git/git.git
7 years agoMerge branch 'cl/p4-use-diff-tree' into maint
Junio C Hamano [Thu, 8 May 2014 17:01:32 +0000 (10:01 -0700)]
Merge branch 'cl/p4-use-diff-tree' into maint

"git p4" dealing with changes in binary files were broken by a
change in 1.9 release.

* cl/p4-use-diff-tree:
  git-p4: format-patch to diff-tree change breaks binary patches

7 years agoMerge branch 'rh/prompt-pcmode-avoid-eval-on-refname' into maint
Junio C Hamano [Thu, 8 May 2014 17:01:18 +0000 (10:01 -0700)]
Merge branch 'rh/prompt-pcmode-avoid-eval-on-refname' into maint

The shell prompt script (in contrib/), when using the PROMPT_COMMAND
interface, used an unsafe construct when showing the branch name in
$PS1.

* rh/prompt-pcmode-avoid-eval-on-refname:
  git-prompt.sh: don't put unsanitized branch names in $PS1

7 years agoMerge branch 'km/avoid-non-function-return-in-rebase' into maint
Junio C Hamano [Thu, 8 May 2014 17:01:06 +0000 (10:01 -0700)]
Merge branch 'km/avoid-non-function-return-in-rebase' into maint

"git rebase" used a POSIX shell construct FreeBSD /bin/sh does not
work well with.

* km/avoid-non-function-return-in-rebase:
  Revert "rebase: fix run_specific_rebase's use of "return" on FreeBSD"
  rebase: avoid non-function use of "return" on FreeBSD

7 years agoMerge branch 'tb/unicode-6.3-zero-width' into maint
Junio C Hamano [Thu, 8 May 2014 17:00:45 +0000 (10:00 -0700)]
Merge branch 'tb/unicode-6.3-zero-width' into maint

Some more Unicode codepoints defined in Unicode 6.3 as having zero
width have been taught to our display column counting logic.

* tb/unicode-6.3-zero-width:
  utf8.c: partially update to version 6.3

7 years agoMerge branch 'km/avoid-bs-in-shell-glob' into maint
Junio C Hamano [Thu, 8 May 2014 17:00:36 +0000 (10:00 -0700)]
Merge branch 'km/avoid-bs-in-shell-glob' into maint

Some tests used shell constructs that did not work well on FreeBSD

* km/avoid-bs-in-shell-glob:
  test: fix t5560 on FreeBSD

7 years agoMerge branch 'km/avoid-cp-a' into maint
Junio C Hamano [Thu, 8 May 2014 16:59:41 +0000 (09:59 -0700)]
Merge branch 'km/avoid-cp-a' into maint

Some tests used shell constructs that did not work well on FreeBSD

* km/avoid-cp-a:
  test: fix t7001 cp to use POSIX options

7 years agogit-p4: format-patch to diff-tree change breaks binary patches
Tolga Ceylan [Wed, 7 May 2014 05:48:54 +0000 (22:48 -0700)]
git-p4: format-patch to diff-tree change breaks binary patches

When applying binary patches a full index is required. format-patch
already handles this, but diff-tree needs '--full-index' argument
to always output full index. When git-p4 runs git-apply to test
the patch, git-apply rejects the patch due to abbreviated blob
object names. This is the error message git-apply emits in this
case:

    error: cannot apply binary patch to '<filename>' without full index line
    error: <filename>: patch does not apply

Signed-off-by: Tolga Ceylan <tolga.ceylan@gmail.com>
Acked-by: Pete Wyckoff <pw@padd.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agogit-prompt.sh: don't put unsanitized branch names in $PS1
Richard Hansen [Mon, 21 Apr 2014 23:53:09 +0000 (19:53 -0400)]
git-prompt.sh: don't put unsanitized branch names in $PS1

Both bash and zsh subject the value of PS1 to parameter expansion,
command substitution, and arithmetic expansion.  Rather than include
the raw, unescaped branch name in PS1 when running in two- or
three-argument mode, construct PS1 to reference a variable that holds
the branch name.  Because the shells do not recursively expand, this
avoids arbitrary code execution by specially-crafted branch names such
as '$(IFS=_;cmd=sudo_rm_-rf_/;$cmd)'.

Signed-off-by: Richard Hansen <rhansen@bbn.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoRevert "rebase: fix run_specific_rebase's use of "return" on FreeBSD"
Kyle J. McKay [Fri, 11 Apr 2014 08:28:18 +0000 (01:28 -0700)]
Revert "rebase: fix run_specific_rebase's use of "return" on FreeBSD"

This reverts commit 99855ddf4bd319cd06a0524e755ab1c1b7d39f3b.

The workaround 99855ddf introduced to deal with problematic
"return" statements in scripts run by "dot" commands located
inside functions only handles one part of the problem.  The
issue has now been addressed by not using "return" statements
in this way in the git-rebase--*.sh scripts.

This workaround is therefore no longer necessary, so clean
up the code by reverting it.

Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
Acked-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agorebase: avoid non-function use of "return" on FreeBSD
Kyle J. McKay [Fri, 11 Apr 2014 08:28:17 +0000 (01:28 -0700)]
rebase: avoid non-function use of "return" on FreeBSD

Since a1549e1015d4bf2e and 01a1e646 (first appearing in v1.8.4)
the git-rebase--*.sh scripts have used a "return" to stop execution
of the dot-sourced file and return to the "dot" command that
dot-sourced it.  The /bin/sh utility on FreeBSD however behaves
poorly under some circumstances when such a "return" is executed.

In particular, if the "dot" command is contained within a function,
then when a "return" is executed by the script it runs (that is not
itself inside a function), control will return from the function
that contains the "dot" command skipping any statements that might
follow the dot command inside that function.  Commit 99855ddf (first
appearing in v1.8.4.1) addresses this by making the "dot" command
the last line in the function.

Unfortunately the FreeBSD /bin/sh may also execute some statements
in the script run by the "dot" command that appear after the
troublesome "return".  The fix in 99855ddf does not address this
problem.

For example, if you have script1.sh with these contents:

run_script2() {
        . "$(dirname -- "$0")/script2.sh"
        _e=$?
        echo only this line should show
        [ $_e -eq 5 ] || echo expected status 5 got $_e
        return 3
}
run_script2
e=$?
[ $e -eq 3 ] || { echo expected status 3 got $e; exit 1; }

And script2.sh with these contents:

if [ 5 -gt 3 ]; then
        return 5
fi
case bad in *)
        echo always shows
esac
echo should not get here
! :

When running script1.sh (e.g. '/bin/sh script1.sh' or './script1.sh'
after making it executable), the expected output from a POSIX shell
is simply the single line:

only this line should show

However, when run using FreeBSD's /bin/sh, the following output
appears instead:

should not get here
expected status 3 got 1

Not only did the lines following the "dot" command in the run_script2
function in script1.sh get skipped, but additional lines in script2.sh
following the "return" got executed -- but not all of them (e.g. the
"echo always shows" line did not run).

