Code

git.git
13 years agoMerge branch 'mg/doc-bisect-tweak-worktree'
Junio C Hamano [Wed, 23 Mar 2011 04:38:15 +0000 (21:38 -0700)]
Merge branch 'mg/doc-bisect-tweak-worktree'

* mg/doc-bisect-tweak-worktree:
  git-bisect.txt: example for bisecting with hot-fix
  git-bisect.txt: streamline run presentation

13 years agoMerge branch 'jn/test-sanitize-git-env'
Junio C Hamano [Wed, 23 Mar 2011 04:38:12 +0000 (21:38 -0700)]
Merge branch 'jn/test-sanitize-git-env'

* jn/test-sanitize-git-env:
  tests: scrub environment of GIT_* variables
  config: drop support for GIT_CONFIG_NOGLOBAL
  gitattributes: drop support for GIT_ATTR_NOGLOBAL
  tests: suppress system gitattributes
  tests: stop worrying about obsolete environment variables

13 years agoMerge branch 'sp/maint-upload-pack-stop-early'
Junio C Hamano [Wed, 23 Mar 2011 04:38:06 +0000 (21:38 -0700)]
Merge branch 'sp/maint-upload-pack-stop-early'

* sp/maint-upload-pack-stop-early:
  upload-pack: Implement no-done capability
  upload-pack: More aggressively send 'ACK %s ready'

13 years agoMerge branch 'sp/maint-fetch-pack-stop-early'
Junio C Hamano [Wed, 23 Mar 2011 04:38:03 +0000 (21:38 -0700)]
Merge branch 'sp/maint-fetch-pack-stop-early'

* sp/maint-fetch-pack-stop-early:
  fetch-pack: Implement no-done capability
  fetch-pack: Finish negotation if remote replies "ACK %s ready"

13 years agoMerge branch 'jc/maint-rev-list-culled-boundary'
Junio C Hamano [Wed, 23 Mar 2011 04:37:59 +0000 (21:37 -0700)]
Merge branch 'jc/maint-rev-list-culled-boundary'

* jc/maint-rev-list-culled-boundary:
  list-objects.c: don't add an unparsed NULL as a pending tree

Conflicts:
list-objects.c

13 years agoMerge branch 'jc/maint-fetch-alt'
Junio C Hamano [Wed, 23 Mar 2011 04:37:53 +0000 (21:37 -0700)]
Merge branch 'jc/maint-fetch-alt'

* jc/maint-fetch-alt:
  fetch-pack: objects in our alternates are available to us
  refs_from_alternate: helper to use refs from alternates

Conflicts:
builtin/receive-pack.c

13 years agoMerge branch 'sg/complete-symmetric-diff'
Junio C Hamano [Wed, 23 Mar 2011 04:37:47 +0000 (21:37 -0700)]
Merge branch 'sg/complete-symmetric-diff'

* sg/complete-symmetric-diff:
  bash: complete 'git diff ...branc<TAB>'
  bash: fix misindented esac statement in __git_complete_file()

13 years agoMerge branch 'svn-fe' of git://repo.or.cz/git/jrn
Junio C Hamano [Wed, 23 Mar 2011 03:51:07 +0000 (20:51 -0700)]
Merge branch 'svn-fe' of git://repo.or.cz/git/jrn

* 'svn-fe' of git://repo.or.cz/git/jrn:
  vcs-svn: use strchr to find RFC822 delimiter
  vcs-svn: implement perfect hash for top-level keys
  vcs-svn: implement perfect hash for node-prop keys
  vcs-svn: use strbuf for author, UUID, and URL
  vcs-svn: use strbuf for revision log
  vcs-svn: improve reporting of input errors
  vcs-svn: make buffer_copy_bytes return length read
  vcs-svn: make buffer_skip_bytes return length read
  vcs-svn: improve support for reading large files
  vcs-svn: allow input errors to be detected promptly
  vcs-svn: simplify repo_modify_path and repo_copy
  vcs-svn: handle_node: use repo_read_path
  vcs-svn: introduce repo_read_path to check the content at a path

13 years agoMerge branch 'db/length-as-hash' into svn-fe
Jonathan Nieder [Tue, 22 Mar 2011 23:44:49 +0000 (18:44 -0500)]
Merge branch 'db/length-as-hash' into svn-fe

* db/length-as-hash:
  vcs-svn: use strchr to find RFC822 delimiter
  vcs-svn: implement perfect hash for top-level keys
  vcs-svn: implement perfect hash for node-prop keys

Conflicts:
vcs-svn/svndump.c

13 years agovcs-svn: use strchr to find RFC822 delimiter
David Barr [Tue, 14 Dec 2010 00:06:43 +0000 (11:06 +1100)]
vcs-svn: use strchr to find RFC822 delimiter

This is a small optimisation (4% reduction in user time) but is the
largest artifact within the parsing portion of svndump.c

Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: implement perfect hash for top-level keys
David Barr [Mon, 13 Dec 2010 08:56:01 +0000 (19:56 +1100)]
vcs-svn: implement perfect hash for top-level keys

Instead of interning property names and comparing their string_pool
keys, look them up in a table by string length, which should be about
as fast.

Another small step towards removing dependence on string_pool
altogether.

Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: implement perfect hash for node-prop keys
David Barr [Mon, 13 Dec 2010 08:13:24 +0000 (19:13 +1100)]
vcs-svn: implement perfect hash for node-prop keys

Instead of interning property names and comparing their string_pool
keys, look them up in a table by string length, which should be about
as fast.

This is a small step towards removing dependence on string_pool.

Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: use strbuf for author, UUID, and URL
David Barr [Tue, 22 Mar 2011 22:52:17 +0000 (17:52 -0500)]
vcs-svn: use strbuf for author, UUID, and URL

Use strbufs and strings instead of interned strings for values of rev,
dump, and node fields that happen to be strings.  After this change,
the only remaining string_pool use is for paths in the repo_tree API
and internals.

Functional change: treat an empty author, UUID, or URL as none at all.
So for example, in repos where the first revision has an empty
svn:author property, the first rev will be treated as by "nobody"
rather than by a person with empty name and email address created by
prepending an @ sign to the repository UUID.

Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agoMerge branch 'maint'
Junio C Hamano [Tue, 22 Mar 2011 21:56:13 +0000 (14:56 -0700)]
Merge branch 'maint'

* maint:
  git-am.txt: advertise 'git am --abort' instead of 'rm .git/rebase-apply'
  bisect: visualize with git-log if gitk is unavailable

13 years agovcs-svn: use strbuf for revision log
David Barr [Mon, 21 Mar 2011 23:49:50 +0000 (10:49 +1100)]
vcs-svn: use strbuf for revision log

obj_pool is overkill for this application: all that is needed is a
buffer that can resize from rev to rev to accomodate differently-sized
strings.  In the spirit of commit deadcef4 (2010-11-06), use a strbuf
instead.

This is a small step towards removing dependence on obj_pool.h.

Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: improve reporting of input errors
Jonathan Nieder [Tue, 28 Dec 2010 10:30:54 +0000 (04:30 -0600)]
vcs-svn: improve reporting of input errors

Catch input errors and exit early enough to print a reasonable
diagnosis based on errno.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: make buffer_copy_bytes return length read
Jonathan Nieder [Tue, 28 Dec 2010 10:26:17 +0000 (04:26 -0600)]
vcs-svn: make buffer_copy_bytes return length read

Currently buffer_copy_bytes does not report to its caller whether
it encountered an early end of file.

Add a return value representing the number of bytes read (but not
the number of bytes copied).  This way all three unusual conditions
can be distinguished: input error with buffer_ferror, output error
with ferror(outfile), early end of input by checking the return
value.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: make buffer_skip_bytes return length read
Jonathan Nieder [Mon, 11 Oct 2010 02:44:21 +0000 (21:44 -0500)]
vcs-svn: make buffer_skip_bytes return length read

Currently there is no way to detect when input ended if it ended
early during buffer_skip_bytes.  Tell the calling program how many
bytes were actually skipped for easier debugging.

Existing callers will still ignore early EOF.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agovcs-svn: improve support for reading large files
Jonathan Nieder [Mon, 11 Oct 2010 02:46:24 +0000 (21:46 -0500)]
vcs-svn: improve support for reading large files

Move from uint32_t to off_t as the fundamental unit of length used by
the line_buffer library.  Performance would get worse if anything but
I think it's worth it for support of deltas that need to skip large
pieces (> 4 GiB).

Exception: buffer_read_string still takes a uint32_t, since it keeps
its result in an in-core obj_pool.

Callers still have to be updated to take advantage of this.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: David Barr <david.barr@cordelta.com>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
13 years agogit-am.txt: advertise 'git am --abort' instead of 'rm .git/rebase-apply'
SZEDER Gábor [Tue, 22 Mar 2011 18:42:25 +0000 (19:42 +0100)]
git-am.txt: advertise 'git am --abort' instead of 'rm .git/rebase-apply'

'git am --abort' is around for quite a long time now, and users should
normally not poke around inside the .git directory, yet the
documentation of 'git am' still recommends the following:

  ... if you decide to start over from scratch,
  run `rm -f -r .git/rebase-apply` ...

Suggest 'git am --abort' instead.

It's not quite the same as the original, because 'git am --abort' will
restore the original branch, while simply removing '.git/rebase-apply'
won't, but that's rather a thinko in the original wording, because
that won't actually "start over _from scratch_".

Signed-off-by: SZEDER Gábor <szeder@ira.uka.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoFix sparse warnings
Stephen Boyd [Tue, 22 Mar 2011 07:51:05 +0000 (00:51 -0700)]
Fix sparse warnings

Fix warnings from 'make check'.

 - These files don't include 'builtin.h' causing sparse to complain that
   cmd_* isn't declared:

   builtin/clone.c:364, builtin/fetch-pack.c:797,
   builtin/fmt-merge-msg.c:34, builtin/hash-object.c:78,
   builtin/merge-index.c:69, builtin/merge-recursive.c:22
   builtin/merge-tree.c:341, builtin/mktag.c:156, builtin/notes.c:426
   builtin/notes.c:822, builtin/pack-redundant.c:596,
   builtin/pack-refs.c:10, builtin/patch-id.c:60, builtin/patch-id.c:149,
   builtin/remote.c:1512, builtin/remote-ext.c:240,
   builtin/remote-fd.c:53, builtin/reset.c:236, builtin/send-pack.c:384,
   builtin/unpack-file.c:25, builtin/var.c:75

 - These files have symbols which should be marked static since they're
   only file scope:

   submodule.c:12, diff.c:631, replace_object.c:92, submodule.c:13,
   submodule.c:14, trace.c:78, transport.c:195, transport-helper.c:79,
   unpack-trees.c:19, url.c:3, url.c:18, url.c:104, url.c:117, url.c:123,
   url.c:129, url.c:136, thread-utils.c:21, thread-utils.c:48

 - These files redeclare symbols to be different types:

   builtin/index-pack.c:210, parse-options.c:564, parse-options.c:571,
   usage.c:49, usage.c:58, usage.c:63, usage.c:72

 - These files use a literal integer 0 when they really should use a NULL
   pointer:

   daemon.c:663, fast-import.c:2942, imap-send.c:1072, notes-merge.c:362

While we're in the area, clean up some unused #includes in builtin files
(mostly exec_cmd.h).

Signed-off-by: Stephen Boyd <bebarino@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agobisect: visualize with git-log if gitk is unavailable
Jeff King [Mon, 21 Mar 2011 13:14:22 +0000 (09:14 -0400)]
bisect: visualize with git-log if gitk is unavailable

If gitk is not available in the PATH, bisect ends up
exiting with the shell's 127 error code, confusing the git
wrapper into thinking that bisect is not a git command.

We already fallback to git-log if there doesn't seem to be a
graphical display available. We should do the same if gitk
is not available in our PATH at all. This not only fixes the
ugly error message, but is a much more sensible default than
failing to show the user anything.

Reported by Maxin John.

