git/git.git
8 years agoMerge branch 'jc/test-say-color-avoid-echo-escape'
Jeff King [Thu, 25 Oct 2012 10:42:49 +0000 (06:42 -0400)]
Merge branch 'jc/test-say-color-avoid-echo-escape'

Recent nd/wildmatch series was the first to reveal this ancient bug
in the test scaffolding.

* jc/test-say-color-avoid-echo-escape:
  test-lib: Fix say_color () not to interpret \a\b\c in the message

8 years agoMerge branch 'nd/attr-match-optim'
Jeff King [Thu, 25 Oct 2012 10:42:36 +0000 (06:42 -0400)]
Merge branch 'nd/attr-match-optim'

Trivial and obvious optimization for finding attributes that match
a given path.

* nd/attr-match-optim:
  attr: avoid searching for basename on every match
  attr: avoid strlen() on every match

8 years agoMerge branch 'jk/peel-ref'
Jeff King [Thu, 25 Oct 2012 10:42:27 +0000 (06:42 -0400)]
Merge branch 'jk/peel-ref'

Speeds up "git upload-pack" (what is invoked by "git fetch" on the
other side of the connection) by reducing the cost to advertise the
branches and tags that are available in the repository.

* jk/peel-ref:
  upload-pack: use peel_ref for ref advertisements
  peel_ref: check object type before loading
  peel_ref: do not return a null sha1
  peel_ref: use faster deref_tag_noverify

8 years agoMerge branch 'bw/config-lift-variable-name-length-limit'
Jeff King [Thu, 25 Oct 2012 10:42:11 +0000 (06:42 -0400)]
Merge branch 'bw/config-lift-variable-name-length-limit'

The configuration parser had an unnecessary hardcoded limit on
variable names that was not checked consistently. Lift the limit.

* bw/config-lift-variable-name-length-limit:
  Remove the hard coded length limit on variable names in config files

8 years agoMerge branch 'fa/remote-svn'
Jeff King [Thu, 25 Oct 2012 10:42:01 +0000 (06:42 -0400)]
Merge branch 'fa/remote-svn'

A GSoC project.

* fa/remote-svn:
  Add a test script for remote-svn
  remote-svn: add marks-file regeneration
  Add a svnrdump-simulator replaying a dump file for testing
  remote-svn: add incremental import
  remote-svn: Activate import/export-marks for fast-import
  Create a note for every imported commit containing svn metadata
  vcs-svn: add fast_export_note to create notes
  Allow reading svn dumps from files via file:// urls
  remote-svn, vcs-svn: Enable fetching to private refs
  When debug==1, start fast-import with "--stats" instead of "--quiet"
  Add documentation for the 'bidi-import' capability of remote-helpers
  Connect fast-import to the remote-helper via pipe, adding 'bidi-import' capability
  Add argv_array_detach and argv_array_free_detached
  Add svndump_init_fd to allow reading dumps from arbitrary FDs
  Add git-remote-testsvn to Makefile
  Implement a remote helper for svn in C

8 years agoMerge branch 'jm/diff-context-config'
Jeff King [Thu, 25 Oct 2012 10:41:57 +0000 (06:41 -0400)]
Merge branch 'jm/diff-context-config'

Teaches a new configuration variable to "git diff" Porcelain and
its friends.

* jm/diff-context-config:
  t4055: avoid use of sed 'a' command
  diff: diff.context configuration gives default to -U

8 years agoMerge branch 'jk/no-more-pre-exec-callback'
Jeff King [Thu, 25 Oct 2012 10:41:15 +0000 (06:41 -0400)]
Merge branch 'jk/no-more-pre-exec-callback'

Removes a workaround for buggy version of less older than version
406.

* jk/no-more-pre-exec-callback:
  pager: drop "wait for output to run less" hack

8 years agoGit 1.8.0 v1.8.0
Junio C Hamano [Sun, 21 Oct 2012 20:56:23 +0000 (13:56 -0700)]
Git 1.8.0

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge tag 'gitgui-0.17.0' of git://repo.or.cz/git-gui
Junio C Hamano [Wed, 17 Oct 2012 22:55:46 +0000 (15:55 -0700)]
Merge tag 'gitgui-0.17.0' of git://repo.or.cz/git-gui

git-gui 0.17.0

* tag 'gitgui-0.17.0' of git://repo.or.cz/git-gui:
  git-gui 0.17
  git-gui: Don't prepend the prefix if value looks like a full path
  git-gui: Detect full path when parsing arguments
  git-gui: remove .git/CHERRY_PICK_HEAD after committing
  git-gui: Fix a loose/lose mistake
  git-gui: Fix semi-working shortcuts for unstage and revert
  git-gui: de.po: translate "remote" as "extern"
  git-gui: de.po: translate "bare" as "bloß"
  git-gui: de.po: consistently add untranslated hook names within braces
  git-gui: preserve commit messages in utf-8
  git-gui: open console when using --trace on windows
  git-gui: fix a typo in po/ files
  git-gui: Use PWD if it exists on Mac OS X
  git-gui: fix git-gui crash due to uninitialized variable

8 years agogit-gui 0.17 gitgui-0.17.0
Pat Thoyts [Wed, 17 Oct 2012 20:57:29 +0000 (21:57 +0100)]
git-gui 0.17

Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
8 years agogit-gui: Don't prepend the prefix if value looks like a full path
Andrew Wong [Tue, 2 Oct 2012 16:25:15 +0000 (12:25 -0400)]
git-gui: Don't prepend the prefix if value looks like a full path

When argument parsing fails to detect a file name, "git-gui" will try to
use the previously detected "head" as the file name. We should avoid
prepending the prefix if "head" looks like a full path.

Signed-off-by: Andrew Wong <andrew.kw.w@gmail.com>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
8 years agogit-gui: Detect full path when parsing arguments
Andrew Wong [Tue, 2 Oct 2012 16:25:14 +0000 (12:25 -0400)]
git-gui: Detect full path when parsing arguments

When running "git-gui blame" from a subfolder (which means prefix is
non-empty), if we pass a full path as argument, the argument parsing
will fail to recognize the argument as a file name, because prefix is
prepended to the argument.

This patch handles that scenario by adding an additional branch that
checks the file name without using the prefix.