These issues can be avoided by not using a top-level "return" in
script2.sh.  If script2.sh is changed to this:

main() {
        if [ 5 -gt 3 ]; then
                return 5
        fi
        case bad in *)
                echo always shows
        esac
        echo should not get here
        ! :
}
main

Then it behaves the same when using FreeBSD's /bin/sh as when using
other more POSIX compliant /bin/sh implementations.

We fix the git-rebase--*.sh scripts in a similar fashion by moving
the top-level code that contains "return" statements into its own
function and then calling that as the last line in the script.

Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
Acked-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agotest: fix t5560 on FreeBSD
Kyle J. McKay [Fri, 11 Apr 2014 08:28:19 +0000 (01:28 -0700)]
test: fix t5560 on FreeBSD

Since fd0a8c2e (first appearing in v1.7.0), the
t/t5560-http-backend-noserver.sh test has used a backslash escape
inside a ${} expansion in order to specify a literal '?' character.

Unfortunately the FreeBSD /bin/sh does not interpret this correctly.

In a POSIX compliant shell, the following:

x='one?two?three'
echo "${x#*\?}"

Would be expected to produce this:

two?three

When using the FreeBSD /bin/sh instead you get this:

one?two?three

In fact the FreeBSD /bin/sh treats the backslash as a literal
character to match so that this:

y='one\two\three'
echo "${y#*\?}"

Produces this unexpected value:

wo\three

In this case the backslash is not only treated literally, it also
fails to defeat the special meaning of the '?' character.

Instead, we can use the [...] construct to defeat the special meaning
of the '?' character and match it exactly in a way that works for the
FreeBSD /bin/sh as well as other POSIX /bin/sh implementations.

Changing the example like so:

x='one?two?three'
echo "${x#*[?]}"

Produces the expected output using the FreeBSD /bin/sh.

Therefore, change the use of \? to [?] in order to be compatible with
the FreeBSD /bin/sh which allows t/t5560-http-backend-noserver.sh to
pass on FreeBSD again.

Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agotest: fix t7001 cp to use POSIX options
Kyle J. McKay [Fri, 11 Apr 2014 08:24:02 +0000 (01:24 -0700)]
test: fix t7001 cp to use POSIX options

Since 11502468 and 04c1ee57 (both first appearing in v1.8.5), the
t7001-mv test has used "cp -a" to perform a copy in several of the
tests.

However, the "-a" option is not required for a POSIX cp utility and
some platforms' cp utilities do not support it.

The POSIX equivalent of -a is -R -P -p.

Change "cp -a" to "cp -R -P -p" so that the t7001-mv test works
on systems with a cp utility that only implements the POSIX
required set of options and not the "-a" option.

Signed-off-by: Kyle J. McKay <mackyle@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoGit 1.9.2 v1.9.2
Junio C Hamano [Wed, 9 Apr 2014 19:04:34 +0000 (12:04 -0700)]
Git 1.9.2

The second maintenance release for Git 1.9; contains all the fixes
that are scheduled to appear in Git 2.0.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoMerge branch 'jl/nor-or-nand-and' into maint
Junio C Hamano [Wed, 9 Apr 2014 19:03:26 +0000 (12:03 -0700)]
Merge branch 'jl/nor-or-nand-and' into maint

* jl/nor-or-nand-and:
  code and test: fix misuses of "nor"
  comments: fix misuses of "nor"
  contrib: fix misuses of "nor"
  Documentation: fix misuses of "nor"

7 years agoMerge branch 'cn/fetch-prune-overlapping-destination' into maint
Junio C Hamano [Wed, 9 Apr 2014 19:02:41 +0000 (12:02 -0700)]
Merge branch 'cn/fetch-prune-overlapping-destination' into maint

* cn/fetch-prune-overlapping-destination:
  fetch: handle overlaping refspecs on --prune
  fetch: add a failing test for prunning with overlapping refspecs

7 years agoMerge branch 'mh/update-ref-batch-create-fix' into maint
Junio C Hamano [Wed, 9 Apr 2014 19:01:28 +0000 (12:01 -0700)]
Merge branch 'mh/update-ref-batch-create-fix' into maint

* mh/update-ref-batch-create-fix:
  update-ref: fail create operation over stdin if ref already exists

7 years agoMerge branch 'jk/commit-dates-parsing-fix' into maint
Junio C Hamano [Wed, 9 Apr 2014 18:59:38 +0000 (11:59 -0700)]
Merge branch 'jk/commit-dates-parsing-fix' into maint

* jk/commit-dates-parsing-fix:
  t4212: loosen far-in-future test for AIX
  date: recognize bogus FreeBSD gmtime output

7 years agoMerge branch 'jc/fix-diff-no-index-diff-opt-parse' into maint
Junio C Hamano [Wed, 9 Apr 2014 18:59:16 +0000 (11:59 -0700)]
Merge branch 'jc/fix-diff-no-index-diff-opt-parse' into maint

* jc/fix-diff-no-index-diff-opt-parse:
  diff-no-index: correctly diagnose error return from diff_opt_parse()

7 years agoMerge commit 'doc/http-backend: missing accent grave in literal mark-up'
Junio C Hamano [Wed, 9 Apr 2014 18:45:04 +0000 (11:45 -0700)]
Merge commit 'doc/http-backend: missing accent grave in literal mark-up'

* commit '5df05146d5cb94628a3dfc53063c802ee1152cec':
  doc/http-backend: missing accent grave in literal mark-up

7 years agodoc/http-backend: missing accent grave in literal mark-up
Thomas Ackermann [Wed, 9 Apr 2014 18:17:38 +0000 (20:17 +0200)]
doc/http-backend: missing accent grave in literal mark-up

Signed-off-by: Thomas Ackermann <th.acker@arcor.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoutf8.c: partially update to version 6.3
Torsten Bögershausen [Mon, 7 Apr 2014 19:39:41 +0000 (21:39 +0200)]
utf8.c: partially update to version 6.3

Unicode 6.3 defines more code points as combining or accents.  For
example, the character "ö" could be expressed as an "o" followed by
U+0308 COMBINING DIARESIS (aka umlaut, double-dot-above).  We should
consider that such a sequence of two codepoints occupies one display
column for the alignment purposes, and for that, git_wcwidth()
should return 0 for them.  Affected codepoints are:

    U+0358..U+035C
    U+0487
    U+05A2, U+05BA, U+05C5, U+05C7
    U+0604, U+0616..U+061A, U+0659..U+065F

Earlier unicode standards had defined these as "reserved".

