remote to notice remote.$nickname.uploadpack

The location to pull from should be converted from the configured nickname
to URL in the message, but ls-remote should be fed the nickname so that
the command uses remote.$nickname.* variables, most notably "uploadpack".

Signed-off-by: Tom Grennan <>
Signed-off-by: Junio C Hamano <>
11 years agot8005: Nobody writes Russian in shift_jis
Junio C Hamano [Fri, 19 Jun 2009 02:18:37 +0000 (19:18 -0700)]
t8005: Nobody writes Russian in shift_jis

The second and third tests of this script expected that Russian strings
are converted between ISO-8859-5 and Shift_JIS in the "blame --porcelain"
format output correctly.

Sure, many platforms may convert between such a combination, but that is
only because one of the base character set of Shift_JIS, JIS X 0208,
defines codepoints for Russian characters (among others); I do not think
anybody uses Shift_JIS when seriously writing Russian, and it is perfectly
understandable if iconv() libraries on some platforms fail converting
between this combination, as it does not matter in reality.

This patch changes the test to verify Japanese strings are converted
correctly between EUC-JP and Shift_JIS in the same procedure.  The point
of the test is not about verifying the platform's iconv() library, but to
see if "git blame" makes correct iconv() library calls when it should.

We could instead use ISO-8859-5 and KOI8-R as the combination, because
they are both meant to represent Russian, in order to make this test
meaningful on more platforms, but we already use Shift_JIS vs EUC-JP
combinations to test other programs in our test suite, so this combination
is safer from the point of view of the portability.  Besides, I do not
read nor write Russian; sorry ;-)

This change allows tests to pass on my (friend's) Solaris 5.11 box.

Signed-off-by: Junio C Hamano <>
11 years agoFix severe breakage in "git-apply --whitespace=fix"
Junio C Hamano [Sat, 25 Jul 2009 08:29:20 +0000 (01:29 -0700)]
Fix severe breakage in "git-apply --whitespace=fix"

735c674 (Trailing whitespace and no newline fix, 2009-07-22) completely
broke --whitespace=fix, causing it to lose all the empty lines in a patch.

Signed-off-by: Junio C Hamano <>
11 years agoSunOS grep does not understand -C<n> nor -e
Junio C Hamano [Fri, 24 Jul 2009 05:30:07 +0000 (22:30 -0700)]
SunOS grep does not understand -C<n> nor -e

The first "grep -C1" test in t7002 does not pass on my SunOS-5.11-i86pc,
and that is not because our way to spawn external grep is broken, but
because the native grep does not understand -C<n>.

It turns out that Peff was also using this option himself because our
Makefile doesn't do that automatically.  Brandon Casey uses SUNWspro
compiler without having to set this, and it turns out that the compiler
does not define preprocessor macro __unix__ which made him always use the
built-in grep, never an external one.

Let's be more explicit and say that we do not use external grep on Suns.

Signed-off-by: Junio C Hamano <>
11 years agoFix export_marks() error handling.
Matthias Andree [Fri, 24 Jul 2009 08:17:13 +0000 (10:17 +0200)]
Fix export_marks() error handling.

- Don't leak one FILE * on error per export_marks() call. Found with
  cppcheck and reported by Martin Ettl.

- Abort the potentially long for(;idnums.size;) loop on write errors.

- Record error if fprintf() fails for reasons not required to set the
  stream error indicator, such as ENOMEM.

- Add a trailing full-stop to error message when fopen() fails.

Signed-off-by: Matthias Andree <>
Signed-off-by: Junio C Hamano <>
11 years agogit branch: clean up detached branch handling
Linus Torvalds [Thu, 23 Jul 2009 19:13:48 +0000 (12:13 -0700)]
git branch: clean up detached branch handling

Make the 'show detached branch info' a routine of its own.  And in the
process, avoid the object lookup that is unnecessary if the current
branch isn't detached.

Signed-off-by: Linus Torvalds <>
Signed-off-by: Junio C Hamano <>
11 years agogit branch: avoid unnecessary object lookups
Linus Torvalds [Thu, 23 Jul 2009 19:05:34 +0000 (12:05 -0700)]
git branch: avoid unnecessary object lookups

They can be expensive in the cold-cache case, so don't bother looking up
the commits for all branches unless we really need them for some reason.

Signed-off-by: Linus Torvalds <>
Signed-off-by: Junio C Hamano <>
11 years agogit branch: fix performance problem
Linus Torvalds [Thu, 23 Jul 2009 17:17:04 +0000 (10:17 -0700)]
git branch: fix performance problem

'git branch' looks at _all_ the refs, and verifies them.  Which means that
during cold-cache situations with a slow disk (and lots of tags, for
example) it can take several very annoying seconds (7.5s according to a
report by Carlos R.  Mafra).

This avoids most of it by simply doing the filtering before looking up
the commits, by using the "raw" version of for_each_ref.

Reported-by: Carlos R. Mafra <>
Signed-off-by: Linus Torvalds <>
Signed-off-by: Junio C Hamano <>
11 years agodo_one_ref(): null_sha1 check is not about broken ref
Junio C Hamano [Thu, 23 Jul 2009 06:07:05 +0000 (23:07 -0700)]
do_one_ref(): null_sha1 check is not about broken ref

f8948e2 (remote prune: warn dangling symrefs, 2009-02-08) introduced a
more dangerous variant of for_each_ref() family that skips the check for
dangling refs, but it also made another unrelated check optional by

The check to see if a ref points at 0{40} is not about brokenness, but is
about a possible future plan to represent a deleted ref by writing 40 "0"
in a loose ref when there is a stale version of the same ref already in
.git/packed-refs, so that we can implement deletion of a ref without
having to rewrite the packed refs file excluding the ref being deleted.