Signed-off-by: Andrew Wong <andrew.kw.w@gmail.com>
Signed-off-by: Pat Thoyts <patthoyts@users.sourceforge.net>
8 years agoGit 1.8.0-rc3 v1.8.0-rc3
Junio C Hamano [Wed, 17 Oct 2012 17:39:44 +0000 (10:39 -0700)]
Git 1.8.0-rc3

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge git://github.com/git-l10n/git-po
Junio C Hamano [Wed, 17 Oct 2012 17:37:40 +0000 (10:37 -0700)]
Merge git://github.com/git-l10n/git-po

* git://github.com/git-l10n/git-po:
  l10n: vi.po: update from v1.8.0-rc2-4-g42e55
  l10n: Update Swedish translation (1964t0f0u)

8 years agoSync with 1.7.12.4
Junio C Hamano [Wed, 17 Oct 2012 17:37:32 +0000 (10:37 -0700)]
Sync with 1.7.12.4

8 years agoGit 1.7.12.4 v1.7.12.4
Junio C Hamano [Wed, 17 Oct 2012 17:36:01 +0000 (10:36 -0700)]
Git 1.7.12.4

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'jk/maint-http-half-auth-push' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:29:24 +0000 (10:29 -0700)]
Merge branch 'jk/maint-http-half-auth-push' into maint

* jk/maint-http-half-auth-push:
  http: fix segfault in handle_curl_result

8 years agoMerge branch 'rr/git-uri-doc' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:28:56 +0000 (10:28 -0700)]
Merge branch 'rr/git-uri-doc' into maint

* rr/git-uri-doc:
  Git url doc: mark ftp/ftps as read-only and deprecate them

8 years agoMerge branch 'bw/cp-a-is-gnuism' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:28:48 +0000 (10:28 -0700)]
Merge branch 'bw/cp-a-is-gnuism' into maint

* bw/cp-a-is-gnuism:
  tests: "cp -a" is a GNUism

8 years agoMerge branch 'nd/doc-ignore' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:28:42 +0000 (10:28 -0700)]
Merge branch 'nd/doc-ignore' into maint

* nd/doc-ignore:
  gitignore.txt: suggestions how to get literal # or ! at the beginning

8 years agoMerge branch 'jc/doc-long-options' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:28:34 +0000 (10:28 -0700)]
Merge branch 'jc/doc-long-options' into maint

* jc/doc-long-options:
  gitcli: parse-options lets you omit tail of long options

8 years agoMerge branch 'jc/maint-t1450-fsck-order-fix' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:28:19 +0000 (10:28 -0700)]
Merge branch 'jc/maint-t1450-fsck-order-fix' into maint

* jc/maint-t1450-fsck-order-fix:
  t1450: the order the objects are checked is undefined

8 years agoMerge branch 'rr/test-use-shell-path-not-shell' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:27:28 +0000 (10:27 -0700)]
Merge branch 'rr/test-use-shell-path-not-shell' into maint

* rr/test-use-shell-path-not-shell:
  test-lib: use $SHELL_PATH, not $SHELL

8 years agoMerge branch 'rr/test-make-sure-we-have-git' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:27:21 +0000 (10:27 -0700)]
Merge branch 'rr/test-make-sure-we-have-git' into maint

* rr/test-make-sure-we-have-git:
  t/test-lib: make sure Git has already been built

8 years agoMerge branch 'po/maint-docs' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:27:10 +0000 (10:27 -0700)]
Merge branch 'po/maint-docs' into maint

* po/maint-docs:
  Doc branch: show -vv option and alternative
  Doc clean: add See Also link
  Doc add: link gitignore
  Doc: separate gitignore pattern sources
  Doc: shallow clone deepens _to_ new depth

8 years agoMerge branch 'jc/ll-merge-binary-ours' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:26:51 +0000 (10:26 -0700)]
Merge branch 'jc/ll-merge-binary-ours' into maint

* jc/ll-merge-binary-ours:
  ll-merge: warn about inability to merge binary files only when we can't
  attr: "binary" attribute should choose built-in "binary" merge driver
  merge: teach -Xours/-Xtheirs to binary ll-merge driver

8 years agoMerge branch 'db/doc-custom-xmlto' into maint
Junio C Hamano [Wed, 17 Oct 2012 17:25:37 +0000 (10:25 -0700)]
Merge branch 'db/doc-custom-xmlto' into maint

* db/doc-custom-xmlto:
  Documentation/Makefile: Allow custom XMLTO binary

8 years agoMerge branch 'master' of https://github.com/nafmo/git-l10n-sv
Jiang Xin [Wed, 17 Oct 2012 01:38:30 +0000 (09:38 +0800)]
Merge branch 'master' of https://github.com/nafmo/git-l10n-sv

* 'master' of https://github.com/nafmo/git-l10n-sv:
  l10n: Update Swedish translation (1964t0f0u)

8 years agoMerge branch 'master' of git://github.com/vnwildman/git
Jiang Xin [Wed, 17 Oct 2012 01:37:30 +0000 (09:37 +0800)]
Merge branch 'master' of git://github.com/vnwildman/git

* 'master' of git://github.com/vnwildman/git:
  l10n: vi.po: update from v1.8.0-rc2-4-g42e55

8 years agol10n: vi.po: update from v1.8.0-rc2-4-g42e55
Tran Ngoc Quan [Wed, 17 Oct 2012 01:15:28 +0000 (08:15 +0700)]
l10n: vi.po: update from v1.8.0-rc2-4-g42e55

* translate 3 new messages
* review quotes

Signed-off-by: Tran Ngoc Quan <vnwildman@gmail.com>
8 years agoDocumentation/RelNotes: remove "updated up to this revision" markers
Thomas Ackermann [Tue, 16 Oct 2012 17:26:23 +0000 (19:26 +0200)]
Documentation/RelNotes: remove "updated up to this revision" markers

These were used to keep track of the last commit a release notes
entry was written for, and should have been removed when cutting
the final release.

Signed-off-by: Thomas Ackermann <th.acker@arcor.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoUpdate draft release notes to 1.8.0
Junio C Hamano [Tue, 16 Oct 2012 20:27:23 +0000 (13:27 -0700)]
Update draft release notes to 1.8.0

We are almost there...

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'jk/maint-http-half-auth-push'
Junio C Hamano [Tue, 16 Oct 2012 18:44:37 +0000 (11:44 -0700)]
Merge branch 'jk/maint-http-half-auth-push'

Fixes a regression in maint-1.7.11 (v1.7.11.7), maint (v1.7.12.1)
and master (v1.8.0-rc0).