Only the range 0..U+07FF has been checked to see which codepoints
need to be marked as 0-width while preparing for this commit; more
updates may be needed.

Signed-off-by: Torsten Bögershausen <tboegi@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoUpdate draft release notes to 1.9.2
Junio C Hamano [Tue, 8 Apr 2014 19:08:34 +0000 (12:08 -0700)]
Update draft release notes to 1.9.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoMerge branch 'mm/status-porcelain-format-i18n-fix' into maint
Junio C Hamano [Tue, 8 Apr 2014 19:07:06 +0000 (12:07 -0700)]
Merge branch 'mm/status-porcelain-format-i18n-fix' into maint

* mm/status-porcelain-format-i18n-fix:
  status: disable translation when --porcelain is used

7 years agoMerge branch 'bp/commit-p-editor' into maint
Junio C Hamano [Tue, 8 Apr 2014 19:07:06 +0000 (12:07 -0700)]
Merge branch 'bp/commit-p-editor' into maint

* bp/commit-p-editor:
  run-command: mark run_hook_with_custom_index as deprecated
  merge hook tests: fix and update tests
  merge: fix GIT_EDITOR override for commit hook
  commit: fix patch hunk editing with "commit -p -m"
  test patch hunk editing with "commit -p -m"
  merge hook tests: use 'test_must_fail' instead of '!'
  merge hook tests: fix missing '&&' in test

7 years agoStart preparing for 1.9.1
Junio C Hamano [Thu, 3 Apr 2014 20:37:29 +0000 (13:37 -0700)]
Start preparing for 1.9.1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoMerge branch 'jk/mv-submodules-fix' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:06 +0000 (13:39 -0700)]
Merge branch 'jk/mv-submodules-fix' into maint

* jk/mv-submodules-fix:
  mv: prevent mismatched data when ignoring errors.
  builtin/mv: fix out of bounds write

Conflicts:
t/t7001-mv.sh

7 years agoMerge branch 'mh/remove-subtree-long-pathname-fix' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:05 +0000 (13:39 -0700)]
Merge branch 'mh/remove-subtree-long-pathname-fix' into maint

* mh/remove-subtree-long-pathname-fix:
  entry.c: fix possible buffer overflow in remove_subtree()
  checkout_entry(): use the strbuf throughout the function

7 years agoMerge branch 'jk/lib-terminal-lazy' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:04 +0000 (13:39 -0700)]
Merge branch 'jk/lib-terminal-lazy' into maint

* jk/lib-terminal-lazy:
  t/lib-terminal: make TTY a lazy prerequisite

7 years agoMerge branch 'nd/index-pack-error-message' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:04 +0000 (13:39 -0700)]
Merge branch 'nd/index-pack-error-message' into maint

* nd/index-pack-error-message:
  index-pack: report error using the correct variable

7 years agoMerge branch 'us/printf-not-echo' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:04 +0000 (13:39 -0700)]
Merge branch 'us/printf-not-echo' into maint

* us/printf-not-echo:
  test-lib.sh: do not "echo" caller-supplied strings
  rebase -i: do not "echo" random user-supplied strings

7 years agoMerge branch 'rr/doc-merge-strategies' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:03 +0000 (13:39 -0700)]
Merge branch 'rr/doc-merge-strategies' into maint

* rr/doc-merge-strategies:
  Documentation/merge-strategies: avoid hyphenated commands

7 years agoMerge branch 'jk/shallow-update-fix' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:03 +0000 (13:39 -0700)]
Merge branch 'jk/shallow-update-fix' into maint

* jk/shallow-update-fix:
  shallow: verify shallow file after taking lock
  shallow: automatically clean up shallow tempfiles
  shallow: use stat_validity to check for up-to-date file

7 years agoMerge branch 'jc/stash-pop-not-popped' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:03 +0000 (13:39 -0700)]
Merge branch 'jc/stash-pop-not-popped' into maint

* jc/stash-pop-not-popped:
  stash pop: mention we did not drop the stash upon failing to apply

7 years agoMerge branch 'jn/wt-status' into maint
Junio C Hamano [Thu, 3 Apr 2014 20:39:02 +0000 (13:39 -0700)]
Merge branch 'jn/wt-status' into maint

* jn/wt-status:
  wt-status: lift the artificual "at least 20 columns" floor
  wt-status: i18n of section labels
  wt-status: extract the code to compute width for labels
  wt-status: make full label string to be subject to l10n

7 years agoupdate-ref: fail create operation over stdin if ref already exists
Aman Gupta [Wed, 2 Apr 2014 08:09:54 +0000 (10:09 +0200)]
update-ref: fail create operation over stdin if ref already exists

Signed-off-by: Aman Gupta <aman@tmm1.net>
Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Acked-by: Brad King <brad.king@kitware.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agot4212: loosen far-in-future test for AIX
Jeff King [Tue, 1 Apr 2014 07:43:06 +0000 (03:43 -0400)]
t4212: loosen far-in-future test for AIX

One of the tests in t4212 checks our behavior when we feed
gmtime a date so far in the future that it gives up and
returns NULL. Some implementations, like AIX, may actually
just provide us a bogus result instead.

It's not worth it for us to come up with heuristics that
guess whether the return value is sensible or not. On good
platforms where gmtime reports the problem to us with NULL,
we will print the epoch value. On bad platforms, we will
print garbage.  But our test should be written for the
lowest common denominator so that it passes everywhere.

Reported-by: Charles Bailey <cbailey32@bloomberg.net>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agodate: recognize bogus FreeBSD gmtime output
Jeff King [Tue, 1 Apr 2014 21:28:42 +0000 (17:28 -0400)]
date: recognize bogus FreeBSD gmtime output

Most gmtime implementations return a NULL value when they
encounter an error (and this behavior is specified by ANSI C
and POSIX).  FreeBSD's implementation, however, will simply
leave the "struct tm" untouched.  Let's also recognize this
and convert it to a NULL (with this patch, t4212 should pass
on FreeBSD).