Tested-by: Maxin B. John <maxin@maxinbjohn.info>
Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoUpdate drart release notes to 1.7.5
Junio C Hamano [Mon, 21 Mar 2011 05:24:24 +0000 (22:24 -0700)]
Update drart release notes to 1.7.5

Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'maint'
Junio C Hamano [Mon, 21 Mar 2011 05:14:47 +0000 (22:14 -0700)]
Merge branch 'maint'

* maint:
  Update draft release notes to 1.7.4.2
  Work around broken ln on solaris as used in t8006
  t/README: Add a note about running commands under valgrind

13 years agoUpdate draft release notes to 1.7.4.2
Junio C Hamano [Mon, 21 Mar 2011 05:14:21 +0000 (22:14 -0700)]
Update draft release notes to 1.7.4.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'sp/maint-fd-limit' into maint
Junio C Hamano [Mon, 21 Mar 2011 05:11:46 +0000 (22:11 -0700)]
Merge branch 'sp/maint-fd-limit' into maint

* sp/maint-fd-limit:
  sha1_file.c: Don't retain open fds on small packs
  mingw: add minimum getrlimit() compatibility stub
  Limit file descriptors used by packs

13 years agoMerge branch 'mr/hpux' into maint
Junio C Hamano [Mon, 21 Mar 2011 05:11:15 +0000 (22:11 -0700)]
Merge branch 'mr/hpux' into maint

* mr/hpux:
  git-compat-util.h: Honor HP C's noreturn attribute
  Makefile: add NO_FNMATCH_CASEFOLD to HP-UX section

13 years agoMerge branch 'so/submodule-no-update-first-time' into maint
Junio C Hamano [Mon, 21 Mar 2011 05:11:02 +0000 (22:11 -0700)]
Merge branch 'so/submodule-no-update-first-time' into maint

* so/submodule-no-update-first-time:
  t7406: "git submodule update {--merge|--rebase]" with new submodules
  submodule: no [--merge|--rebase] when newly cloned

13 years agoMerge branch 'mo/perl-bidi-pipe-envfix' into maint
Junio C Hamano [Mon, 21 Mar 2011 05:10:31 +0000 (22:10 -0700)]
Merge branch 'mo/perl-bidi-pipe-envfix' into maint

* mo/perl-bidi-pipe-envfix:
  perl: command_bidi_pipe() method should set-up git environmens

13 years agoMerge branch 'ae/better-template-failure-report' into maint
Junio C Hamano [Mon, 21 Mar 2011 05:09:39 +0000 (22:09 -0700)]
Merge branch 'ae/better-template-failure-report' into maint

* ae/better-template-failure-report:
  Improve error messages when temporary file creation fails

13 years agoWork around broken ln on solaris as used in t8006
Ben Walton [Mon, 21 Mar 2011 01:12:26 +0000 (21:12 -0400)]
Work around broken ln on solaris as used in t8006

The test setup in t8006-blame-textconv.sh uses "ln -sf" to
overwrite an existing symlink.  Unfortunately, both /usr/bin/ln
and /usr/xpg4/bin/ln on solaris 9 don't properly handle -f and -s
used at the same time.  This caused the test setup and subsequent
checks to fail.

Instead, remove the symlink and then create a new one in the
setup code.

The upstream Solaris bug (fixed in 10, but not 9) is documented
here:

  http://bugs.opensolaris.org/view_bug.do?bug_id=4372462

Signed-off-by: Ben Walton <bwalton@artsci.utoronto.ca>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agot/README: Add a note about running commands under valgrind
Carlos Martín Nieto [Tue, 15 Mar 2011 09:32:11 +0000 (10:32 +0100)]
t/README: Add a note about running commands under valgrind

The test suite runs valgrind with certain options activated. Add a
note saying how to run commands under the same conditions as the test
suite does.

Signed-off-by: Carlos Martín Nieto <cmn@elego.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'mm/maint-log-n-with-diff-filtering'
Junio C Hamano [Sun, 20 Mar 2011 06:25:38 +0000 (23:25 -0700)]
Merge branch 'mm/maint-log-n-with-diff-filtering'

* mm/maint-log-n-with-diff-filtering:
  log: fix --max-count when used together with -S or -G

13 years agoMerge branch 'mg/rev-list-n-reverse-doc'
Junio C Hamano [Sun, 20 Mar 2011 06:25:32 +0000 (23:25 -0700)]
Merge branch 'mg/rev-list-n-reverse-doc'

* mg/rev-list-n-reverse-doc:
  git-log.txt,rev-list-options.txt: put option blocks in proper order
  git-log.txt,rev-list-options.txt: -n/--max-count is commit limiting

13 years agoMerge branch 'js/rerere-forget-always-take-pathspec'
Junio C Hamano [Sun, 20 Mar 2011 06:24:48 +0000 (23:24 -0700)]
Merge branch 'js/rerere-forget-always-take-pathspec'

* js/rerere-forget-always-take-pathspec:
  rerere forget: deprecate invocation without pathspec

13 years agoMerge branch 'ab/i18n-basic'
Junio C Hamano [Sun, 20 Mar 2011 06:24:42 +0000 (23:24 -0700)]
Merge branch 'ab/i18n-basic'

* ab/i18n-basic:
  i18n: "make distclean" should clean up after "make pot"
  i18n: Makefile: "pot" target to extract messages marked for translation
  i18n: add stub Q_() wrapper for ngettext
  i18n: do not poison translations unless GIT_GETTEXT_POISON envvar is set
  i18n: add GETTEXT_POISON to simulate unfriendly translator
  i18n: add no-op _() and N_() wrappers
  commit, status: use status_printf{,_ln,_more} helpers
  commit: refer to commit template as s->fp
  wt-status: add helpers for printing wt-status lines

Conflicts:
builtin/commit.c

13 years agoMerge branch 'uk/ls-remote-in-get-remote-url'
Junio C Hamano [Sun, 20 Mar 2011 06:24:34 +0000 (23:24 -0700)]
Merge branch 'uk/ls-remote-in-get-remote-url'