* jk/maint-http-half-auth-push:
  http: fix segfault in handle_curl_result

8 years agol10n: Update Swedish translation (1964t0f0u)
Peter Krefting [Tue, 16 Oct 2012 07:38:11 +0000 (08:38 +0100)]
l10n: Update Swedish translation (1964t0f0u)

Signed-off-by: Peter Krefting <peter@softwolves.pp.se>
8 years agol10n: Update git.pot (3 new, 6 removed messages)
Jiang Xin [Tue, 16 Oct 2012 00:39:10 +0000 (08:39 +0800)]
l10n: Update git.pot (3 new, 6 removed messages)

This po/git.pot update is generated from v1.8.0-rc2-4-g42e55.

Signed-off-by: Jiang Xin <worldhello.net@gmail.com>
8 years agoMerge branch 'master' of git://github.com/vnwildman/git
Jiang Xin [Tue, 16 Oct 2012 00:31:22 +0000 (08:31 +0800)]
Merge branch 'master' of git://github.com/vnwildman/git

* 'master' of git://github.com/vnwildman/git:
  l10n: vi.po: update translation upto cc76011

8 years agol10n: vi.po: update translation upto cc76011
Tran Ngoc Quan [Mon, 15 Oct 2012 07:35:49 +0000 (14:35 +0700)]
l10n: vi.po: update translation upto cc76011

* translate all new messages (100%)
* review some others

Signed-off-by: Tran Ngoc Quan <vnwildman@gmail.com>
8 years agoMerge branch 'maint'
Junio C Hamano [Sun, 14 Oct 2012 06:05:54 +0000 (23:05 -0700)]
Merge branch 'maint'

* maint:
  Fix spelling error in post-receive-email hook

8 years agoFix spelling error in post-receive-email hook
Richard Fearn [Sat, 13 Oct 2012 22:32:50 +0000 (23:32 +0100)]
Fix spelling error in post-receive-email hook

Signed-off-by: Richard Fearn <richardfearn@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'maint'
Junio C Hamano [Fri, 12 Oct 2012 18:36:06 +0000 (11:36 -0700)]
Merge branch 'maint'

* maint:
  gitweb.cgi: fix "comitter_tz" typo in feed

8 years agohttp: fix segfault in handle_curl_result
Jeff King [Fri, 12 Oct 2012 06:22:49 +0000 (02:22 -0400)]
http: fix segfault in handle_curl_result

When we create an http active_request_slot, we can set its
"results" pointer back to local storage. The http code will
fill in the details of how the request went, and we can
access those details even after the slot has been cleaned
up.

Commit 8809703 (http: factor out http error code handling)
switched us from accessing our local results struct directly
to accessing it via the "results" pointer of the slot. That
means we're accessing the slot after it has been marked as
finished, defeating the whole purpose of keeping the results
storage separate.

Most of the time this doesn't matter, as finishing the slot
does not actually clean up the pointer. However, when using
curl's multi interface with the dumb-http revision walker,
we might actually start a new request before handing control
back to the original caller. In that case, we may reuse the
slot, zeroing its results pointer, and leading the original
caller to segfault while looking for its results inside the
slot.

Instead, we need to pass a pointer to our local results
storage to the handle_curl_result function, rather than
relying on the pointer in the slot struct. This matches what
the original code did before the refactoring (which did not
use a separate function, and therefore just accessed the
results struct directly).

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogitweb.cgi: fix "comitter_tz" typo in feed
Dylan Alex Simon [Thu, 11 Oct 2012 20:40:35 +0000 (16:40 -0400)]
gitweb.cgi: fix "comitter_tz" typo in feed

gitweb's feeds sometimes contained committer timestamps in the wrong timezone
due to a misspelling.

Signed-off-by: Dylan Simon <dylan@dylex.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoGit 1.8.0-rc2 v1.8.0-rc2
Junio C Hamano [Thu, 11 Oct 2012 18:18:56 +0000 (11:18 -0700)]
Git 1.8.0-rc2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agotest-lib: Fix say_color () not to interpret \a\b\c in the message
Junio C Hamano [Thu, 11 Oct 2012 17:40:36 +0000 (10:40 -0700)]
test-lib: Fix say_color () not to interpret \a\b\c in the message

When running with color disabled (e.g. under prove to produce TAP
output), say_color() helper function is defined to use echo to show
the message.  With a message that ends with "\c", echo is allowed to
interpret it as "Do not end the line with LF".

Use printf "%s\n" to emit the message literally.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'rr/git-uri-doc'
Junio C Hamano [Wed, 10 Oct 2012 21:57:26 +0000 (14:57 -0700)]
Merge branch 'rr/git-uri-doc'

* rr/git-uri-doc:
  Git url doc: mark ftp/ftps as read-only and deprecate them

8 years agoMerge branch 'bw/cp-a-is-gnuism'
Junio C Hamano [Wed, 10 Oct 2012 21:57:09 +0000 (14:57 -0700)]
Merge branch 'bw/cp-a-is-gnuism'

* bw/cp-a-is-gnuism:
  tests: "cp -a" is a GNUism

8 years agoMerge branch 'nd/doc-ignore'
Junio C Hamano [Wed, 10 Oct 2012 21:57:02 +0000 (14:57 -0700)]
Merge branch 'nd/doc-ignore'

* nd/doc-ignore:
  gitignore.txt: suggestions how to get literal # or ! at the beginning

8 years agoMerge branch 'jc/doc-long-options'
Junio C Hamano [Wed, 10 Oct 2012 21:56:52 +0000 (14:56 -0700)]
Merge branch 'jc/doc-long-options'

* jc/doc-long-options:
  gitcli: parse-options lets you omit tail of long options

8 years agoSync with maint
Junio C Hamano [Wed, 10 Oct 2012 21:54:25 +0000 (14:54 -0700)]
Sync with maint

* maint:
  attr: a note about the order of .gitattributes lookup

8 years agoMerge git://bogomips.org/git-svn
Junio C Hamano [Wed, 10 Oct 2012 20:53:56 +0000 (13:53 -0700)]
Merge git://bogomips.org/git-svn

* git://bogomips.org/git-svn:
  svn test: escape peg revision separator using empty peg rev
  git svn: work around SVN 1.7 mishandling of svn:special changes

8 years agosvn test: escape peg revision separator using empty peg rev
Jonathan Nieder [Tue, 9 Oct 2012 08:41:45 +0000 (01:41 -0700)]
svn test: escape peg revision separator using empty peg rev

This test script uses "svn cp" to create a branch with an @-sign in
its name:

svn cp "pr ject/trunk" "pr ject/branches/not-a@{0}reflog"

That sets up for later tests that fetch the branch and check that git
svn mangles the refname appropriately.