Reported-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agocode and test: fix misuses of "nor"
Justin Lebar [Mon, 31 Mar 2014 22:11:47 +0000 (15:11 -0700)]
code and test: fix misuses of "nor"

Signed-off-by: Justin Lebar <jlebar@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agocomments: fix misuses of "nor"
Justin Lebar [Mon, 31 Mar 2014 22:11:46 +0000 (15:11 -0700)]
comments: fix misuses of "nor"

Signed-off-by: Justin Lebar <jlebar@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agocontrib: fix misuses of "nor"
Justin Lebar [Mon, 31 Mar 2014 22:11:45 +0000 (15:11 -0700)]
contrib: fix misuses of "nor"

Signed-off-by: Justin Lebar <jlebar@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoDocumentation: fix misuses of "nor"
Justin Lebar [Mon, 31 Mar 2014 22:11:44 +0000 (15:11 -0700)]
Documentation: fix misuses of "nor"

Signed-off-by: Justin Lebar <jlebar@google.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agodiff-no-index: correctly diagnose error return from diff_opt_parse()
Junio C Hamano [Mon, 31 Mar 2014 18:47:17 +0000 (11:47 -0700)]
diff-no-index: correctly diagnose error return from diff_opt_parse()

diff_opt_parse() returns the number of options parsed, or often
returns error() which is defined to return -1.  Yes, return value of
0 is "I did not process that option at all", which should cause the
caller to say that, but negative return should not be forgotten.

This bug caused "diff --no-index" to infinitely show the same error
message because the returned value was used to decrement the loop
control variable, e.g.

        $ git diff --no-index --color=words a b
        error: option `color' expects "always", "auto", or "never"
        error: option `color' expects "always", "auto", or "never"
        ...

Instead, make it act like so:

        $ git diff --no-index --color=words a b
        error: option `color' expects "always", "auto", or "never"
        fatal: invalid diff option/value: --color=words

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agofetch: handle overlaping refspecs on --prune
Carlos Martín Nieto [Thu, 27 Feb 2014 09:00:10 +0000 (10:00 +0100)]
fetch: handle overlaping refspecs on --prune

We need to consider that a remote-tracking branch may match more than
one rhs of a fetch refspec. In such a case, it is not enough to stop at
the first match but look at all of the matches in order to determine
whether a head is stale.

To this goal, introduce a variant of query_refspecs which returns all of
the matching refspecs and loop over those answers to check for
staleness.

Signed-off-by: Carlos Martín Nieto <cmn@elego.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agostatus: disable translation when --porcelain is used
Matthieu Moy [Thu, 20 Mar 2014 12:12:41 +0000 (13:12 +0100)]
status: disable translation when --porcelain is used

"git status --branch --porcelain" displays the status of the branch
(ahead, behind, gone), and used gettext to translate the string.

Use hardcoded strings when --porcelain is used, but keep the gettext
translation for "git status --short" which is essentially the same, but
meant to be read by a human.

Reported-by: Anarky <ghostanarky@gmail.com>
Signed-off-by: Matthieu Moy <Matthieu.Moy@imag.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoGit 1.9.1 v1.9.1
Junio C Hamano [Tue, 18 Mar 2014 21:06:49 +0000 (14:06 -0700)]
Git 1.9.1

The version numbering scheme has changed since Git 1.9 and we
dropped the third dewey-decimal from the traditional numbering
(e.g. both 1.8.4 and 1.8.5 were major feature releases).  This
release 1.9.1 is the first maintenance relase for Git 1.9.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoMerge branch 'jk/clean-d-pathspec' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:04:59 +0000 (14:04 -0700)]
Merge branch 'jk/clean-d-pathspec' into maint

"git clean -d pathspec" did not use the given pathspec correctly
and ended up cleaning too much.

* jk/clean-d-pathspec:
  clean: simplify dir/not-dir logic
  clean: respect pathspecs with "-d"

7 years agoMerge branch 'da/difftool-git-files' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:04:36 +0000 (14:04 -0700)]
Merge branch 'da/difftool-git-files' into maint

"git difftool" misbehaved when the repository is bound to the
working tree with the ".git file" mechanism, where a textual file
".git" tells us where it is.

* da/difftool-git-files:
  t7800: add a difftool test for .git-files
  difftool: support repositories with .git-files

7 years agoMerge branch 'jk/remote-pushremote-config-reading' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:04:16 +0000 (14:04 -0700)]
Merge branch 'jk/remote-pushremote-config-reading' into maint

"git push" did not pay attention to branch.*.pushremote if it is
defined earlier than remote.pushdefault; the order of these two
variables in the configuration file should not matter, but it did by
mistake.

* jk/remote-pushremote-config-reading:
  remote: handle pushremote config in any order

7 years agoMerge branch 'jk/commit-dates-parsing-fix' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:04:01 +0000 (14:04 -0700)]
Merge branch 'jk/commit-dates-parsing-fix' into maint

Codepaths that parse timestamps in commit objects have been
tightened.

* jk/commit-dates-parsing-fix:
  show_ident_date: fix tz range check
  log: do not segfault on gmtime errors
  log: handle integer overflow in timestamps
  date: check date overflow against time_t
  fsck: report integer overflow in author timestamps
  t4212: test bogus timestamps with git-log

7 years agoMerge branch 'tr/diff-submodule-no-reuse-worktree' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:03:41 +0000 (14:03 -0700)]
Merge branch 'tr/diff-submodule-no-reuse-worktree' into maint

"git diff --external-diff" incorrectly fed the submodule directory
in the working tree to the external diff driver when it knew it is
the same as one of the versions being compared.

* tr/diff-submodule-no-reuse-worktree:
  diff: do not reuse_worktree_file for submodules

7 years agoMerge branch 'nd/reset-setup-worktree' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:03:24 +0000 (14:03 -0700)]
Merge branch 'nd/reset-setup-worktree' into maint

"git reset" needs to refresh the index when working in a working
tree (it can also be used to match the index to the HEAD in an
otherwise bare repository), but it failed to set up the working
tree properly, causing GIT_WORK_TREE to be ignored.

* nd/reset-setup-worktree:
  reset: optionally setup worktree and refresh index on --mixed

7 years agoMerge branch 'jc/check-attr-honor-working-tree' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:03:03 +0000 (14:03 -0700)]
Merge branch 'jc/check-attr-honor-working-tree' into maint

"git check-attr" when working on a repository with a working tree
did not work well when the working tree was specified via the
--work-tree (and obviously with --git-dir) option.

* jc/check-attr-honor-working-tree:
  check-attr: move to the top of working tree when in non-bare repository
  t0003: do not chdir the whole test process

7 years agoMerge branch 'bk/refresh-missing-ok-in-merge-recursive' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:02:37 +0000 (14:02 -0700)]
Merge branch 'bk/refresh-missing-ok-in-merge-recursive' into maint

"merge-recursive" was broken in 1.7.7 era and stopped working in an
empty (temporary) working tree, when there are renames involved.
This has been corrected.