* uk/ls-remote-in-get-remote-url:
  git-request-pull: open-code the only invocation of get_remote_url
  get_remote_url(): use the same data source as ls-remote to get remote urls

13 years agoMerge branch 'ss/mergetool--lib'
Junio C Hamano [Sun, 20 Mar 2011 06:24:27 +0000 (23:24 -0700)]
Merge branch 'ss/mergetool--lib'

* ss/mergetool--lib:
  mergetool--lib: Add Beyond Compare 3 as a tool
  mergetool--lib: Sort tools alphabetically for easier lookup

13 years agoMerge branch 'jn/status-translatable'
Junio C Hamano [Sun, 20 Mar 2011 06:24:19 +0000 (23:24 -0700)]
Merge branch 'jn/status-translatable'

* jn/status-translatable:
  commit, status: use status_printf{,_ln,_more} helpers
  commit: refer to commit template as s->fp
  wt-status: add helpers for printing wt-status lines

Conflicts:
builtin/commit.c

13 years agoMerge branch 'jk/trace-sifter'
Junio C Hamano [Sun, 20 Mar 2011 06:24:12 +0000 (23:24 -0700)]
Merge branch 'jk/trace-sifter'

* jk/trace-sifter:
  trace: give repo_setup trace its own key
  add packet tracing debug code
  trace: add trace_strbuf
  trace: factor out "do we want to trace" logic
  trace: refactor to support multiple env variables
  trace: add trace_vprintf

13 years agoMerge branch 'jk/format-patch-multiline-header'
Junio C Hamano [Sun, 20 Mar 2011 06:24:08 +0000 (23:24 -0700)]
Merge branch 'jk/format-patch-multiline-header'

* jk/format-patch-multiline-header:
  format-patch: rfc2047-encode newlines in headers
  format-patch: wrap long header lines
  strbuf: add fixed-length version of add_wrapped_text

13 years agoMerge branch 'jk/merge-rename-ux'
Junio C Hamano [Sun, 20 Mar 2011 06:23:56 +0000 (23:23 -0700)]
Merge branch 'jk/merge-rename-ux'

* jk/merge-rename-ux:
  pull: propagate --progress to merge
  merge: enable progress reporting for rename detection
  add inexact rename detection progress infrastructure
  commit: stop setting rename limit
  bump rename limit defaults (again)
  merge: improve inexact rename limit warning

13 years agoMerge branch 'maint'
Junio C Hamano [Sun, 20 Mar 2011 06:21:10 +0000 (23:21 -0700)]
Merge branch 'maint'

* maint:
  gitweb: Always call parse_date with timezone parameter
  bisect: explain the rationale behind 125

13 years agogrep: read patterns from stdin with -f -
René Scharfe [Sat, 19 Mar 2011 18:33:15 +0000 (19:33 +0100)]
grep: read patterns from stdin with -f -

Support the well-know convention of reading standard input instead of a
named file if "-" (dash) is specified.  GNU grep does the same.

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agogitweb: Always call parse_date with timezone parameter
Jakub Narebski [Sat, 19 Mar 2011 22:53:55 +0000 (23:53 +0100)]
gitweb: Always call parse_date with timezone parameter

Timezone is required to correctly set local time, which would be needed
for future 'localtime' feature.

While at it, remove unnecessary call to the function from git_log_body,
as its return value is not used anywhere.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agobisect: explain the rationale behind 125
Junio C Hamano [Sun, 20 Mar 2011 04:46:06 +0000 (21:46 -0700)]
bisect: explain the rationale behind 125

Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'maint'
Junio C Hamano [Fri, 18 Mar 2011 23:20:49 +0000 (16:20 -0700)]
Merge branch 'maint'

* maint:
  docs: fix grammar in gitattributes.txt

13 years agodocs: fix grammar in gitattributes.txt
Alexei Sholik [Fri, 18 Mar 2011 13:14:27 +0000 (15:14 +0200)]
docs: fix grammar in gitattributes.txt

[jc: with a fixlet from Marc Branchaud]

Signed-off-by: Alexei Sholik <alcosholik@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agofetch-pack: objects in our alternates are available to us
Junio C Hamano [Fri, 11 Mar 2011 19:53:52 +0000 (11:53 -0800)]
fetch-pack: objects in our alternates are available to us

Use the helper function split from the receiving end of "git push" to
allow the same optimization on the receiving end of "git fetch".

Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
13 years agorefs_from_alternate: helper to use refs from alternates
Junio C Hamano [Fri, 11 Mar 2011 19:32:53 +0000 (11:32 -0800)]
refs_from_alternate: helper to use refs from alternates

The receiving end of "git push" advertises the objects that the repository
itself does not use, but are at the tips of refs in other repositories
whose object databases are used as alternates for it. This helps it avoid
having to receive (and the pusher having to send) objects that are already
available to the receiving repository via the alternates mechanism.

Tweak the helper function that implements this feature, and move it to
transport.[ch] for future reuse by other programs.

The additional test demonstrates how this optimization is helping "git push",
and "git fetch" is ignorant about it.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
13 years agoName make_*_path functions more accurately
Carlos Martín Nieto [Thu, 17 Mar 2011 11:26:46 +0000 (12:26 +0100)]
Name make_*_path functions more accurately

Rename the make_*_path functions so it's clearer what they do, in
particlar make clear what the differnce between make_absolute_path and
make_nonrelative_path is by renaming them real_path and absolute_path
respectively. make_relative_path has an understandable name and is
renamed to relative_path to maintain the name convention.

The function calls have been replaced 1-to-1 in their usage.

Signed-off-by: Carlos Martín Nieto <cmn@elego.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoRelNotes/1.7.5.txt: typo and language fixes
Michael J Gruber [Wed, 16 Mar 2011 09:23:23 +0000 (10:23 +0100)]
RelNotes/1.7.5.txt: typo and language fixes

Also remove entries for fixes that are already present in the
maintenance track.