Unfortunately, modern svn versions interpret path arguments with an
@-sign as an example of path@revision syntax (which pegs a path to a
particular revision) and truncate the path or error out with message
"svn: E205000: Syntax error parsing peg revision '{0}reflog'".

When using subversion 1.6.x, escaping the @ sign as %40 avoids trouble
(see 08fd28bb, 2010-07-08).  Newer versions are stricter:

$ svn cp "$repo/pr ject/trunk" "$repo/pr ject/branches/not-a%40{reflog}"
svn: E205000: Syntax error parsing peg revision '%7B0%7Dreflog'

The recommended method for escaping a literal @ sign in a path passed
to subversion is to add an empty peg revision at the end of the path
("branches/not-a@{0}reflog@").  Do that.

Pre-1.6.12 versions of Subversion probably treat the trailing @ as
another literal @-sign (svn issue 3651).  Luckily ever since
v1.8.0-rc0~155^2~7 (t9118: workaround inconsistency between SVN
versions, 2012-07-28) the test can survive that.

Tested with Debian Subversion 1.6.12dfsg-6 and 1.7.5-1 and r1395837
of Subversion trunk (1.8.x).

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Tested-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agogit svn: work around SVN 1.7 mishandling of svn:special changes
Jonathan Nieder [Tue, 9 Oct 2012 10:12:39 +0000 (03:12 -0700)]
git svn: work around SVN 1.7 mishandling of svn:special changes

Subversion represents symlinks as ordinary files with content starting
with "link " and the svn:special property set to "*".  Thus a file can
switch between being a symlink and a non-symlink simply by toggling
its svn:special property, and new checkouts will automatically write a
file of the appropriate type.  Likewise, in subversion 1.6 and older,
running "svn update" would notice changes in filetype and update the
working copy appropriately.

Starting in subversion 1.7 (issue 4091), changes to the svn:special
property trip an assertion instead:

$ svn up svn-tree
Updating 'svn-tree':
svn: E235000: In file 'subversion/libsvn_wc/update_editor.c' \
line 1583: assertion failed (action == svn_wc_conflict_action_edit \
|| action == svn_wc_conflict_action_delete || action == \
svn_wc_conflict_action_replace)

Revisions prepared with ordinary svn commands ("svn add" and not "svn
propset") don't trip this because they represent these filetype
changes using a replace operation, which is approximately equivalent
to removal followed by adding a new file and works fine.  Follow suit.

Noticed using t9100.  After this change, git-svn's file-to-symlink
changes are sent in a format that modern "svn update" can handle and
tests t9100.11-13 pass again.

[ew: s,git-svn\.perl,perl/Git/SVN/Editor.pm,g]

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agoMALLOC_CHECK: Allow checking to be disabled from config.mak
Ramsay Jones [Sat, 6 Oct 2012 17:33:08 +0000 (18:33 +0100)]
MALLOC_CHECK: Allow checking to be disabled from config.mak

The malloc checks can be disabled using the TEST_NO_MALLOC_CHECK
variable, either from the environment or command line of an
'make test' invocation. In order to allow the malloc checks to be
disabled from the 'config.mak' file, we add TEST_NO_MALLOC_CHECK
to the environment using an export directive.

Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoattr: a note about the order of .gitattributes lookup
Nguyen Thai Ngoc Duy [Wed, 10 Oct 2012 13:55:52 +0000 (20:55 +0700)]
attr: a note about the order of .gitattributes lookup

This is the documentation part of

1a9d7e9 (attr.c: read .gitattributes from index as well. - 2007-08-14)
06f33c1 (Read attributes from the index that is being checked out - 2009-03-13)

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'maint'
Junio C Hamano [Tue, 9 Oct 2012 21:23:45 +0000 (14:23 -0700)]
Merge branch 'maint'

* maint:
  l10n: de.po: fix a few minor typos

8 years agoconfigure.ac: Add missing comma to CC_LD_DYNPATH
Øyvind A. Holm [Tue, 9 Oct 2012 14:26:11 +0000 (16:26 +0200)]
configure.ac: Add missing comma to CC_LD_DYNPATH

40bfbde ("build: don't duplicate substitution of make variables",
2012-09-11) by mistake removed a necessary comma at the end of
"CC_LD_DYNPATH=-Wl,rpath," in line 414.

When executing "./configure --with-zlib=PATH", this resulted in

      [...]
      CC xdiff/xhistogram.o
      AR xdiff/lib.a
      LINK git-credential-store
  /usr/bin/ld: bad -rpath option
  collect2: ld returned 1 exit status
  make: *** [git-credential-store] Error 1
  $

during make.

Signed-off-by: Øyvind A. Holm <sunny@sunbase.org>
Acked-by: Stefano Lattarini <stefano.lattarini@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'maint' of git://github.com/git-l10n/git-po into maint
Junio C Hamano [Tue, 9 Oct 2012 18:48:53 +0000 (11:48 -0700)]
Merge branch 'maint' of git://github.com/git-l10n/git-po into maint

* 'maint' of git://github.com/git-l10n/git-po:
  l10n: de.po: fix a few minor typos

8 years agotests: "cp -a" is a GNUism
Ben Walton [Mon, 8 Oct 2012 08:08:01 +0000 (09:08 +0100)]
tests: "cp -a" is a GNUism

These tests just want a bit-for-bit identical copy; they do not need
even -H (there is no symbolic link involved) nor -p (there is no
funny permission or ownership issues involved).

Just use "cp -R" instead.

Signed-off-by: Ben Walton <bdwalton@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoGit url doc: mark ftp/ftps as read-only and deprecate them
Ramkumar Ramachandra [Mon, 8 Oct 2012 17:26:53 +0000 (10:26 -0700)]
Git url doc: mark ftp/ftps as read-only and deprecate them

It is not even worth mentioning their removal; just discourage
people from using them.