This check has to live outside of the conditional.

Signed-off-by: Junio C Hamano <>
11 years agoTrailing whitespace and no newline fix
SZEDER Gábor [Thu, 23 Jul 2009 00:24:38 +0000 (19:24 -0500)]
Trailing whitespace and no newline fix

If a patch adds a new line to the end of a file and this line ends with
one trailing whitespace character and has no newline, then
'--whitespace=fix' currently does not remove that trailing whitespace.

This patch fixes this by removing the check for trailing whitespace at
the end of the line at a hardcoded offset which does not take the
eventual absence of newline into account.

Signed-off-by: SZEDER Gábor <>
Signed-off-by: Junio C Hamano <>
11 years agodiff --cc: a lost line at the beginning of the file is shown incorrectly
Junio C Hamano [Wed, 22 Jul 2009 21:48:29 +0000 (14:48 -0700)]
diff --cc: a lost line at the beginning of the file is shown incorrectly

When combine-diff inspected the diff from one parent to the merge result,
it misinterpreted a header in the form @@ -l,k +0,0 @@.

This hunk header means that K lines were removed from the beginning of the
file, so the lost lines must be queued to the sline that represents the
first line of the merge result, but we incremented our pointer incorrectly
and ended up queuing it to the second line, which in turn made the lossage
appear _after_ the first line.

Signed-off-by: Junio C Hamano <>
11 years agocombine-diff.c: fix performance problem when folding common deleted lines
Junio C Hamano [Wed, 22 Jul 2009 21:48:28 +0000 (14:48 -0700)]
combine-diff.c: fix performance problem when folding common deleted lines

For a deleted line in a patch with the parent we are looking at, the
append_lost() function finds the same line among a run of lines that were
deleted from the same location by patches from parents we previously
checked.  This is so that patches with two parents

    @@ -1,4 +1,3 @@    @@ -1,4 +1,3 @@
     one                   one
    -two                  -two
     three                 three
    -quatro               -fyra
    +four                 +four

can be coalesced into this sequence, reusing one line that describes the
removal of "two" for both parents.

   @@@ -1,4 -1,4 +1,3 @@@
   - quatro

While reading the second patch (that removes "two" and then "fyra"), after
finding where removal of the "two" matches, we need to find existing
removal of "fyra" (if exists) in the removal list, but the match has to
happen after all the existing matches (in this case "two").  The code used
a naïve O(n^2) algorithm to compute this by scanning the whole removal
list over and over again.

This patch remembers where the next scan should be started in the existing
removal list to avoid this.

Noticed by Linus Torvalds.

Signed-off-by: Junio C Hamano <>
11 years agocheckout -f: deal with a D/F conflict entry correctly
Junio C Hamano [Sat, 18 Jul 2009 19:26:38 +0000 (12:26 -0700)]
checkout -f: deal with a D/F conflict entry correctly

When we switch branches with "checkout -f", unpack_trees() feeds two
cache_entries to oneway_merge() function in its src[] array argument.  The
zeroth entry comes from the current index, and the first entry represents
what the merge result should be, taken from the tree recorded in the
commit we are switching to.

When we have a blob (either regular file or a symlink) in the index and in
the work tree at path "foo", and the switched-to tree has "foo/bar",
i.e. "foo" becomes a directory, src[0] is obviously that blob currently
registered at "foo".  Even though we do not have anything at "foo" in the
switched-to tree, src[1] is _not_ NULL in this case.

The unpack_trees() machinery places a special marker df_conflict_entry
to signal that no blob exists at "foo", but it will become a directory
that may have somthing underneath it (namely "foo/bar"), so a usual 3-way
merge can notice the situation.

But oneway_merge() codepath failed to notice this and passed the special
marker directly to merged_entry().  This happens to remove the "foo" in
the end because the df_conflict_entry does not have any name (hence the
"error" message) and its addition in add_index_entry() is rejected, but it
is wrong.

Signed-off-by: Junio C Hamano <>
Acked-by: Linus Torvalds <>
11 years agosha1_name.c: avoid unnecessary strbuf_release
Brandon Casey [Thu, 16 Jul 2009 21:25:19 +0000 (16:25 -0500)]
sha1_name.c: avoid unnecessary strbuf_release

When we fall back to a standard for_each_reflog_ent() after failing to find
the nth branch switch (or if we had a short reflog) with the call to
for_each_recent_reflog_ent(), we do not need to free the memory allocated
for our strbuf's since a strbuf_reset() will be performed in
grab_nth_branch_switch() before assigning to the entry.

Plus, the strbuf_release() negates the non-zero hint we initially gave to
strbuf_init() just above these lines.