* bk/refresh-missing-ok-in-merge-recursive:
  merge-recursive.c: tolerate missing files while refreshing index
  read-cache.c: extend make_cache_entry refresh flag with options
  read-cache.c: refactor --ignore-missing implementation
  t3030-merge-recursive: test known breakage with empty work tree

7 years agoMerge branch 'ds/rev-parse-required-args' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:01:05 +0000 (14:01 -0700)]
Merge branch 'ds/rev-parse-required-args' into maint

"git rev-parse" was loose in rejecting command line arguments that
do not make sense, e.g. "--default" without the required value for
that option.

* ds/rev-parse-required-args:
  rev-parse: check i before using argv[i] against argc

7 years agoMerge branch 'jk/config-path-include-fix' into maint
Junio C Hamano [Tue, 18 Mar 2014 21:00:15 +0000 (14:00 -0700)]
Merge branch 'jk/config-path-include-fix' into maint

include.path variable (or any variable that expects a path that can
use ~username expansion) in the configuration file is not a boolean,
but the code failed to check it.

* jk/config-path-include-fix:
  handle_path_include: don't look at NULL value
  expand_user_path: do not look at NULL path

7 years agoMerge branch 'nd/diff-quiet-stat-dirty' into maint
Junio C Hamano [Tue, 18 Mar 2014 20:59:55 +0000 (13:59 -0700)]
Merge branch 'nd/diff-quiet-stat-dirty' into maint

"git diff --quiet -- pathspec1 pathspec2" sometimes did not return
correct status value.

* nd/diff-quiet-stat-dirty:
  diff: do not quit early on stat-dirty files
  diff.c: move diffcore_skip_stat_unmatch core logic out for reuse later

7 years agoMerge branch 'nd/http-fetch-shallow-fix' into maint
Junio C Hamano [Tue, 18 Mar 2014 20:59:37 +0000 (13:59 -0700)]
Merge branch 'nd/http-fetch-shallow-fix' into maint

Attempting to deepen a shallow repository by fetching over smart
HTTP transport failed in the protocol exchange, when no-done
extension was used.  The fetching side waited for the list of
shallow boundary commits after the sending end stopped talking to
it.

* nd/http-fetch-shallow-fix:
  t5537: move http tests out to t5539
  fetch-pack: fix deepen shallow over smart http with no-done cap
  protocol-capabilities.txt: document no-done
  protocol-capabilities.txt: refer multi_ack_detailed back to pack-protocol.txt
  pack-protocol.txt: clarify 'obj-id' in the last ACK after 'done'
  test: rename http fetch and push test files
  tests: auto-set LIB_HTTPD_PORT from test name

7 years agoMerge branch 'nd/submodule-pathspec-ending-with-slash' into maint
Junio C Hamano [Tue, 18 Mar 2014 20:58:58 +0000 (13:58 -0700)]
Merge branch 'nd/submodule-pathspec-ending-with-slash' into maint

Allow "git cmd path/", when the 'path' is where a submodule is
bound to the top-level working tree, to match 'path', despite the
extra and unnecessary trailing slash (such a slash is often
given by command line completion).

* nd/submodule-pathspec-ending-with-slash:
  clean: use cache_name_is_other()
  clean: replace match_pathspec() with dir_path_match()
  pathspec: pass directory indicator to match_pathspec_item()
  match_pathspec: match pathspec "foo/" against directory "foo"
  dir.c: prepare match_pathspec_item for taking more flags
  pathspec: rename match_pathspec_depth() to match_pathspec()
  pathspec: convert some match_pathspec_depth() to dir_path_match()
  pathspec: convert some match_pathspec_depth() to ce_path_match()

7 years agotest-lib.sh: do not "echo" caller-supplied strings
Uwe Storbeck [Tue, 18 Mar 2014 00:14:11 +0000 (01:14 +0100)]
test-lib.sh: do not "echo" caller-supplied strings

In some places we "echo" a string that is supplied by the calling
test script and may contain backslash sequences. The echo command
of some shells, most notably "dash", interprets these backslash
sequences (POSIX.1 allows this) which may scramble the test
output.

Signed-off-by: Uwe Storbeck <uwe@ibr.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agorun-command: mark run_hook_with_custom_index as deprecated
Benoit Pierre [Tue, 18 Mar 2014 10:00:56 +0000 (11:00 +0100)]
run-command: mark run_hook_with_custom_index as deprecated

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomerge hook tests: fix and update tests
Benoit Pierre [Tue, 18 Mar 2014 10:00:55 +0000 (11:00 +0100)]
merge hook tests: fix and update tests

- update 'no editor' hook test and add 'editor' hook test
- make sure the tree is reset to a clean state after running a test
  (using test_when_finished) so later tests are not impacted

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomerge: fix GIT_EDITOR override for commit hook
Benoit Pierre [Tue, 18 Mar 2014 10:00:54 +0000 (11:00 +0100)]
merge: fix GIT_EDITOR override for commit hook

Don't set GIT_EDITOR to ":" when calling prepare-commit-msg hook if the
editor is going to be called (e.g. with "merge -e").

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agocommit: fix patch hunk editing with "commit -p -m"
Benoit Pierre [Tue, 18 Mar 2014 10:00:53 +0000 (11:00 +0100)]
commit: fix patch hunk editing with "commit -p -m"

Don't change git environment: move the GIT_EDITOR=":" override to the
hook command subprocess, like it's already done for GIT_INDEX_FILE.

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agotest patch hunk editing with "commit -p -m"
Benoit Pierre [Tue, 18 Mar 2014 10:00:52 +0000 (11:00 +0100)]
test patch hunk editing with "commit -p -m"

Add (failing) tests: with commit changing the environment to let hooks
know that no editor will be used (by setting GIT_EDITOR to ":"), the
"edit hunk" functionality does not work (no editor is launched and the
whole hunk is committed).

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoDocumentation/merge-strategies: avoid hyphenated commands
Ramkumar Ramachandra [Sun, 16 Mar 2014 22:54:56 +0000 (18:54 -0400)]
Documentation/merge-strategies: avoid hyphenated commands

Replace git-pull and git-merge with the corresponding un-hyphenated
versions. While at it, use ` to mark it up instead of '.

Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoindex-pack: report error using the correct variable
Junio C Hamano [Mon, 17 Mar 2014 22:08:36 +0000 (15:08 -0700)]
index-pack: report error using the correct variable

We feed a string pointer that is potentially NULL to die() when
showing the message.  Don't.