Signed-off-by: Ramkumar Ramachandra <artagnon@gmail.com>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoGit 1.8.0-rc1 v1.8.0-rc1
Junio C Hamano [Mon, 8 Oct 2012 18:45:41 +0000 (11:45 -0700)]
Git 1.8.0-rc1

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'jc/maint-t1450-fsck-order-fix'
Junio C Hamano [Mon, 8 Oct 2012 18:43:10 +0000 (11:43 -0700)]
Merge branch 'jc/maint-t1450-fsck-order-fix'

The fsck test assumed too much on what kind of error it will
detect. The only important thing is the inconsistency is detected
as an error.

* jc/maint-t1450-fsck-order-fix:
  t1450: the order the objects are checked is undefined

8 years agoMerge branch 'jc/merge-bases-paint-fix'
Junio C Hamano [Mon, 8 Oct 2012 18:42:15 +0000 (11:42 -0700)]
Merge branch 'jc/merge-bases-paint-fix'

"git fmt-merge-msg" (an internal helper reduce_heads() it uses) had
a severe performance regression; an empty "git pull" took forever to
finish as the result.

* jc/merge-bases-paint-fix:
  paint_down_to_common(): parse commit before relying on its timestamp

8 years agoSync with 1.7.12.3
Junio C Hamano [Mon, 8 Oct 2012 18:41:21 +0000 (11:41 -0700)]
Sync with 1.7.12.3

8 years agoGit 1.7.12.3 v1.7.12.3
Junio C Hamano [Mon, 8 Oct 2012 18:40:43 +0000 (11:40 -0700)]
Git 1.7.12.3

Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoMerge branch 'os/commit-submodule-ignore' into maint
Junio C Hamano [Mon, 8 Oct 2012 18:34:34 +0000 (11:34 -0700)]
Merge branch 'os/commit-submodule-ignore' into maint

"git status" honored the ignore=dirty settings in .gitmodules but
"git commit" didn't.

* os/commit-submodule-ignore:
  commit: pay attention to submodule.$name.ignore in .gitmodules

8 years agoMerge branch 'jk/receive-pack-unpack-error-to-pusher' into maint
Junio C Hamano [Mon, 8 Oct 2012 18:34:19 +0000 (11:34 -0700)]
Merge branch 'jk/receive-pack-unpack-error-to-pusher' into maint

"git receive-pack" (the counterpart to "git push") did not give
progress output while processing objects it received to the puser
when run over the smart-http protocol.

* jk/receive-pack-unpack-error-to-pusher:
  receive-pack: drop "n/a" on unpacker errors
  receive-pack: send pack-processing stderr over sideband
  receive-pack: redirect unpack-objects stdout to /dev/null

8 years agoMerge branch 'rt/maint-clone-single' into maint
Junio C Hamano [Mon, 8 Oct 2012 18:34:02 +0000 (11:34 -0700)]
Merge branch 'rt/maint-clone-single' into maint

A repository created with "git clone --single" had its fetch
refspecs set up just like a clone without "--single", leading the
subsequent "git fetch" to slurp all the other branches, defeating
the whole point of specifying "only this branch".

* rt/maint-clone-single:
  clone --single: limit the fetch refspec to fetched branch

8 years agoMerge branch 'jc/blame-follows-renames' into maint
Junio C Hamano [Mon, 8 Oct 2012 18:33:35 +0000 (11:33 -0700)]
Merge branch 'jc/blame-follows-renames' into maint

It was unclear in the documentation for "git blame" that it is
unnecessary for users to use the "--follow" option.

* jc/blame-follows-renames:
  git blame: document that it always follows origin across whole-file renames

8 years agoMerge branch 'lt/mailinfo-handle-attachment-more-sanely' into maint
Junio C Hamano [Mon, 8 Oct 2012 18:32:59 +0000 (11:32 -0700)]
Merge branch 'lt/mailinfo-handle-attachment-more-sanely' into maint

A patch attached as application/octet-stream (e.g. not text/*) were
mishandled, not correctly honoring Content-Transfer-Encoding
(e.g. base64).

* lt/mailinfo-handle-attachment-more-sanely:
  mailinfo: don't require "text" mime type for attachments

8 years agogitignore.txt: suggestions how to get literal # or ! at the beginning
Nguyễn Thái Ngọc Duy [Sun, 7 Oct 2012 03:13:54 +0000 (10:13 +0700)]
gitignore.txt: suggestions how to get literal # or ! at the beginning

We support backslash escape, but we hide the details behind the phrase
"a shell glob suitable for consumption by fnmatch(3)". So it may not
be obvious how one can get literal # or ! at the beginning of pattern.
Add a few lines on how to work around the magic characters.

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAdd a test script for remote-svn
Florian Achleitner [Wed, 19 Sep 2012 15:21:30 +0000 (17:21 +0200)]
Add a test script for remote-svn

Use svnrdump_sim.py to emulate svnrdump without an svn server.
Tests fetching, incremental fetching, fetching from file://,
and the regeneration of fast-import's marks file.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoremote-svn: add marks-file regeneration
Florian Achleitner [Wed, 19 Sep 2012 15:21:29 +0000 (17:21 +0200)]
remote-svn: add marks-file regeneration

fast-import mark files are stored outside the object database and are
therefore not fetched and can be lost somehow else.  marks provide a
svn revision --> git sha1 mapping, while the notes that are attached
to each commit when it is imported provide a git sha1 --> svn revision
mapping.

If the marks file is not available or not plausible, regenerate it by
walking through the notes tree.  , i.e.  The plausibility check tests
if the highest revision in the marks file matches the revision of the
top ref. It doesn't ensure that the mark file is completely correct.
This could only be done with an effort equal to unconditional
regeneration.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAdd a svnrdump-simulator replaying a dump file for testing
Florian Achleitner [Wed, 19 Sep 2012 15:21:28 +0000 (17:21 +0200)]
Add a svnrdump-simulator replaying a dump file for testing

To ease testing without depending on a reachable svn server, this
compact python script mimics parts of svnrdumps behaviour.  It
requires the remote url to start with sim://.

Start and end revisions are evaluated.  If the requested revision
doesn't exist, as it is the case with incremental imports, if no new
commit was added, it returns 1 (like svnrdump).