Signed-off-by: Brandon Casey <>
Signed-off-by: Junio C Hamano <>
11 years agorefs.c: release file descriptor on error return
Brandon Casey [Thu, 16 Jul 2009 21:25:18 +0000 (16:25 -0500)]
refs.c: release file descriptor on error return

Signed-off-by: Brandon Casey <>
Signed-off-by: Junio C Hamano <>
11 years agoMerge branch 'cb/maint-fetch-refspec-wo-dst' into maint
Junio C Hamano [Wed, 8 Jul 2009 16:52:25 +0000 (09:52 -0700)]
Merge branch 'cb/maint-fetch-refspec-wo-dst' into maint

* cb/maint-fetch-refspec-wo-dst:
  fetch: do not create ref from empty name

11 years agoMerge branch 'cf/maint-remote-uploadpack-useconfig-fix' into maint
Junio C Hamano [Wed, 8 Jul 2009 16:52:14 +0000 (09:52 -0700)]
Merge branch 'cf/maint-remote-uploadpack-useconfig-fix' into maint

* cf/maint-remote-uploadpack-useconfig-fix:
  git-remote: fix missing .uploadpack usage for show command

11 years agoDocumentation: update description of shell aliases
Sitaram Chamarty [Wed, 1 Jul 2009 16:00:31 +0000 (21:30 +0530)]
Documentation: update description of shell aliases

Aliases that invoke shell commands start from the top-level directory,
but this was not documented.

Signed-off-by: Sitaram Chamarty <>
Signed-off-by: Junio C Hamano <>
11 years agorequest-pull: really really disable pager
Michal Marek [Wed, 1 Jul 2009 09:40:30 +0000 (11:40 +0200)]
request-pull: really really disable pager

Earlier 476cc72 (request-pull: really disable pager, 2009-06-30)
tried to use the correct environment variable to disable paging
from multiple calls to "git log" and friends, but there was one
extra call to "git log" that was not covered by the trick.

Move the setting and exporting of GIT_PAGER much earlier in the
script to cover everybody.

Signed-off-by: Junio C Hamano <>
11 years agoattr: plug minor memory leak
René Scharfe [Tue, 30 Jun 2009 22:30:00 +0000 (00:30 +0200)]
attr: plug minor memory leak

Free the memory allocated for struct strbuf pathbuf when we're done.

Signed-off-by: Rene Scharfe <>
Signed-off-by: Junio C Hamano <>
11 years agorequest-pull: really disable pager
Junio C Hamano [Tue, 30 Jun 2009 18:29:36 +0000 (11:29 -0700)]
request-pull: really disable pager

ff06c74 (Improve request-pull to handle non-rebased branches, 2007-05-01)
attempted to disable pager when running subcommands in this script, but
with a wrong variable.  If GIT_PAGER is set, it takes precedence over

Noticed by Michal Marek.

Signed-off-by: Junio C Hamano <>
11 years agoMakes some cleanup/review in gittutorial
Thadeu Lima de Souza Cascardo [Mon, 29 Jun 2009 15:13:58 +0000 (12:13 -0300)]
Makes some cleanup/review in gittutorial

There are some different but little cleanup changes to fix some missing
quotes, to fix what seemed to be an unended sentence, to reident a
little paragraph with too large a sentence and fix a branch name that
was referred to twice later by another name.

Signed-off-by: Thadeu Lima de Souza Cascardo <>
Signed-off-by: Junio C Hamano <>
11 years agoMakefile: git.o depends on library headers
Johannes Sixt [Mon, 29 Jun 2009 19:42:08 +0000 (21:42 +0200)]
Makefile: git.o depends on library headers

This dependency was not yet specified anywhere else.

Signed-off-by: Johannes Sixt <>
Signed-off-by: Junio C Hamano <>
11 years agogit-submodule documentation: fix foreach example
Miklos Vajna [Sun, 28 Jun 2009 12:55:45 +0000 (14:55 +0200)]
git-submodule documentation: fix foreach example

Backtick and apostrophe are asciidoc markup, so they should be escaped
in order to get the expected result in the rendered manual page.

Signed-off-by: Miklos Vajna <>
Signed-off-by: Junio C Hamano <>
11 years agogitweb/README: fix AliasMatch in example
Giuseppe Bilotta [Sat, 27 Jun 2009 16:24:11 +0000 (18:24 +0200)]
gitweb/README: fix AliasMatch in example

When combining "dumb client" and human-friendly access by using the
'.git' extension to switch between the two, make sure the AliasMatch
covers the entire request. Without a full match, a request for

would result in a 404 because the server would try to access the
the project 'project/shortlog/branch.'

The solution is still not bulletproof, so document the possible failing

Signed-off-by: Giuseppe Bilotta <>
Signed-off-by: Junio C Hamano <>
11 years agoTest grep --and/--or/--not
Thomas Rast [Sat, 27 Jun 2009 18:47:44 +0000 (20:47 +0200)]
Test grep --and/--or/--not

Signed-off-by: Thomas Rast <>
Signed-off-by: Junio C Hamano <>
11 years agoTest git archive --remote
Thomas Rast [Sat, 27 Jun 2009 18:47:43 +0000 (20:47 +0200)]
Test git archive --remote

Add a small test case for git archive --remote (and thus
git-upload-archive), which so far went untested.

Signed-off-by: Thomas Rast <>
Signed-off-by: Junio C Hamano <>
11 years agogit-remote: fix missing .uploadpack usage for show command
Chris Frey [Thu, 25 Jun 2009 21:21:35 +0000 (17:21 -0400)]
git-remote: fix missing .uploadpack usage for show command

For users pulling from machines with self compiled git installs,
in non-PATH locations, they can set the config option
remote.<name>.uploadpack to set the location of git-upload-pack.

When using 'git remote show <name>', the remote HEAD check
did not use the uploadpack configuration setting, and would
not use the configured program.

In builtin-remote.c, the config setting is already loaded
with the call to remote_get(), so this patch passes that remote
along to transport_get().

Signed-off-by: Chris Frey <>
Signed-off-by: Junio C Hamano <>
11 years agofread does not return negative on error
Roel Kluin [Mon, 22 Jun 2009 16:42:33 +0000 (18:42 +0200)]
fread does not return negative on error

size_t res cannot be less than 0. fread returns 0 on error.