Noticed-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoshallow: verify shallow file after taking lock
Jeff King [Sat, 15 Mar 2014 03:47:06 +0000 (23:47 -0400)]
shallow: verify shallow file after taking lock

Before writing the shallow file, we stat() the existing file
to make sure it has not been updated since our operation
began. However, we do not do so under a lock, so there is a
possible race:

  1. Process A takes the lock.

  2. Process B calls check_shallow_file_for_update and finds
     no update.

  3. Process A commits the lockfile.

  4. Process B takes the lock, then overwrite's process A's
     changes.

We can fix this by doing our check while we hold the lock.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agorebase -i: do not "echo" random user-supplied strings
Uwe Storbeck [Fri, 14 Mar 2014 23:56:43 +0000 (00:56 +0100)]
rebase -i: do not "echo" random user-supplied strings

In some places we "echo" a string that comes from a commit log
message, which may have a backslash sequence that is interpreted by
the command (POSIX.1 allows this), most notably "dash"'s built-in
'echo'.

A commit message which contains the string '\n' (or ends with the
string '\c') may result in a garbage line in the todo list of an
interactive rebase which causes the rebase to fail.

To reproduce the behavior (with dash as /bin/sh):

  mkdir test && cd test && git init
  echo 1 >foo && git add foo
  git commit -m"this commit message ends with '\n'"
  echo 2 >foo && git commit -a --fixup HEAD
  git rebase -i --autosquash --root

Now the editor opens with garbage in line 3 which has to be
removed or the rebase fails.

Signed-off-by: Uwe Storbeck <uwe@ibr.ch>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomv: prevent mismatched data when ignoring errors.
brian m. carlson [Sat, 15 Mar 2014 18:56:52 +0000 (18:56 +0000)]
mv: prevent mismatched data when ignoring errors.

We shrink the source and destination arrays, but not the modes or
submodule_gitfile arrays, resulting in potentially mismatched data.  Shrink
all the arrays at the same time to prevent this.  Add tests to ensure the
problem does not recur.

Signed-off-by: brian m. carlson <sandals@crustytoothpaste.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agot/lib-terminal: make TTY a lazy prerequisite
Jeff King [Fri, 14 Mar 2014 21:57:23 +0000 (17:57 -0400)]
t/lib-terminal: make TTY a lazy prerequisite

When lib-terminal.sh is sourced by a test script, we
immediately set up the TTY prerequisite. We do so inside a
test_expect_success, because that nicely isolates any
generated output.

However, this early test can interfere with a script that
later wants to skip all tests (e.g., t5541 then goes on to
set up the httpd server, and wants to skip_all if that
fails). TAP output doesn't let us skip everything after we
have already run at least one test.

We could fix this by reordering the inclusion of
lib-terminal.sh in t5541 to go after the httpd setup.  That
solves this case, but we might eventually hit a case with
circular dependencies, where either lib-*.sh include might
want to skip_all after the other has run a test.  So
instead, let's just remove the ordering constraint entirely
by doing the setup inside a test_lazy_prereq construct,
rather than in a regular test.  We never cared about the
test outcome anyway (it was written to always succeed).

Note that in addition to setting up the prerequisite, the
current test also defines test_terminal. Since we can't
affect the environment from a lazy_prereq, we have to hoist
that out. We previously depended on it _not_ being defined
when the TTY prereq isn't set as a way to ensure that tests
properly declare their dependency on TTY. However, we still
cover the case (see the in-code comment for details).

Reported-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoentry.c: fix possible buffer overflow in remove_subtree()
Michael Haggerty [Thu, 13 Mar 2014 09:19:08 +0000 (10:19 +0100)]
entry.c: fix possible buffer overflow in remove_subtree()

remove_subtree() manipulated path in a fixed-size buffer even though
the length of the input, let alone the length of entries within the
directory, were not known in advance.  Change the function to take a
strbuf argument and use that object as its scratch space.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agocheckout_entry(): use the strbuf throughout the function
Michael Haggerty [Thu, 13 Mar 2014 09:19:07 +0000 (10:19 +0100)]
checkout_entry(): use the strbuf throughout the function

There is no need to break out the "buf" and "len" members into
separate temporary variables.  Rename path_buf to path and use
path.buf and path.len directly.  This makes it easier to reason about
the data flow in the function.

Signed-off-by: Michael Haggerty <mhagger@alum.mit.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agowt-status: lift the artificual "at least 20 columns" floor
Junio C Hamano [Wed, 12 Mar 2014 20:43:51 +0000 (13:43 -0700)]
wt-status: lift the artificual "at least 20 columns" floor

When we show unmerged paths, we had an artificial 20 columns floor
for the width of labels (e.g. "both deleted:") shown next to the
pathnames.  Depending on the locale, this may result in a label that
is too wide when all the label strings are way shorter than 20
columns, or no-op when a label string is longer than 20 columns.

Just drop the artificial floor.  The screen real estate is better
utilized this way when all the strings are shorter.

Adjust the tests to this change.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agowt-status: i18n of section labels
Jonathan Nieder [Thu, 19 Dec 2013 19:43:19 +0000 (11:43 -0800)]
wt-status: i18n of section labels

The original code assumes that:

 (1) the number of bytes written is the width of a string, so they
     can line up;

 (2) the "how" string is always <= 19 bytes.

Neither of which we should assume.

Using the same approach as the earlier 3651e45c (wt-status: take the
alignment burden off translators, 2013-11-05), compute the necessary
column width to hold the longest label and use that for alignment.

cf. http://bugs.debian.org/725777

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Helped-by: Sandy Carter
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agowt-status: extract the code to compute width for labels
Jonathan Nieder [Thu, 19 Dec 2013 19:43:19 +0000 (11:43 -0800)]
wt-status: extract the code to compute width for labels

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agowt-status: make full label string to be subject to l10n
Junio C Hamano [Wed, 12 Mar 2014 20:51:22 +0000 (13:51 -0700)]
wt-status: make full label string to be subject to l10n

Earlier in 3651e45c (wt-status: take the alignment burden off
translators, 2013-11-05), we assumed that it is OK to make the
string before the colon in a label string we give as the section
header of various kinds of changes (e.g. "new file:") translatable.

This assumption apparently does not hold for some languages,
e.g. ones that want to have spaces around the colon.

Also introduce a static label_width to avoid having to run
strlen(padding) over and over.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agobuiltin/mv: fix out of bounds write
John Keeping [Sat, 8 Mar 2014 19:29:17 +0000 (19:29 +0000)]
builtin/mv: fix out of bounds write

When commit a88c915 (mv: move submodules using a gitfile, 2013-07-30)
added the submodule_gitfile array, it was not added to the block that
enlarges the arrays when we are moving a directory so that we do not
have to worry about it being a directory when we perform the actual
move.  After this, the loop continues over the enlarged set of sources.