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'maint'
Junio C Hamano [Wed, 16 Mar 2011 23:59:30 +0000 (16:59 -0700)]
Merge branch 'maint'

* maint:
  Prepare draft release notes to 1.7.4.2
  gitweb: highlight: replace tabs with spaces
  make_absolute_path: return the input path if it points to our buffer
  valgrind: ignore SSE-based strlen invalid reads
  diff --submodule: split into bite-sized pieces
  cherry: split off function to print output lines
  branch: split off function that writes tracking info and commit subject
  standardize brace placement in struct definitions
  compat: make gcc bswap an inline function
  enums: omit trailing comma for portability

Conflicts:
RelNotes

13 years agoPrepare draft release notes to 1.7.4.2
Junio C Hamano [Wed, 16 Mar 2011 23:29:03 +0000 (16:29 -0700)]
Prepare draft release notes to 1.7.4.2

Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'jn/maint-commit-missing-template' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'jn/maint-commit-missing-template' into maint

* jn/maint-commit-missing-template:
  commit: error out for missing commit message template

13 years agoMerge branch 'lt/rename-no-extra-copy-detection' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'lt/rename-no-extra-copy-detection' into maint

* lt/rename-no-extra-copy-detection:
  diffcore-rename: improve estimate_similarity() heuristics
  diffcore-rename: properly honor the difference between -M and -C
  for_each_hash: allow passing a 'void *data' pointer to callback

13 years agoMerge branch 'jk/diffstat-binary' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'jk/diffstat-binary' into maint

* jk/diffstat-binary:
  diff: don't retrieve binary blobs for diffstat
  diff: handle diffstat of rewritten binary files

13 years agoMerge branch 'mg/maint-difftool-vim-readonly' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'mg/maint-difftool-vim-readonly' into maint

* mg/maint-difftool-vim-readonly:
  mergetool-lib: call vim in readonly mode for diffs

13 years agoMerge branch 'jn/test-terminal-punt-on-osx-breakage' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'jn/test-terminal-punt-on-osx-breakage' into maint

* jn/test-terminal-punt-on-osx-breakage:
  tests: skip terminal output tests on OS X

13 years agoMerge branch 'jk/fail-null-clone' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'jk/fail-null-clone' into maint

* jk/fail-null-clone:
  clone: die when trying to clone missing local path

13 years agoMerge branch 'jh/push-default-upstream-configname' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:26 +0000 (16:47 -0700)]
Merge branch 'jh/push-default-upstream-configname' into maint

* jh/push-default-upstream-configname:
  push.default: Rename 'tracking' to 'upstream'

13 years agoMerge branch 'mg/placeholders-are-lowercase' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:25 +0000 (16:47 -0700)]
Merge branch 'mg/placeholders-are-lowercase' into maint

* mg/placeholders-are-lowercase:
  Make <identifier> lowercase in Documentation
  Make <identifier> lowercase as per CodingGuidelines
  Make <identifier> lowercase as per CodingGuidelines
  Make <identifier> lowercase as per CodingGuidelines
  CodingGuidelines: downcase placeholders in usage messages

13 years agoMerge branch 'mg/patch-id' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:25 +0000 (16:47 -0700)]
Merge branch 'mg/patch-id' into maint

* mg/patch-id:
  git-patch-id: do not trip over "no newline" markers
  git-patch-id: test for "no newline" markers

13 years agoMerge branch 'js/maint-merge-use-prepare-commit-msg-hook' into maint
Junio C Hamano [Wed, 16 Mar 2011 23:47:25 +0000 (16:47 -0700)]
Merge branch 'js/maint-merge-use-prepare-commit-msg-hook' into maint

* js/maint-merge-use-prepare-commit-msg-hook:
  merge: honor prepare-commit-msg hook

13 years agogitweb: highlight: replace tabs with spaces
Kevin Cernekee [Wed, 16 Mar 2011 22:34:13 +0000 (15:34 -0700)]
gitweb: highlight: replace tabs with spaces

Consider the following code fragment:

        /*
         * test
         */

vim ":set list" mode shows that the first character on each line is a
tab:

^I/*$
^I * test$
^I */$

By default, the "highlight" program will retain the tabs in the HTML
output:

$ highlight --fragment --syntax c test.c
        <span class="hl com">/*</span>
<span class="hl com">    * test</span>
<span class="hl com">    */</span>

vim list mode:

^I<span class="hl com">/*</span>$
<span class="hl com">^I * test</span>$
<span class="hl com">^I */</span>$

In gitweb, this winds up looking something like:

   1         /*
   2     * test
   3     */

I tried both Firefox and Opera and saw the same behavior.

The desired output is:

   1         /*
   2          * test
   3          */

This can be accomplished by specifying "--replace-tabs=8" on the
highlight command line.

Signed-off-by: Kevin Cernekee <cernekee@gmail.com>
Acked-by: John 'Warthog9' Hawley <warthog9@eaglescrag.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agomake_absolute_path: return the input path if it points to our buffer
Carlos Martín Nieto [Wed, 16 Mar 2011 16:06:17 +0000 (17:06 +0100)]
make_absolute_path: return the input path if it points to our buffer

Some codepaths call make_absolute_path with its own return value as
input. In such a cases, return the path immediately.

This fixes a valgrind-discovered error, whereby we tried to copy a
string onto itself.

Signed-off-by: Carlos Martín Nieto <cmn@elego.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agovalgrind: ignore SSE-based strlen invalid reads
Carlos Martín Nieto [Wed, 16 Mar 2011 11:46:01 +0000 (12:46 +0100)]
valgrind: ignore SSE-based strlen invalid reads

Some versions of strlen use SSE to speed up the calculation and load 4
bytes at a time, even if it means reading past the end of the
allocated memory. This read is safe and when the strlen function is
inlined, it is not replaced by valgrind, which reports a
false-possitive.

Tell valgrind to ignore this particular error, as the read is, in
fact, safe. Current upstream-released version 3.6.1 is affected. Some
distributions have this fixed in their latest versions.