Reported-by: Ingo Molnar <>
Signed-off-by: Roel Kluin <>
Signed-off-by: Junio C Hamano <>
11 years agot3700-add: add a POSIXPERM prerequisite to a new test
Johannes Sixt [Mon, 22 Jun 2009 07:30:38 +0000 (09:30 +0200)]
t3700-add: add a POSIXPERM prerequisite to a new test

The new test does a 'chmod 0', which does not have the intended
effect on Windows.

Signed-off-by: Johannes Sixt <>
Signed-off-by: Junio C Hamano <>
11 years agoMerge branch 'sb/maint-1.6.0-add-config-fix' into maint
Junio C Hamano [Mon, 22 Jun 2009 07:44:09 +0000 (00:44 -0700)]
Merge branch 'sb/maint-1.6.0-add-config-fix' into maint

* sb/maint-1.6.0-add-config-fix:
  add: allow configurations to be overriden by command line

11 years agoGIT v1.6.3.3
Junio C Hamano [Mon, 22 Jun 2009 04:02:49 +0000 (21:02 -0700)]

Signed-off-by: Junio C Hamano <>
11 years agoMerge branch 'ak/maint-for-each-ref-no-lookup' into maint
Junio C Hamano [Mon, 22 Jun 2009 04:15:39 +0000 (21:15 -0700)]
Merge branch 'ak/maint-for-each-ref-no-lookup' into maint

* ak/maint-for-each-ref-no-lookup:
  for-each-ref: Do not lookup objects when they will not be used

11 years agoMerge branch 'rc/maint-http-local-slot-fix' into maint
Junio C Hamano [Mon, 22 Jun 2009 04:15:31 +0000 (21:15 -0700)]
Merge branch 'rc/maint-http-local-slot-fix' into maint

* rc/maint-http-local-slot-fix:
  http*: cleanup slot->local after fclose

11 years agoMerge branch 'cb/maint-no-double-merge' into maint
Junio C Hamano [Mon, 22 Jun 2009 04:15:27 +0000 (21:15 -0700)]
Merge branch 'cb/maint-no-double-merge' into maint

* cb/maint-no-double-merge:
  refuse to merge during a merge

11 years agoMerge branch 'mn/maint-iconv-autoconf' into maint
Junio C Hamano [Mon, 22 Jun 2009 04:14:25 +0000 (21:14 -0700)]
Merge branch 'mn/maint-iconv-autoconf' into maint

* mn/maint-iconv-autoconf:
  fix handling of iconv configuration options

11 years agoMerge branch 'lt/maint-unsigned-left-shift' into maint
Junio C Hamano [Mon, 22 Jun 2009 04:14:09 +0000 (21:14 -0700)]
Merge branch 'lt/maint-unsigned-left-shift' into maint

* lt/maint-unsigned-left-shift:
  Fix big left-shifts of unsigned char

11 years agoMerge branch 'pb/maint-1.6.2-userdiff-fix' into maint
Junio C Hamano [Mon, 22 Jun 2009 04:08:05 +0000 (21:08 -0700)]
Merge branch 'pb/maint-1.6.2-userdiff-fix' into maint

* pb/maint-1.6.2-userdiff-fix:
  upload-archive: fix infinite loop on Cygwin
  avoid exponential regex match for java and objc function names

11 years agoattribute: whitespace set to true detects all errors known to git
Junio C Hamano [Sun, 21 Jun 2009 09:35:18 +0000 (02:35 -0700)]
attribute: whitespace set to true detects all errors known to git

That is what the documentation says, but the code pretends as if all the
known whitespace error tokens were given.

Among the whitespace error tokens, there is one kind that loosens the rule
when set: cr-at-eol.  Which means that whitespace error token that is set
to true ignores a newly introduced CR at the end, which is inconsistent
with the documentation.

Signed-off-by: Junio C Hamano <>
11 years ago.gitattributes: CR at the end of the line is an error
Nanako Shiraishi [Fri, 19 Jun 2009 10:42:53 +0000 (19:42 +0900)]
.gitattributes: CR at the end of the line is an error

When a CR is accidentally added at the end of a C source file in the git
project tree, "git diff --check" doesn't detect it as an error.

    $ echo abQ | tr Q '\015' >>fast-import.c
    $ git diff --check

I think this is because the "whitespace" attribute is set to *.[ch] files
without specifying what kind of errors are caught. It makes git "notice
all types of errors" (as described in the documentation), but I think it
is incorrectly setting cr-at-eol, too, and hides this error.

Signed-off-by: Nanako Shiraishi <>
Signed-off-by: Junio C Hamano <>
11 years agot3505: fix abuse of test_expect_code
Junio C Hamano [Sun, 21 Jun 2009 09:01:28 +0000 (02:01 -0700)]
t3505: fix abuse of test_expect_code

The test wanted to make sure that cherry-pick exits with status 1,
but with the way it was placed after "git checkout master &&" meant
that it could have misjudged success if checkout barfed with the
same failure status.