Since we assume that submodule_gitfile has size argc, if any of the
items in the source directory are submodules we are guaranteed to write
beyond the end of submodule_gitfile.

Fix this by realloc'ing submodule_gitfile at the same time as the other
arrays.

Reported-by: Guillaume Gelin <contact@ramnes.eu>
Signed-off-by: John Keeping <john@keeping.me.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomerge hook tests: use 'test_must_fail' instead of '!'
Benoit Pierre [Mon, 10 Mar 2014 18:49:32 +0000 (19:49 +0100)]
merge hook tests: use 'test_must_fail' instead of '!'

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomerge hook tests: fix missing '&&' in test
Benoit Pierre [Mon, 10 Mar 2014 18:49:31 +0000 (19:49 +0100)]
merge hook tests: fix missing '&&' in test

Signed-off-by: Benoit Pierre <benoit.pierre@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoclean: simplify dir/not-dir logic
Jeff King [Mon, 10 Mar 2014 17:24:47 +0000 (13:24 -0400)]
clean: simplify dir/not-dir logic

When we get a list of paths from read_directory, we further
prune it to create the final list of items to remove. The
code paths for directories and non-directories repeat the
same "add to list" code.

This patch restructures the code so that we don't repeat
ourselves. Also, by following a "if (condition) continue"
pattern like the pathspec check above, it makes it more
obvious that the conditional is about excluding directories
under certain circumstances.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoclean: respect pathspecs with "-d"
Jeff King [Mon, 10 Mar 2014 20:37:30 +0000 (16:37 -0400)]
clean: respect pathspecs with "-d"

git-clean uses read_directory to fill in a `struct dir` with
potential hits. However, read_directory does not actually
check against our pathspec. It uses a simplified version
that may turn up false positives. As a result, we need to
check that any hits match our pathspec. We do so reliably
for non-directories. For directories, if "-d" is not given
we check that the pathspec matched exactly (i.e., we are
even stricter, and require an explicit "git clean foo" to
clean "foo/"). But if "-d" is given, rather than relaxing
the exact match to allow a recursive match, we do not check
the pathspec at all.

This regression was introduced in 113f10f (Make git-clean a
builtin, 2007-11-11).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoshow_ident_date: fix tz range check
Jeff King [Fri, 7 Mar 2014 17:15:01 +0000 (12:15 -0500)]
show_ident_date: fix tz range check

Commit 1dca155fe3fa (log: handle integer overflow in
timestamps, 2014-02-24) tried to catch integer overflow
coming from strtol() on the timezone field by comparing against
LONG_MIN/LONG_MAX. However, the intermediate "tz" variable
is an "int", which means it can never be LONG_MAX on LP64
systems; we would truncate the output from strtol before the
comparison.

Clang's -Wtautological-constant-out-of-range-compare notices
this and rightly complains.

Let's instead store the result of strtol in a long, and then
compare it against INT_MIN/INT_MAX. This will catch overflow
from strtol, and also overflow when we pass the result as an
int to show_date.

Reported-by: Eric Sunshine <sunshine@sunshineco.com>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agot7800: add a difftool test for .git-files
Junio C Hamano [Wed, 5 Mar 2014 09:23:35 +0000 (01:23 -0800)]
t7800: add a difftool test for .git-files

Signed-off-by: David Aguilar <davvid@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoi18n: proposed command missing leading dash
Sandy Carter [Mon, 3 Mar 2014 14:55:53 +0000 (09:55 -0500)]
i18n: proposed command missing leading dash

Add missing leading dash to proposed commands in french output when
using the command:
    git branch --set-upstream remotename/branchname
and when upstream is gone

Signed-off-by: Sandy Carter <sandy.carter@savoirfairelinux.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agofetch: add a failing test for prunning with overlapping refspecs
Carlos Martín Nieto [Thu, 27 Feb 2014 09:00:09 +0000 (10:00 +0100)]
fetch: add a failing test for prunning with overlapping refspecs

When a remote has multiple fetch refspecs and these overlap in the
target namespace, fetch may prune a remote-tracking branch which still
exists in the remote. The test uses a popular form of this, by putting
pull requests as stored in a popular hosting platform alongside "real"
remote-tracking branches.

The fetch command makes a decision of whether to prune based
on the first matching refspec, which in this case is insufficient, as it
covers the pull request names. This pair of refspecs does work as
expected if the more "specific" refspec is the first in the list.

Signed-off-by: Carlos Martín Nieto <cmn@elego.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoshallow: automatically clean up shallow tempfiles
Jeff King [Thu, 27 Feb 2014 11:25:20 +0000 (06:25 -0500)]
shallow: automatically clean up shallow tempfiles

We sometimes write tempfiles of the form "shallow_XXXXXX"
during fetch/push operations with shallow repositories.
Under normal circumstances, we clean up the result when we
are done. However, we do no take steps to clean up after
ourselves when we exit due to die() or signal death.

This patch teaches the tempfile creation code to register
handlers to clean up after ourselves. To handle this, we
change the ownership semantics of the filename returned by
setup_temporary_shallow. It now keeps a copy of the filename
itself, and returns only a const pointer to it.

We can also do away with explicit tempfile removal in the
callers. They all exit not long after finishing with the
file, so they can rely on the auto-cleanup, simplifying the
code.

Note that we keep things simple and maintain only a single
filename to be cleaned. This is sufficient for the current
caller, but we future-proof it with a die("BUG").

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoshallow: use stat_validity to check for up-to-date file
Jeff King [Thu, 27 Feb 2014 10:56:31 +0000 (05:56 -0500)]
shallow: use stat_validity to check for up-to-date file

When we are about to write the shallow file, we check that
it has not changed since we last read it. Instead of
hand-rolling this, we can use stat_validity. This is built
around the index stat-check, so it is more robust than just
checking the mtime, as we do now (it uses the same check as
we do for index files).

The new code also handles the case of a shallow file
appearing unexpectedly. With the current code, two
simultaneous processes making us shallow (e.g., two "git
fetch --depth=1" running at the same time in a non-shallow
repository) can race to overwrite each other.