To allow using the same dump file for simulating multiple incremental
imports, the highest revision can be limited by setting the environment
variable SVNRMAX to that value. This simulates the situation where
higher revs don't exist yet.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoremote-svn: add incremental import
Florian Achleitner [Wed, 19 Sep 2012 15:21:27 +0000 (17:21 +0200)]
remote-svn: add incremental import

Search for a note attached to the ref to update and read it's
'Revision-number:'-line. Start import from the next svn revision.

If there is no next revision in the svn repo, svnrdump terminates with
a message on stderr an non-zero return value. This looks a little
weird, but there is no other way to know whether there is a new
revision in the svn repo.

On the start of an incremental import, the parent of the first commit
in the fast-import stream is set to the branch name to update. All
following commits specify their parent by a mark number. Previous mark
files are currently not reused.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoremote-svn: Activate import/export-marks for fast-import
Florian Achleitner [Wed, 19 Sep 2012 15:21:26 +0000 (17:21 +0200)]
remote-svn: Activate import/export-marks for fast-import

Enable import and export of a marks file by sending the appropriate
feature commands to fast-import before sending data.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoCreate a note for every imported commit containing svn metadata
Florian Achleitner [Wed, 19 Sep 2012 15:21:25 +0000 (17:21 +0200)]
Create a note for every imported commit containing svn metadata

To provide metadata from svn dumps for further processing, e.g.
branch detection, attach a note to each imported commit that stores
additional information.  The notes are currently hard-coded in
refs/notes/svn/revs.  Currently the following lines from the svn dump
are directly accumulated in the note. This can be refined as needed.

 - "Revision-number"
 - "Node-path"
 - "Node-kind"
 - "Node-action"
 - "Node-copyfrom-path"
 - "Node-copyfrom-rev"

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agovcs-svn: add fast_export_note to create notes
Dmitry Ivankov [Wed, 19 Sep 2012 15:21:24 +0000 (17:21 +0200)]
vcs-svn: add fast_export_note to create notes

fast_export lacked a method to writes notes to fast-import stream.
Add two new functions fast_export_note which is similar to
fast_export_modify. And also add fast_export_buf_to_data to be able to
write inline blobs that don't come from a line_buffer or from delta
application.

To be used like this:

  fast_export_begin_commit("refs/notes/somenotes", ...)
  fast_export_note("refs/heads/master", "inline")
  fast_export_buf_to_data(&data)

or maybe

  fast_export_note("refs/heads/master", sha1)

Signed-off-by: Dmitry Ivankov <divanorama@gmail.com>
Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAllow reading svn dumps from files via file:// urls
Florian Achleitner [Wed, 19 Sep 2012 15:21:23 +0000 (17:21 +0200)]
Allow reading svn dumps from files via file:// urls

For testing as well as for importing large, already available dumps,
it's useful to bypass svnrdump and replay the svndump from a file
directly.

Add support for file:// urls in the remote url, e.g.

  svn::file:///path/to/dump

When the remote helper finds an url starting with file:// it tries to
open that file instead of invoking svnrdump.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoremote-svn, vcs-svn: Enable fetching to private refs
Florian Achleitner [Wed, 19 Sep 2012 15:21:22 +0000 (17:21 +0200)]
remote-svn, vcs-svn: Enable fetching to private refs

The reference to update by the fast-import stream is hard-coded.  When
fetching from a remote the remote-helper shall update refs in a
private namespace, i.e. a private subdir of refs/.  This namespace is
defined by the 'refspec' capability, that the remote-helper advertises
as a reply to the 'capabilities' command.

Extend svndump and fast-export to allow passing the target ref.
Update svn-fe to be compatible.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoWhen debug==1, start fast-import with "--stats" instead of "--quiet"
Florian Achleitner [Wed, 19 Sep 2012 15:21:21 +0000 (17:21 +0200)]
When debug==1, start fast-import with "--stats" instead of "--quiet"

fast-import prints statistics that could be interesting to the
developer of remote helpers.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAdd documentation for the 'bidi-import' capability of remote-helpers
Florian Achleitner [Wed, 19 Sep 2012 15:21:20 +0000 (17:21 +0200)]
Add documentation for the 'bidi-import' capability of remote-helpers

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoConnect fast-import to the remote-helper via pipe, adding 'bidi-import' capability
Florian Achleitner [Wed, 19 Sep 2012 15:21:19 +0000 (17:21 +0200)]
Connect fast-import to the remote-helper via pipe, adding 'bidi-import' capability

The fast-import commands 'cat-blob' and 'ls' can be used by
remote-helpers to retrieve information about blobs and trees that
already exist in fast-import's memory. This requires a channel from
fast-import to the remote-helper.

remote-helpers that use these features shall advertise the new
'bidi-import' capability to signal that they require the communication
channel.  When forking fast-import in transport-helper.c connect it to
a dup of the remote-helper's stdin-pipe. The additional file
descriptor is passed to fast-import via its command line
(--cat-blob-fd).  It follows that git and fast-import are connected to
the remote-helpers's stdin.

Because git can send multiple commands to the remote-helper on it's
stdin, it is required that helpers that advertise 'bidi-import' buffer
all input commands until the batch of 'import' commands is ended by a
newline before sending data to fast-import.  This is to prevent mixing
commands and fast-import responses on the helper's stdin.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAdd argv_array_detach and argv_array_free_detached
Florian Achleitner [Wed, 19 Sep 2012 15:21:18 +0000 (17:21 +0200)]
Add argv_array_detach and argv_array_free_detached

Allow detaching of ownership of the argv_array's contents and add a
function to free those detached argv_arrays later.

This makes it possible to use argv_array efficiently with the exiting
struct child_process which only contains a member char **argv.

Add to documentation.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAdd svndump_init_fd to allow reading dumps from arbitrary FDs
Florian Achleitner [Wed, 19 Sep 2012 15:21:17 +0000 (17:21 +0200)]
Add svndump_init_fd to allow reading dumps from arbitrary FDs

The existing function only allows reading from a filename or from
stdin. Allow passing of a FD and an additional FD for the back report
pipe. This allows us to retrieve the name of the pipe in the caller.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoAdd git-remote-testsvn to Makefile
Florian Achleitner [Wed, 19 Sep 2012 15:21:16 +0000 (17:21 +0200)]
Add git-remote-testsvn to Makefile

The link-rule is a copy of the standard git$X rule but adds VCSSVN_LIB.
Add executable to .gitignore.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoImplement a remote helper for svn in C
Florian Achleitner [Wed, 19 Sep 2012 15:21:15 +0000 (17:21 +0200)]
Implement a remote helper for svn in C

Enable basic fetching from subversion repositories. When processing
remote URLs starting with testsvn::, git invokes this remote-helper.
It starts svnrdump to extract revisions from the subversion repository
in the 'dump file format', and converts them to a git-fast-import stream
using the functions of vcs-svn/.