Signed-off-by: Junio C Hamano <>
11 years agoMerge branch 'maint-1.6.2' into maint
Junio C Hamano [Sun, 21 Jun 2009 06:48:46 +0000 (23:48 -0700)]
Merge branch 'maint-1.6.2' into maint

* maint-1.6.2:
  git-show-ref.txt: remove word and make consistent
  git-svn documentation: fix typo in 'rebase vs. pull/merge' section
  use xstrdup, not strdup in ll-merge.c

11 years agoMerge branch 'maint-1.6.1' into maint-1.6.2
Junio C Hamano [Sun, 21 Jun 2009 06:48:28 +0000 (23:48 -0700)]
Merge branch 'maint-1.6.1' into maint-1.6.2

* maint-1.6.1:
  git-show-ref.txt: remove word and make consistent
  git-svn documentation: fix typo in 'rebase vs. pull/merge' section
  use xstrdup, not strdup in ll-merge.c

11 years agoMerge branch 'maint-1.6.0' into maint-1.6.1
Junio C Hamano [Sun, 21 Jun 2009 06:48:21 +0000 (23:48 -0700)]
Merge branch 'maint-1.6.0' into maint-1.6.1

* maint-1.6.0:
  git-show-ref.txt: remove word and make consistent
  git-svn documentation: fix typo in 'rebase vs. pull/merge' section
  use xstrdup, not strdup in ll-merge.c

11 years agogit-show-ref.txt: remove word and make consistent
Stephen Boyd [Sun, 21 Jun 2009 04:40:45 +0000 (21:40 -0700)]
git-show-ref.txt: remove word and make consistent

Under is better than in because of the nested nature of the .git

"also using" sounds a little odd, plus we say combined with later on so
just use that.

Signed-off-by: Stephen Boyd <>
Signed-off-by: Junio C Hamano <>
11 years agogit-svn documentation: fix typo in 'rebase vs. pull/merge' section
Miklos Vajna [Sat, 20 Jun 2009 11:27:15 +0000 (13:27 +0200)]
git-svn documentation: fix typo in 'rebase vs. pull/merge' section

Signed-off-by: Miklos Vajna <>
Signed-off-by: Junio C Hamano <>
11 years agoadd: allow configurations to be overriden by command line
Stephen Boyd [Thu, 18 Jun 2009 09:17:54 +0000 (02:17 -0700)]
add: allow configurations to be overriden by command line

Don't call git_config after parsing the command line options, otherwise
the config settings will override any settings made by the command line.

This can be seen by setting add.ignore_errors and then specifying
--no-ignore-errors when using git-add.

Signed-off-by: Stephen Boyd <>
Signed-off-by: Junio C Hamano <>
11 years agohttp.c: fix compiling with libcurl 7.9.2
Mark Lodato [Mon, 15 Jun 2009 02:39:00 +0000 (22:39 -0400)]
http.c: fix compiling with libcurl 7.9.2

Change the minimimum required libcurl version for the http.sslKey option
to 7.9.3.  Previously, preprocessor macros checked for >= 7.9.2, which
is incorrect because CURLOPT_SSLKEY was introduced in 7.9.3.  This now
allows git to compile with libcurl 7.9.2.

Signed-off-by: Mark Lodato <>
Signed-off-by: Junio C Hamano <>
11 years agofetch: do not create ref from empty name
Clemens Buchacher [Wed, 17 Jun 2009 13:38:36 +0000 (15:38 +0200)]
fetch: do not create ref from empty name

Previously, the refspec "<src>:" would be expanded to
"<src>:refs/heads/". Instead, treat an empty <dst> just like refspecs
without a colon.

Signed-off-by: Clemens Buchacher <>
Signed-off-by: Junio C Hamano <>
11 years agoimport-tars: support symlinks
Johannes Schindelin [Wed, 17 Jun 2009 12:49:39 +0000 (14:49 +0200)]
import-tars: support symlinks

Without this patch, symbolic links are turned into empty files.

Signed-off-by: Johannes Schindelin <>
Signed-off-by: Junio C Hamano <>
11 years agoupload-archive: fix infinite loop on Cygwin
René Scharfe [Wed, 17 Jun 2009 10:11:10 +0000 (12:11 +0200)]
upload-archive: fix infinite loop on Cygwin

On Cygwin, poll() reports POLLIN even for file descriptors that have
reached their end.  This caused git upload-archive to be stuck in an
infinite loop, as it only looked at the POLLIN flag.

In addition to POLLIN, check if read() returned 0, which indicates
end-of-file, and keep looping only as long as at least one of the file
descriptors has input.  This lets the following command finish on its
own when run in a git repository on Cygwin, instead of it getting stuck
after printing all file names:

$ git archive -v --remote . HEAD >/dev/null

Reported-by: Bob Kagy <>
Signed-off-by: Rene Scharfe <>
Signed-off-by: Junio C Hamano <>
11 years agoavoid exponential regex match for java and objc function names
Paolo Bonzini [Wed, 17 Jun 2009 14:26:06 +0000 (16:26 +0200)]
avoid exponential regex match for java and objc function names

In the old regex

^[ \t]*(([ \t]*[A-Za-z_][A-Za-z_0-9]*){2,}[ \t]*\([^;]*)$

you can backtrack arbitrarily from [A-Za-z_0-9]* into [A-Za-z_], thus
causing an exponential number of backtracks.  Ironically it also causes
the regex not to work as intended; for example "catch" can match the
underlined part of the regex, the first repetition matching "c" and
the second matching "atch".

The replacement regex avoids this problem, because it makes sure that
at least a space/tab is eaten on each repetition.  In other words,
a suffix of a repetition can never be a prefix of the next repetition.

Signed-off-by: Paolo Bonzini <>
Signed-off-by: Junio C Hamano <>
11 years agoFix big left-shifts of unsigned char
Linus Torvalds [Thu, 18 Jun 2009 00:22:27 +0000 (17:22 -0700)]
Fix big left-shifts of unsigned char

Shifting 'unsigned char' or 'unsigned short' left can result in sign
extension errors, since the C integer promotion rules means that the
unsigned char/short will get implicitly promoted to a signed 'int' due to
the shift (or due to other operations).