Signed-off-by: Carlos Martín Nieto <cmn@elego.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agovcs-svn: remove spurious semicolons
Jonathan Nieder [Wed, 16 Mar 2011 07:02:42 +0000 (02:02 -0500)]
vcs-svn: remove spurious semicolons

trp_gen is not a statement or function call, so it should not be
followed with a semicolon.  Noticed by gcc -pedantic.

 vcs-svn/repo_tree.c:41:81: warning: ISO C does not allow extra ';'
  outside of a function [-pedantic]

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agodiff --submodule: split into bite-sized pieces
Jonathan Nieder [Wed, 16 Mar 2011 07:14:11 +0000 (02:14 -0500)]
diff --submodule: split into bite-sized pieces

Introduce two functions:

 - prepare_submodule_summary prepares the revision walker
   to list changes in a submodule.  That is, it:

   * finds merge bases between the commits pointed to this
     path from before ("left") and after ("right") the change;
   * checks whether this is a fast-forward or fast-backward;
   * prepares a revision walk to list commits in the symmetric
     difference between the commits at each endpoint.

   It returns nonzero on error.

 - print_submodule_summary runs the revision walk and saves
   the result to a strbuf in --left-right format.

The goal is just readability.  No functional change intended.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Acked-by: Jens Lehmann <Jens.Lehmann@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agocherry: split off function to print output lines
Jonathan Nieder [Wed, 16 Mar 2011 07:12:48 +0000 (02:12 -0500)]
cherry: split off function to print output lines

Readers uninterested in the details of "git cherry"'s output format
can see

print_commit('-', commit, verbose, abbrev);

and ignore the details.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agobranch: split off function that writes tracking info and commit subject
Jonathan Nieder [Wed, 16 Mar 2011 07:10:14 +0000 (02:10 -0500)]
branch: split off function that writes tracking info and commit subject

Introduce a add_verbose_info function that takes care of adding

 - an abbreviated object name;
 - a summary of the form [ahead x, behind y] of the relationship
   to the corresponding upstream branch;
 - a one line commit subject

for the tip commit of a branch, for use in "git branch -v" output.

No functional change intended.  This just unindents the code a little
and makes it easier to skip on first reading.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agostandardize brace placement in struct definitions
Jonathan Nieder [Wed, 16 Mar 2011 07:08:34 +0000 (02:08 -0500)]
standardize brace placement in struct definitions

In a struct definitions, unlike functions, the prevailing style is for
the opening brace to go on the same line as the struct name, like so:

 struct foo {
int bar;
char *baz;
 };

Indeed, grepping for 'struct [a-z_]* {$' yields about 5 times as many
matches as 'struct [a-z_]*$'.

Linus sayeth:

 Heretic people all over the world have claimed that this inconsistency
 is ...  well ...  inconsistent, but all right-thinking people know that
 (a) K&R are _right_ and (b) K&R are right.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agocompat: make gcc bswap an inline function
Jonathan Nieder [Wed, 16 Mar 2011 07:00:49 +0000 (02:00 -0500)]
compat: make gcc bswap an inline function

Without this change, gcc -pedantic warns:

 cache.h: In function 'ce_to_dtype':
 cache.h:270:21: warning: ISO C forbids braced-groups within expressions [-pedantic]

An inline function is more readable anyway.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoenums: omit trailing comma for portability
Jonathan Nieder [Wed, 16 Mar 2011 06:59:10 +0000 (01:59 -0500)]
enums: omit trailing comma for portability

Since v1.7.2-rc0~23^2~2 (Add per-repository eol normalization,
2010-05-19), building with gcc -std=gnu89 -pedantic produces warnings
like the following:

 convert.c:21:11: warning: comma at end of enumerator list [-pedantic]

gcc is right to complain --- these commas are not permitted in C89.
In the spirit of v1.7.2-rc0~32^2~16 (2010-05-14), remove them.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'nd/struct-pathspec'
Junio C Hamano [Wed, 16 Mar 2011 07:17:05 +0000 (00:17 -0700)]
Merge branch 'nd/struct-pathspec'

* nd/struct-pathspec:
  declare 1-bit bitfields to be unsigned

13 years agodeclare 1-bit bitfields to be unsigned
Jonathan Nieder [Wed, 16 Mar 2011 03:42:32 +0000 (22:42 -0500)]
declare 1-bit bitfields to be unsigned

As "gcc -pedantic" notices, a two's complement 1-bit signed integer
cannot represent the value '1'.

 dir.c: In function 'init_pathspec':
 dir.c:1291:4: warning: overflow in implicit constant conversion [-Woverflow]

In the spirit of v1.7.1-rc1~10 (2010-04-06), 'unsigned' is what was
intended, so let's make the flags unsigned.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agogit-bisect.txt: example for bisecting with hot-fix
Michael J Gruber [Tue, 15 Mar 2011 21:24:56 +0000 (22:24 +0100)]
git-bisect.txt: example for bisecting with hot-fix

Give an example on how to bisect when older revisions need a hot-fix to
build, run or test. Triggered by the binutils/kernel issue at

http://thread.gmane.org/gmane.comp.gnu.binutils/52601/focus=1112779

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agogit-bisect.txt: streamline run presentation
Michael J Gruber [Tue, 15 Mar 2011 21:24:55 +0000 (22:24 +0100)]
git-bisect.txt: streamline run presentation

Streamline the presentation of "bisect run" by removing one example
which does not introduce new concepts.

Signed-off-by: Michael J Gruber <git@drmicha.warpmail.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoUpdate draft release notes to 1.7.5
Junio C Hamano [Tue, 15 Mar 2011 22:17:54 +0000 (15:17 -0700)]
Update draft release notes to 1.7.5

Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'jk/strbuf-vaddf'
Junio C Hamano [Tue, 15 Mar 2011 21:23:04 +0000 (14:23 -0700)]
Merge branch 'jk/strbuf-vaddf'

* jk/strbuf-vaddf:
  compat: fall back on __va_copy if available
  strbuf: add strbuf_vaddf
  compat: provide a fallback va_copy definition

13 years agoMerge branch 'mm/push-default-advice'
Junio C Hamano [Tue, 15 Mar 2011 21:22:55 +0000 (14:22 -0700)]
Merge branch 'mm/push-default-advice'

* mm/push-default-advice:
  push: better error message when no remote configured
  push: better error messages when push.default = tracking