As a bonus, we also remove a race in determining the stat
information of what we read (we stat and then open, leaving
a race window; instead we should open and then fstat the
descriptor).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agostash pop: mention we did not drop the stash upon failing to apply
Junio C Hamano [Wed, 26 Feb 2014 22:18:54 +0000 (14:18 -0800)]
stash pop: mention we did not drop the stash upon failing to apply

Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agodiff: do not quit early on stat-dirty files
Nguyễn Thái Ngọc Duy [Sat, 25 Jan 2014 06:46:50 +0000 (13:46 +0700)]
diff: do not quit early on stat-dirty files

When QUICK is set (i.e. with --quiet) we try to do as little work as
possible, stopping after seeing the first change. stat-dirty is
considered a "change" but it may turn out not, if no actual content is
changed. The actual content test is performed too late in the process
and the shortcut may be taken prematurely, leading to incorrect return
code.

Assume we do "git diff --quiet". If we have a stat-dirty file "a" and
a really dirty file "b". We break the loop in run_diff_files() and
stop after "a" because we have got a "change". Later in
diffcore_skip_stat_unmatch() we find out "a" is actually not
changed. But there's nothing else in the diff queue, we incorrectly
declare "no change", ignoring the fact that "b" is changed.

This also happens to "git diff --quiet HEAD" when it hits
diff_can_quit_early() in oneway_diff().

This patch does the content test earlier in order to keep going if "a"
is unchanged. The test result is cached so that when
diffcore_skip_stat_unmatch() is done in the end, we spend no cycles on
re-testing "a".

Reported-by: IWAMOTO Toshihiro <iwamoto@valinux.co.jp>
Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agodiff.c: move diffcore_skip_stat_unmatch core logic out for reuse later
Nguyễn Thái Ngọc Duy [Sat, 25 Jan 2014 06:46:49 +0000 (13:46 +0700)]
diff.c: move diffcore_skip_stat_unmatch core logic out for reuse later

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoclean: use cache_name_is_other()
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:35 +0000 (20:40 +0700)]
clean: use cache_name_is_other()

cmd_clean() has the exact same code of index_name_is_other(). Reduce
code duplication.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoclean: replace match_pathspec() with dir_path_match()
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:34 +0000 (20:40 +0700)]
clean: replace match_pathspec() with dir_path_match()

This instance was left out when many match_pathspec() call sites that
take input from dir_entry were converted to dir_path_match() because
it passed a path with the trailing slash stripped out to match_pathspec()
while the others did not. Stripping for all call sites back then would
be a regression because match_pathspec() did not know how to match
pathspec foo/ against _directory_ foo (the stripped version of path
"foo/").

match_pathspec() knows how to do it now. And dir_path_match() strips
the trailing slash also. Use the new function, because the stripping
code is removed in the next patch.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agopathspec: pass directory indicator to match_pathspec_item()
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:33 +0000 (20:40 +0700)]
pathspec: pass directory indicator to match_pathspec_item()

This patch activates the DO_MATCH_DIRECTORY code in m_p_i(), which
makes "git diff HEAD submodule/" and "git diff HEAD submodule" produce
the same output. Previously only the version without trailing slash
returns the difference (if any).

That's the effect of new ce_path_match(). dir_path_match() is not
executed by the new tests. And it should not introduce regressions.

Previously if path "dir/" is passed in with pathspec "dir/", they
obviously match. With new dir_path_match(), the path becomes
_directory_ "dir" vs pathspec "dir/", which is not executed by the old
code path in m_p_i(). The new code path is executed and produces the
same result.

The other case is pathspec "dir" and path "dir/" is now turned to
"dir" (with DO_MATCH_DIRECTORY). Still the same result before or after
the patch.

So why change? Because of the next patch about clean.c.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomatch_pathspec: match pathspec "foo/" against directory "foo"
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:32 +0000 (20:40 +0700)]
match_pathspec: match pathspec "foo/" against directory "foo"

Currently we do support matching pathspec "foo/" against directory
"foo". That is because match_pathspec() has no way to tell "foo" is a
directory and matching "foo/" against _file_ "foo" is wrong.

The callers can now tell match_pathspec if "foo" is a directory, we
could make an exception for this case. Code is not executed though
because no callers pass the flag yet.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agodir.c: prepare match_pathspec_item for taking more flags
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:31 +0000 (20:40 +0700)]
dir.c: prepare match_pathspec_item for taking more flags

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agopathspec: rename match_pathspec_depth() to match_pathspec()
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:30 +0000 (20:40 +0700)]
pathspec: rename match_pathspec_depth() to match_pathspec()

A long time ago, for some reason I was not happy with
match_pathspec(). I created a better version, match_pathspec_depth()
that was suppose to replace match_pathspec()
eventually. match_pathspec() has finally been gone since 6 months
ago. Use the shorter name for match_pathspec_depth().

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agopathspec: convert some match_pathspec_depth() to dir_path_match()
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:29 +0000 (20:40 +0700)]
pathspec: convert some match_pathspec_depth() to dir_path_match()

This helps reduce the number of match_pathspec_depth() call sites and
show how m_p_d() is used. And it usage is:

 - match against an index entry (ce_path_match or match_pathspec_depth
   in ls-files)

 - match against a dir_entry from read_directory (dir_path_match and
   match_pathspec_depth in clean.c, which will be converted later)

 - resolve-undo (rerere.c and ls-files.c)

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agopathspec: convert some match_pathspec_depth() to ce_path_match()
Nguyễn Thái Ngọc Duy [Fri, 24 Jan 2014 13:40:28 +0000 (20:40 +0700)]
pathspec: convert some match_pathspec_depth() to ce_path_match()

This helps reduce the number of match_pathspec_depth() call sites and
show how match_pathspec_depth() is used.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agomerge-recursive.c: tolerate missing files while refreshing index
Brad King [Mon, 27 Jan 2014 14:45:09 +0000 (09:45 -0500)]
merge-recursive.c: tolerate missing files while refreshing index

Teach add_cacheinfo to tell make_cache_entry to skip refreshing stat
information when a file is missing from the work tree.  We do not want
the index to be stat-dirty after the merge but also do not want to fail
when a file happens to be missing.

This fixes the 'merge-recursive w/ empty work tree - ours has rename'
case in t3030-merge-recursive.

Suggested-by: Elijah Newren <newren@gmail.com>
Signed-off-by: Brad King <brad.king@kitware.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
7 years agoread-cache.c: extend make_cache_entry refresh flag with options
Brad King [Mon, 27 Jan 2014 14:45:08 +0000 (09:45 -0500)]
read-cache.c: extend make_cache_entry refresh flag with options

Convert the make_cache_entry boolean 'refresh' argument to a more
general 'refresh_options' argument.  Pass the value through to the
underlying refresh_cache_ent call.  Add option CE_MATCH_REFRESH to
enable stat refresh.  Update call sites to use the new signature.

Signed-off-by: Brad King <brad.king@kitware.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>