This normally doesn't matter, but if you shift things up sufficiently, it
will now set the sign bit in 'int', and a subsequent cast to a bigger type
(eg 'long' or 'unsigned long') will now sign-extend the value despite the
original expression being unsigned.

One example of this would be something like

unsigned long size;
unsigned char c;

size += c << 24;

where despite all the variables being unsigned, 'c << 24' ends up being a
signed entity, and will get sign-extended when then doing the addition in
an 'unsigned long' type.

Since git uses 'unsigned char' pointers extensively, we actually have this
bug in a couple of places.

I may have missed some, but this is the result of looking at

git grep '[^0-9  ][  ]*<<[  ][a-z]' -- '*.c' '*.h'
git grep '<<[   ]*24'

which catches at least the common byte cases (shifting variables by a
variable amount, and shifting by 24 bits).

I also grepped for just 'unsigned char' variables in general, and
converted the ones that most obviously ended up getting implicitly cast
immediately anyway (eg hash_name(), encode_85()).

In addition to just avoiding 'unsigned char', this patch also tries to use
a common idiom for the delta header size thing. We had three different
variations on it: "& 0x7fUL" in one place (getting the sign extension
right), and "& ~0x80" and "& 0x7f" in two other places (not getting it
right). Apart from making them all just avoid using "unsigned char" at
all, I also unified them to then use a simple "& 0x7f".

I considered making a sparse extension which warns about doing implicit
casts from unsigned types to signed types, but it gets rather complex very
quickly, so this is just a hack.

Signed-off-by: Linus Torvalds <>
Signed-off-by: Junio C Hamano <>
12 years agopull, rebase: simplify to use die()
Stephen Boyd [Sun, 14 Jun 2009 23:08:56 +0000 (16:08 -0700)]
pull, rebase: simplify to use die()

Signed-off-by: Stephen Boyd <>
Signed-off-by: Junio C Hamano <>
12 years agouse xstrdup, not strdup in ll-merge.c
Jim Meyering [Sun, 14 Jun 2009 19:47:54 +0000 (21:47 +0200)]
use xstrdup, not strdup in ll-merge.c

Otherwise, a fluky allocation failure would cause merge
configuration settings to be silently ignored.

Signed-off-by: Jim Meyering <>
Signed-off-by: Junio C Hamano <>
12 years agoMerge branch 'maint-1.6.2' into maint
Junio C Hamano [Sun, 14 Jun 2009 00:10:08 +0000 (17:10 -0700)]
Merge branch 'maint-1.6.2' into maint

* maint-1.6.2:
  git-rerere.txt: grammatical fixups and cleanups

12 years agoMerge branch 'maint-1.6.1' into maint-1.6.2
Junio C Hamano [Sun, 14 Jun 2009 00:09:50 +0000 (17:09 -0700)]
Merge branch 'maint-1.6.1' into maint-1.6.2

* maint-1.6.1:
  git-rerere.txt: grammatical fixups and cleanups

12 years agoMerge branch 'maint-1.6.0' into maint-1.6.1
Junio C Hamano [Sun, 14 Jun 2009 00:09:45 +0000 (17:09 -0700)]
Merge branch 'maint-1.6.0' into maint-1.6.1

* maint-1.6.0:
  git-rerere.txt: grammatical fixups and cleanups
  http-push.c::remove_locks(): fix use after free

12 years agogit-rerere.txt: grammatical fixups and cleanups
Stephen Boyd [Sat, 13 Jun 2009 18:20:00 +0000 (11:20 -0700)]
git-rerere.txt: grammatical fixups and cleanups

Rewrite the gc section using unresolved and resolved instead of "not
recorded". Add plurals and missing articles. Make some sentences have
consistent tense. Try and be more active by removing "that" and
simplifying sentences.

The terms "hand-resolve" and "hand resolve" were used, so just use "hand
resolve" to be more consistent.

Signed-off-by: Stephen Boyd <>
Signed-off-by: Junio C Hamano <>
12 years agoDocumentation: git-send-mail can take rev-list arg to drive format-patch
Paolo Bonzini [Thu, 11 Jun 2009 07:30:27 +0000 (09:30 +0200)]
Documentation: git-send-mail can take rev-list arg to drive format-patch

The git-send-email docs do not mention except in the usage lines
the combined patch formatting/sending ability of git-send-email.
This patch expands on the possible arguments to git-send-email
and explains the meaning of the rev-list argument.

Signed-off-by: Paolo Bonzini <>
Signed-off-by: Junio C Hamano <>
12 years agoMerge branch 'uk/maint-1.5.3-rebase-i-reflog' into maint
Junio C Hamano [Thu, 11 Jun 2009 21:14:00 +0000 (14:14 -0700)]
Merge branch 'uk/maint-1.5.3-rebase-i-reflog' into maint

* uk/maint-1.5.3-rebase-i-reflog:
  rebase--interactive: remote stray closing parenthesis


12 years agorebase--interactive: remote stray closing parenthesis
Uwe Kleine-König [Thu, 11 Jun 2009 20:27:55 +0000 (22:27 +0200)]
rebase--interactive: remote stray closing parenthesis

it was introduced in 68a163c9b483ae352fcfee8c4505d113213daa73

Signed-off-by: Uwe Kleine-König <>
Cc: Jöhännës "Dschö" Schindëlin <>
Signed-off-by: Junio C Hamano <>
12 years agodiff.c: plug a memory leak in an error path
Johannes Sixt [Mon, 8 Jun 2009 20:34:30 +0000 (22:34 +0200)]
diff.c: plug a memory leak in an error path

Signed-off-by: Johannes Sixt <>
Signed-off-by: Junio C Hamano <>
12 years agofetch-pack: close output channel after sideband demultiplexer terminates
Johannes Sixt [Mon, 8 Jun 2009 08:51:22 +0000 (10:51 +0200)]
fetch-pack: close output channel after sideband demultiplexer terminates

fetch-pack runs the sideband demultiplexer using start_async(). This
facility requires that the asynchronously executed function closes the
output file descriptor (see Documentation/technical/api-run-command.txt).
But the sideband demultiplexer did not do that. This fixes it.