13 years agoMerge branch 'jn/maint-instaweb-plack-fix'
Junio C Hamano [Tue, 15 Mar 2011 21:22:37 +0000 (14:22 -0700)]
Merge branch 'jn/maint-instaweb-plack-fix'

* jn/maint-instaweb-plack-fix:
  git-instaweb: Change how gitweb.psgi is made runnable as standalone app

13 years agoMerge branch 'sp/maint-fd-limit'
Junio C Hamano [Tue, 15 Mar 2011 21:22:23 +0000 (14:22 -0700)]
Merge branch 'sp/maint-fd-limit'

* sp/maint-fd-limit:
  sha1_file.c: Don't retain open fds on small packs
  mingw: add minimum getrlimit() compatibility stub
  Limit file descriptors used by packs

13 years agoMerge branch 'jc/checkout-orphan-warning'
Junio C Hamano [Tue, 15 Mar 2011 21:22:16 +0000 (14:22 -0700)]
Merge branch 'jc/checkout-orphan-warning'

* jc/checkout-orphan-warning:
  commit: give final warning when reattaching HEAD to leave commits behind

13 years agoMerge branch 'jh/maint-do-not-track-non-branches'
Junio C Hamano [Tue, 15 Mar 2011 21:22:13 +0000 (14:22 -0700)]
Merge branch 'jh/maint-do-not-track-non-branches'

* jh/maint-do-not-track-non-branches:
  branch/checkout --track: Ensure that upstream branch is indeed a branch

13 years agoMerge branch 'fk/maint-cvsimport-early-failure'
Junio C Hamano [Tue, 15 Mar 2011 21:22:09 +0000 (14:22 -0700)]
Merge branch 'fk/maint-cvsimport-early-failure'

* fk/maint-cvsimport-early-failure:
  git-cvsimport.perl: Bail out right away when reading from the server fails

13 years agoMerge branch 'jc/maint-apply-report-offset'
Junio C Hamano [Tue, 15 Mar 2011 21:22:02 +0000 (14:22 -0700)]
Merge branch 'jc/maint-apply-report-offset'

* jc/maint-apply-report-offset:
  apply -v: show offset count when patch did not apply exactly

13 years agoMerge branch 'jc/maint-apply-no-double-patch'
Junio C Hamano [Tue, 15 Mar 2011 21:22:00 +0000 (14:22 -0700)]
Merge branch 'jc/maint-apply-no-double-patch'

* jc/maint-apply-no-double-patch:
  apply: do not patch lines that were already patched

13 years agoMerge branch 'mr/hpux'
Junio C Hamano [Tue, 15 Mar 2011 21:21:47 +0000 (14:21 -0700)]
Merge branch 'mr/hpux'

* mr/hpux:
  git-compat-util.h: Honor HP C's noreturn attribute
  Makefile: add NO_FNMATCH_CASEFOLD to HP-UX section

13 years agoDocumentation: running test with --debug keeps "trash" directory
Piotr Krukowiecki [Tue, 15 Mar 2011 19:58:14 +0000 (20:58 +0100)]
Documentation: running test with --debug keeps "trash" directory

Signed-off-by: Piotr Krukowiecki <piotr.krukowiecki@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoMerge branch 'jk/doc-credits' of git://github.com/peff/git
Junio C Hamano [Tue, 15 Mar 2011 20:32:09 +0000 (13:32 -0700)]
Merge branch 'jk/doc-credits' of git://github.com/peff/git

* 'jk/doc-credits' of git://github.com/peff/git:
  docs: point git.txt author credits to git-scm.com
  doc: add missing git footers
  doc: drop author/documentation sections from most pages

13 years agotests: scrub environment of GIT_* variables
Jonathan Nieder [Tue, 15 Mar 2011 10:10:45 +0000 (05:10 -0500)]
tests: scrub environment of GIT_* variables

Variables from the inherited environment that are meaningful to git
can break tests in undesirable ways.  For example,

GIT_PAGER=more sh t5400-send-pack.sh -v -i

hangs.  So unset all environment variables in the GIT_ namespace in
test-lib, with a few exceptions:

- GIT_TRACE* are useful for tracking down bugs exhibited by a failing
  test;

- GIT_DEBUG* are GIT_TRACE variables by another name, practically
  speaking.  They should probably be tweaked to follow the
  GIT_TRACE_foo scheme and use trace_printf machinery some time.

- GIT_USE_LOOKUP from v1.5.6-rc0~134^2~1 (sha1-lookup: more memory
  efficient search in sorted list of SHA-1, 2007-12-29) is about
  trying an alternate implementation strategy rather than changing
  semantics and it can be useful to compare performance with and
  without it set.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoconfig: drop support for GIT_CONFIG_NOGLOBAL
Jonathan Nieder [Tue, 15 Mar 2011 09:04:49 +0000 (04:04 -0500)]
config: drop support for GIT_CONFIG_NOGLOBAL

Now that test-lib sets $HOME to protect against pollution from user
settings, GIT_CONFIG_NOGLOBAL is not needed for use by the test
suite any more.  And as luck would have it, a quick code search
reveals no other users in the wild.

This patch does not affect GIT_CONFIG_NOSYSTEM, which is still
needed.

Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agogitattributes: drop support for GIT_ATTR_NOGLOBAL
Jonathan Nieder [Tue, 15 Mar 2011 09:04:15 +0000 (04:04 -0500)]
gitattributes: drop support for GIT_ATTR_NOGLOBAL

test-lib sets $HOME to protect against pollution from user settings,
so setting GIT_ATTR_NOGLOBAL would be redundant.  Simplify by
eliminating support for that environment variable altogether.

GIT_ATTR_NOGLOBAL was introduced in v1.7.4-rc0~208^2 (Add global and
system-wide gitattributes, 2010-09-01) as an undocumented feature for
use by the test suite.  It never ended up being used (neither within
git.git nor in other projects).

This patch does not affect GIT_ATTR_NOSYSTEM, which should still be
useful.