Imported refs are created in a private namespace at
refs/svn/<remote-name>/master.  The revision history is imported
linearly (no branch detection) and completely, i.e. from revision 0 to
HEAD.

The 'bidi-import' capability is used. The remote-helper expects data
from fast-import on its stdin. It buffers a batch of 'import' command
lines in a string_list before starting to process them.

Signed-off-by: Florian Achleitner <florian.achleitner.2.6.31@gmail.com>
Acked-by: David Michael Barr <b@rr-dav.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogit-svn: keep leading slash when canonicalizing paths (fallback case)
Jonathan Nieder [Fri, 5 Oct 2012 07:04:31 +0000 (00:04 -0700)]
git-svn: keep leading slash when canonicalizing paths (fallback case)

Subversion's svn_dirent_canonicalize() and svn_path_canonicalize()
APIs keep a leading slash in the return value if one was present on
the argument, which can be useful since it allows relative and
absolute paths to be distinguished.

When git-svn's canonicalize_path() learned to use these functions if
available, its semantics changed in the corresponding way.  Some new
callers rely on the leading slash --- for example, if the slash is
stripped out then _canonicalize_url_ourselves() will transform
"proto://host/path/to/resource" to "proto://hostpath/to/resource".

Unfortunately the fallback _canonicalize_path_ourselves(), used when
the appropriate SVN APIs are not usable, still follows the old
semantics, so if that code path is exercised then it breaks.  Fix it
to follow the new convention.

Noticed by forcing the fallback on and running tests.  Without this
patch, t9101.4 fails:

 Bad URL passed to RA layer: Unable to open an ra_local session to \
 URL: Local URL 'file://homejrnsrcgit-scratch/t/trash%20directory.\
 t9101-git-svn-props/svnrepo' contains unsupported hostname at \
 /home/jrn/src/git-scratch/perl/blib/lib/Git/SVN.pm line 148

With it, the git-svn tests pass again.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agoGit::SVN: rename private path field
Jonathan Nieder [Mon, 17 Sep 2012 09:13:31 +0000 (02:13 -0700)]
Git::SVN: rename private path field

All users of $gs->{path} should have been converted to use the
accessor by now.  Check our work by renaming the underlying variable
to break callers that try to use it directly.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agogit-svn: use path accessor for Git::SVN objects
Eric Wong [Tue, 18 Sep 2012 00:09:31 +0000 (00:09 +0000)]
git-svn: use path accessor for Git::SVN objects

The accessors should improve maintainability and enforce
consistent access to Git::SVN objects.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Reviewed-by: Jonathan Nieder <jrnieder@gmail.com>
8 years agoMake git-svn branch patterns match complete URL
Ammon Riley [Thu, 30 Aug 2012 22:53:57 +0000 (15:53 -0700)]
Make git-svn branch patterns match complete URL

When using the {word,[...]} style of configuration for tags and branches,
it appears the intent is to only match whole path parts, since the words
in the {} pattern are meta-character quoted.

When the pattern word appears in the beginning or middle of the url,
it's matched completely, since the left side, pattern, and (non-empty)
right side are joined together with path separators.

However, when the pattern word appears at the end of the URL, the
right side is an empty pattern, and the resulting regex matches
more than just the specified pattern.

For example, if you specify something along the lines of

    branches = branches/project/{release_1,release_2}

and your repository also contains "branches/project/release_1_2", you
will also get the release_1_2 branch.  By restricting the match regex
with anchors, this is avoided.

Signed-off-by: Ammon Riley <ammon.riley@gmail.com>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agot9164: Add missing quotes in test
Robert Luberda [Sun, 19 Aug 2012 22:43:19 +0000 (00:43 +0200)]
t9164: Add missing quotes in test

This fixes `ambiguous redirect' error given by bash.

[ew: fix misspelled test name,
     also eliminate space after ">>" to conform to guidelines]

Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agogit-svn.perl: keep processing all commits in parents_exclude
Steven Walter [Mon, 20 Aug 2012 01:39:41 +0000 (21:39 -0400)]
git-svn.perl: keep processing all commits in parents_exclude

This fixes a bug where git finds the incorrect merge parent.  Consider a
repository with trunk, branch1 of trunk, and branch2 of branch1.
Without this change, git interprets a merge of branch2 into trunk as a
merge of branch1 into trunk.

Signed-off-by: Steven Walter <stevenrwalter@gmail.com>
Reviewed-by: Sam Vilain <sam@vilain.net>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agogit-svn.perl: consider all ranges for a given merge, instead of only tip-by-tip
Steven Walter [Mon, 20 Aug 2012 01:39:40 +0000 (21:39 -0400)]
git-svn.perl: consider all ranges for a given merge, instead of only tip-by-tip

Consider the case where you have trunk, branch1 of trunk, and branch2 of
branch1.  trunk is merged back into branch2, and then branch2 is
reintegrated into trunk.  The merge of branch2 into trunk will have
svn:mergeinfo property references to both branch1 and branch2.  When
git-svn fetches the commit that merges branch2 (check_cherry_pick),
it is necessary to eliminate the merged contents of branch1 as well as
branch2, or else the merge will be incorrectly ignored as a cherry-pick.

Signed-off-by: Steven Walter <stevenrwalter@gmail.com>
Reviewed-by: Sam Vilain <sam@vilain.net>
Signed-off-by: Eric Wong <normalperson@yhbt.net>
8 years agoattr: avoid searching for basename on every match
Nguyễn Thái Ngọc Duy [Fri, 5 Oct 2012 04:41:02 +0000 (11:41 +0700)]
attr: avoid searching for basename on every match

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoattr: avoid strlen() on every match
Nguyễn Thái Ngọc Duy [Fri, 5 Oct 2012 04:41:01 +0000 (11:41 +0700)]
attr: avoid strlen() on every match

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agoupload-pack: use peel_ref for ref advertisements
Jeff King [Thu, 4 Oct 2012 08:03:33 +0000 (04:03 -0400)]
upload-pack: use peel_ref for ref advertisements

When upload-pack advertises refs, we attempt to peel tags
and advertise the peeled version. We currently hand-roll the
tag dereferencing, and use as many optimizations as we can
to avoid loading non-tag objects into memory.