In certain error situations this could lock up a fetch operation on
Windows because the asynchronous function is run in a thread; by not
closing the output fd the reading end never got EOF and waited for more
data indefinitely. On Unix this is not a problem because the asynchronous
function is run in a separate process, which exits after the function ends
and so implicitly closes the output.

Since the pack that is sent over the wire encodes the number of objects in
the stream, during normal operation the reading end knows when the stream
ends and terminates by itself, and does not lock up.

Signed-off-by: Johannes Sixt <>
Signed-off-by: Junio C Hamano <>
12 years agofix handling of iconv configuration options
Marco Nelissen [Tue, 9 Jun 2009 03:46:38 +0000 (20:46 -0700)]
fix handling of iconv configuration options

Fix the way in which the configure script handles --without-iconv
(and --with-iconv=no), which it  used to essentially ignore.
Also fix the way the configure script determines the value of
NEEDS_LIBICONV, which would be incorrectly set to 'YesPlease' on
systems that lack iconv entirely.

Signed-off-by: Marco Nelissen <>
Signed-off-by: Junio C Hamano <>
12 years agobuiltin-remote: Make "remote show" display all urls
Michael J Gruber [Sat, 6 Jun 2009 15:16:30 +0000 (17:16 +0200)]
builtin-remote: Make "remote show" display all urls

Currently, "git remote -v" lists all urls whereas "git remote show
$remote" shows only the first. Make it so that both show all.

Signed-off-by: Michael J Gruber <>
Signed-off-by: Junio C Hamano <>
12 years agoDocumentation: refer to gitworkflows(7) from tutorial and git(1)
Thomas Rast [Sat, 6 Jun 2009 13:11:07 +0000 (15:11 +0200)]
Documentation: refer to gitworkflows(7) from tutorial and git(1)

Add references to the gitworkflows(7) manpage added in f948dd8
(Documentation: add manpage about workflows, 2008-10-19) to both
gittutorial(1) and git(1), so that new users might actually discover
and read it.

Noticed by Randal L. Schwartz.

Signed-off-by: Thomas Rast <>
Signed-off-by: Junio C Hamano <>
12 years agodaemon: Strictly parse the "extra arg" part of the command
Shawn O. Pearce [Fri, 5 Jun 2009 01:33:32 +0000 (18:33 -0700)]
daemon: Strictly parse the "extra arg" part of the command

Since (49ba83fb67 "Add virtualization support to git-daemon")
git daemon enters an infinite loop and never terminates if a client
hides any extra arguments in the initial request line which is not
exactly "\0host=blah\0".

Since that change, a client must never insert additional extra
arguments, or attempt to use any argument other than "host=", as
any daemon will get stuck parsing the request line and will never
complete the request.

Since the client can't tell if the daemon is patched or not, it
is not possible to know if additional extra args might actually be
able to be safely requested.

If we ever need to extend the git daemon protocol to support a new
feature, we may have to do something like this to the exchange:

  # If both support git:// v2
  C: 000cgit://v2
  S: 0010ok host user
  C: 0018host
  C: 0027git-upload-pack /pub/linux-2.6.git
  S: ...git-upload-pack header...

  # If client supports git:// v2, server does not:
  C: 000cgit://v2
  S: <EOF>

  C: 003bgit-upload-pack /pub/linux-2.6.git\\0
  S: ...git-upload-pack header...

This requires the client to create two TCP connections to talk to
an older git daemon, however all daemons since the introduction of
daemon.c will safely reject the unknown "git://v2" command request,
so the client can quite easily determine the server supports an
older protocol.

Signed-off-by: Shawn O. Pearce <>
Signed-off-by: Junio C Hamano <>
12 years agohttp*: cleanup slot->local after fclose
Tay Ray Chuan [Sat, 6 Jun 2009 08:43:26 +0000 (16:43 +0800)]
http*: cleanup slot->local after fclose

Set slot->local to NULL after doing a fclose() on the file it points
to. This prevents the passing of a FILE* pointer to a fclose()'d file
to ftell() in http.c::run_active_slot().

This issue was raised by Clemens Buchacher on 30th May 2009:

Signed-off-by: Tay Ray Chuan <>
Signed-off-by: Junio C Hamano <>
12 years agoGIT v1.6.3.2
Junio C Hamano [Thu, 4 Jun 2009 05:42:15 +0000 (22:42 -0700)]

Signed-off-by: Junio C Hamano <>
12 years agogrep: fix empty word-regexp matches
René Scharfe [Wed, 3 Jun 2009 16:19:01 +0000 (18:19 +0200)]
grep: fix empty word-regexp matches

The command "git grep -w ''" dies as soon as it encounters an empty line,
reporting (wrongly) that "regexp returned nonsense".  The first hunk of
this patch relaxes the sanity check that is responsible for that,
allowing matches to start at the end.