Helped-by: Jeff King <peff@peff.net>
Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agotests: suppress system gitattributes
Jonathan Nieder [Tue, 15 Mar 2011 09:05:10 +0000 (04:05 -0500)]
tests: suppress system gitattributes

Set GIT_ATTR_NOSYSTEM in test-lib to make tests more reliable in two
ways:

 - an invalid GIT_ATTR_NOSYSTEM setting should not cause tests to fail
   with "fatal: bad config value for 'GIT_ATTR_NOSYSTEM'".

 - /etc/gitattributes should not change the outcome of tests.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Improved-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agotests: stop worrying about obsolete environment variables
Jonathan Nieder [Tue, 15 Mar 2011 10:09:24 +0000 (05:09 -0500)]
tests: stop worrying about obsolete environment variables

After v0.99.7~99 (Retire support for old environment variables,
2005-09-09), there is no more need to unset a stray AUTHOR_NAME
variable that might have entered the test environment.

Signed-off-by: Jonathan Nieder <jrnieder@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoupload-pack: Implement no-done capability
Shawn O. Pearce [Tue, 15 Mar 2011 00:59:40 +0000 (17:59 -0700)]
upload-pack: Implement no-done capability

If the client requests both multi_ack_detailed and no-done then
upload-pack is free to immediately send a PACK following its first
'ACK %s ready' message.  The upload-pack response actually winds
up being:

  ACK %s common
  ... (maybe more) ...
  ACK %s ready
  NAK
  ACK %s
  PACK.... the pack stream ....

For smart HTTP connections this saves one HTTP RPC, reducing
the overall latency for a trivial fetch.  For git:// and ssh://
a no-done option slightly reduces latency by removing one
server->client->server round-trip at the end of the common
ancestor negotiation.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agofetch-pack: Implement no-done capability
Shawn O. Pearce [Tue, 15 Mar 2011 00:59:39 +0000 (17:59 -0700)]
fetch-pack: Implement no-done capability

If enabled on the connection "multi_ack_detailed no-done" as a
pair allows the remote upload-pack process to send a PACK down
to the client as soon as a "ACK %s ready" message was also sent.

Over git:// and ssh:// where a bi-directional stream is in place
this has very little difference over the classical version that
waits for the client to send a "done\n" line by itself.  It does
slightly reduce the latency involved to start the pack stream as
there is one less round-trip from client->server required.

Over smart HTTP this avoids needing to send a final RPC that has
all of the prior common objects.  Instead the server is able to
return a pack as soon as its ready to.  For many common users the
smart HTTP fetch is now just 2 requests: GET .../info/refs, and
a POST .../git-upload-pack to not only negotiate but also receive
the pack stream.  Only users who have more than 32 local unshared
commits with the remote will need additional requests to negotiate
a common merge base.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agoupload-pack: More aggressively send 'ACK %s ready'
Shawn O. Pearce [Mon, 14 Mar 2011 23:48:39 +0000 (16:48 -0700)]
upload-pack: More aggressively send 'ACK %s ready'

If a client is merely following the remote (and has not made any
new commits itself), all "have %s" lines sent by the client will be
common to the server.  As all lines are common upload-pack never
calls ok_to_give_up() and does not compute if it has a good cut
point in the commit graph.

Without this computation the following client is going to send all
tagged commits, as these were determined to be COMMON_REF during the
initial advertisement, but the client does not parse their history
to transitively pass the COMMON flag and empty its queue of commits.

For git.git with 339 commit tags, it takes clients 11 rounds of
negotation to fully send all tagged commits and exhaust its queue
of things to send as common.  This is pretty slow for a client that
has not done any local development activity.

Force computing ok_to_give_up() and send "ACK %s ready" at the end
of the current round if this round only contained common objects
and ok_to_give_up() was therefore not called.  This may allow the
client to break early, avoiding transmission of the COMMON_REFs.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
13 years agofetch-pack: Finish negotation if remote replies "ACK %s ready"
Shawn O. Pearce [Mon, 14 Mar 2011 23:48:38 +0000 (16:48 -0700)]
fetch-pack: Finish negotation if remote replies "ACK %s ready"

If multi_ack_detailed was selected in the protocol capabilities
(both client and server are >= Git 1.6.6) the upload-pack side will
send "ACK %s ready" when it knows how to safely cut the graph and
produce a reasonable pack for the want list that was already sent
on the connection.

Upon receiving "ACK %s ready" there is no point in looking at
the remaining commits inside of rev_list.  Sending additional
"have %s" lines to the remote will not construct a smaller pack.
It is unlikely a commit older than the current cut point will have
a better delta base than the cut point itself has.

The original design of this code had fetch-pack empty rev_list by
marking a commit and its transitive ancestors COMMON whenever the
remote side said "ACK %s {continue,common}" and skipping over any
already COMMON commits during get_rev().  This approach does not
work when most of rev_list is actually COMMON_REF, commits that
are pointed to by a reference on the remote, which exist locally,
and which have not yet been sent to the remote as a "have %s" line.

Most of the common references are tags in the ref/tags namespace,
using points in the commit graph that are more than 1 commit apart.
In git.git itself, this is currently 340 tags, 339 of which point to
commits in the commit graph.  fetch-pack pushes all of these into
rev_list, but is unable to mark them COMMON and discard during a
remote's "ACK %s {continue,common}" because it does not parse through
the entire parent chain.  Not parsing the entire parent chain is
an optimization to avoid walking back to the roots of the repository.

Assuming the client is only following the remote (and does not make
its own local commits), the client needs 11 rounds to spin through
the entire list of tags (32 commits per round, ceil(339/32) == 11).
Unfortunately the server knows on the first "have %s" line that
it can produce a good pack, and does not need to see the remaining
320 tags in the other 10 rounds.

Over git:// and ssh:// this isn't as bad as it sounds, the client is
only transmitting an extra 16,000 bytes that it doesn't need to send.

Over smart HTTP, the client must do an additional 10 HTTP POST
requests, each of which incurs round-trip latency, and must upload
the entire state vector of all known common objects.  On the final
POST request, this is 16 KiB worth of data.

Fix all of this by clearing rev_list as soon as the remote side
says it can construct a pack.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>