Not only has peel_ref recently learned these optimizations,
too, but it also contains an even more important one: it
has access to the "peeled" data from the pack-refs file.
That means we can avoid not only loading annotated tags
entirely, but also avoid doing any kind of object lookup at
all.

This cut the CPU time to advertise refs by 50% in the
linux-2.6 repo, as measured by:

  echo 0000 | git-upload-pack . >/dev/null

best-of-five, warm cache, objects and refs fully packed:

  [before]             [after]
  real    0m0.026s     real    0m0.013s
  user    0m0.024s     user    0m0.008s
  sys     0m0.000s     sys     0m0.000s

Those numbers are irrelevantly small compared to an actual
fetch. Here's a larger repo (400K refs, of which 12K are
unique, and of which only 107 are unique annotated tags):

  [before]             [after]
  real    0m0.704s     real    0m0.596s
  user    0m0.600s     user    0m0.496s
  sys     0m0.096s     sys     0m0.092s

This shows only a 15% speedup (mostly because it has fewer
actual tags to parse), but a larger absolute value (100ms,
which isn't a lot compared to a real fetch, but this
advertisement happens on every fetch, even if the client is
just finding out they are completely up to date).

In truly pathological cases, where you have a large number
of unique annotated tags, it can make an even bigger
difference. Here are the numbers for a linux-2.6 repository
that has had every seventh commit tagged (so about 50K
tags):

  [before]             [after]
  real    0m0.443s     real    0m0.097s
  user    0m0.416s     user    0m0.080s
  sys     0m0.024s     sys     0m0.012s

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agopeel_ref: check object type before loading
Jeff King [Thu, 4 Oct 2012 08:02:53 +0000 (04:02 -0400)]
peel_ref: check object type before loading

The point of peel_ref is to dereference tags; if the base
object is not a tag, then we can return early without even
loading the object into memory.

This patch accomplishes that by checking sha1_object_info
for the type. For a packed object, we can get away with just
looking in the pack index. For a loose object, we only need
to inflate the first couple of header bytes.

This is a bit of a gamble; if we do find a tag object, then
we will end up loading the content anyway, and the extra
lookup will have been wasteful. However, if it is not a tag
object, then we save loading the object entirely. Depending
on the ratio of non-tags to tags in the input, this can be a
minor win or minor loss.

However, it does give us one potential major win: if a ref
points to a large blob (e.g., via an unannotated tag), then
we can avoid looking at it entirely.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agopeel_ref: do not return a null sha1
Jeff King [Thu, 4 Oct 2012 08:00:19 +0000 (04:00 -0400)]
peel_ref: do not return a null sha1

The idea of the peel_ref function is to dereference tag
objects recursively until we hit a non-tag, and return the
sha1. Conceptually, it should return 0 if it is successful
(and fill in the sha1), or -1 if there was nothing to peel.

However, the current behavior is much more confusing. For a
regular loose ref, the behavior is as described above. But
there is an optimization to reuse the peeled-ref value for a
ref that came from a packed-refs file. If we have such a
ref, we return its peeled value, even if that peeled value
is null (indicating that we know the ref definitely does
_not_ peel).

It might seem like such information is useful to the caller,
who would then know not to bother loading and trying to peel
the object. Except that they should not bother loading and
trying to peel the object _anyway_, because that fallback is
already handled by peel_ref. In other words, the whole point
of calling this function is that it handles those details
internally, and you either get a sha1, or you know that it
is not peel-able.

This patch catches the null sha1 case internally and
converts it into a -1 return value (i.e., there is nothing
to peel). This simplifies callers, which do not need to
bother checking themselves.

Two callers are worth noting:

  - in pack-objects, a comment indicates that there is a
    difference between non-peelable tags and unannotated
    tags. But that is not the case (before or after this
    patch). Whether you get a null sha1 has to do with
    internal details of how peel_ref operated.

  - in show-ref, if peel_ref returns a failure, the caller
    tries to decide whether to try peeling manually based on
    whether the REF_ISPACKED flag is set. But this doesn't
    make any sense. If the flag is set, that does not
    necessarily mean the ref came from a packed-refs file
    with the "peeled" extension. But it doesn't matter,
    because even if it didn't, there's no point in trying to
    peel it ourselves, as peel_ref would already have done
    so. In other words, the fallback peeling is guaranteed
    to fail.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agopeel_ref: use faster deref_tag_noverify
Jeff King [Thu, 4 Oct 2012 07:58:15 +0000 (03:58 -0400)]
peel_ref: use faster deref_tag_noverify

When we are asked to peel a ref to a sha1, we internally call
deref_tag, which will recursively parse each tagged object
until we reach a non-tag. This has the benefit that we will
verify our ability to load and parse the pointed-to object.

However, there is a performance downside: we may not need to
load that object at all (e.g., if we are listing peeled
simply listing peeled refs), or it may be a large object
that should follow a streaming code path (e.g., an annotated
tag of a large blob).

It makes more sense for peel_ref to choose the fast thing
rather than performing the extra check, for two reasons:

  1. We will already sometimes short-circuit the tag parsing
     in favor of a peeled entry from a packed-refs file. So
     we are already favoring speed in some cases, and it is
     not wise for a caller to rely on peel_ref to detect
     corruption.

  2. We already silently ignore much larger corruptions,
     like a ref that points to a non-existent object, or a
     tag object that exists but is corrupted.

  2. peel_ref is not the right place to check for such a
     database corruption. It is returning only the sha1
     anyway, not the actual object. Any callers which use
     that sha1 to load an object will soon discover the
     corruption anyway, so we are really just pushing back
     the discovery to later in the program.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
8 years agogitcli: parse-options lets you omit tail of long options
Junio C Hamano [Thu, 4 Oct 2012 17:13:49 +0000 (10:13 -0700)]
gitcli: parse-options lets you omit tail of long options

Describe the behaviour, but do warn people against taking it too
literally and expect an abbreviation valid today will stay valid
forever.

Signed-off-by: Junio C Hamano <gitster@pobox.com>