The second hunk complements it by making sure that empty matches are
rejected if -w was specified, as they are not really words.

GNU grep does the same:

$ echo foo | grep -c ''
$ echo foo | grep -c -w ''

Signed-off-by: Rene Scharfe <>
Signed-off-by: Junio C Hamano <>
12 years agoblame: correctly handle a path that used to be a directory
Junio C Hamano [Wed, 3 Jun 2009 07:43:22 +0000 (00:43 -0700)]
blame: correctly handle a path that used to be a directory

When trying to see if the same path exists in the parent, we ran
"diff-tree" with pathspec set to the path we are interested in with the
parent, and expect either to have exactly one resulting filepair (either
"changed from the parent", "created when there was none") or nothing (when
there is no change from the parent).

If the path used to be a directory, however, we will also see unbounded
number of entries that talk about the files that used to exist underneath
the directory in question.  Correctly pick only the entry that describes
the path we are interested in in such a case (namely, the creation of the
path as a regular file).

Noticed by Ben Willard.

Signed-off-by: Junio C Hamano <>
12 years agoadd -i: do not dump patch during application
Thomas Rast [Tue, 2 Jun 2009 21:34:27 +0000 (23:34 +0200)]
add -i: do not dump patch during application

Remove a debugging print that snuck in at 7a26e65 (Revert
"git-add--interactive: remove hunk coalescing", 2009-05-16).

Signed-off-by: Thomas Rast <>
Signed-off-by: Junio C Hamano <>
12 years agoUpdate draft release notes for
Junio C Hamano [Tue, 2 Jun 2009 14:57:39 +0000 (07:57 -0700)]
Update draft release notes for

Signed-off-by: Junio C Hamano <>
12 years agoMerge branch 'cb/maint-1.6.0-xdl-merge-fix' into maint
Junio C Hamano [Tue, 2 Jun 2009 14:48:44 +0000 (07:48 -0700)]
Merge branch 'cb/maint-1.6.0-xdl-merge-fix' into maint

* cb/maint-1.6.0-xdl-merge-fix:
  Change xdl_merge to generate output even for null merges
  t6023: merge-file fails to output anything for a degenerate merge


12 years agoMerge branch 'rs/maint-grep-word-regexp-fix' into maint
Junio C Hamano [Tue, 2 Jun 2009 14:48:09 +0000 (07:48 -0700)]
Merge branch 'rs/maint-grep-word-regexp-fix' into maint

* rs/maint-grep-word-regexp-fix:
  grep: fix colouring of matches with zero length
  grep: fix word-regexp at the beginning of lines

12 years agoMerge branch 'sb/maint-1.6.2-opt-filename-fix' into maint
Junio C Hamano [Tue, 2 Jun 2009 14:47:03 +0000 (07:47 -0700)]
Merge branch 'sb/maint-1.6.2-opt-filename-fix' into maint

* sb/maint-1.6.2-opt-filename-fix:
  apply, fmt-merge-msg: use relative filenames
  commit: -F overrides -t

12 years agoMerge branch 'jc/maint-add-p-coalesce-fix' into maint
Junio C Hamano [Tue, 2 Jun 2009 14:46:52 +0000 (07:46 -0700)]
Merge branch 'jc/maint-add-p-coalesce-fix' into maint

* jc/maint-add-p-coalesce-fix:
  t3701: ensure correctly set up repository after skipped tests
  Revert "git-add--interactive: remove hunk coalescing"
  Splitting a hunk that adds a line at the top fails in "add -p"

12 years agoMerge branch 'tr/maint-doc-stash-pop' into maint
Junio C Hamano [Tue, 2 Jun 2009 14:43:45 +0000 (07:43 -0700)]
Merge branch 'tr/maint-doc-stash-pop' into maint

* tr/maint-doc-stash-pop:
  Documentation: teach stash/pop workflow instead of stash/apply

12 years agogrep: fix colouring of matches with zero length
René Scharfe [Mon, 1 Jun 2009 21:53:05 +0000 (23:53 +0200)]
grep: fix colouring of matches with zero length

If a zero-length match is encountered, break out of loop and show the rest
of the line uncoloured.  Otherwise we'd be looping forever, trying to make
progress by advancing the pointer by zero characters.

Signed-off-by: Rene Scharfe <>
Signed-off-by: Junio C Hamano <>
12 years agorefuse to merge during a merge
Clemens Buchacher [Mon, 1 Jun 2009 09:20:56 +0000 (11:20 +0200)]
refuse to merge during a merge

The following is an easy mistake to make for users coming from version
control systems with an "update and commit"-style workflow.

        1. git pull
        2. resolve conflicts
        3. git pull

Step 3 overrides MERGE_HEAD, starting a new merge with dirty index.
IOW, probably not what the user intended. Instead, refuse to merge
again if a merge is in progress.

Reported-by: Dave Olszewski <>
Signed-off-by: Clemens Buchacher <>
Signed-off-by: Junio C Hamano <>
12 years agofor-each-ref: Do not lookup objects when they will not be used
Anders Kaseorg [Wed, 27 May 2009 19:23:12 +0000 (15:23 -0400)]
for-each-ref: Do not lookup objects when they will not be used

This makes commands such as `git for-each-ref --format='%(refname)'`,
which are used heavily by the bash_completion code, run about 6 times
faster on an uncached repository (3 s intead of 18 s on my linux-2.6
repository with several remotes).

Signed-off-by: Anders Kaseorg <>
Signed-off-by: Junio C Hamano <>