Code

git.git
18 years agogit-ls-files: Fix, document, and add test for --error-unmatch option.
Carl Worth [Tue, 21 Feb 2006 20:48:30 +0000 (12:48 -0800)]
git-ls-files: Fix, document, and add test for --error-unmatch option.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix typo in git-rebase.sh.
Jason Riedy [Tue, 21 Feb 2006 20:56:14 +0000 (12:56 -0800)]
Fix typo in git-rebase.sh.

s/upsteram/upstream in git-rebase.sh.

Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoNew test to verify that when git-clone fails it cleans up the new directory.
Carl Worth [Fri, 17 Feb 2006 21:33:28 +0000 (13:33 -0800)]
New test to verify that when git-clone fails it cleans up the new directory.

Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'pj/portable'
Junio C Hamano [Wed, 22 Feb 2006 02:16:29 +0000 (18:16 -0800)]
Merge branch 'pj/portable'

* pj/portable:
  Makefile tweaks: Solaris 9+ dont need iconv / move up uname variables

18 years agogit-rebase: Clarify usage statement and copy it into the actual documentation.
Carl Worth [Wed, 22 Feb 2006 01:10:12 +0000 (17:10 -0800)]
git-rebase: Clarify usage statement and copy it into the actual documentation.

I found a paper thin man page for git-rebase, but was quite happy to
see something much more useful in the usage statement of the script
when I went there to find out how this thing worked. Here it is
cleaned up slightly and expanded a bit into the actual documentation.

Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMakefile tweaks: Solaris 9+ dont need iconv / move up uname variables
Paul Jakma [Mon, 20 Feb 2006 23:36:28 +0000 (23:36 +0000)]
Makefile tweaks: Solaris 9+ dont need iconv / move up uname variables

- Solaris 9 and up do not need -liconv, so NEEDS_LIBICONV should be set
   only for S8.
- Move the declaration of the uname variables to early in the Makefile
   so they can be referenced by prefix and gitexecdir variables.
- gitexecdir defaults to being same as bindir, it might as well reference
   that variable.

[jc: corrupt patch, sneakily tried to remove inclusion of GIT-VERSION-FILE
 I do not know why I am applying this...]

Signed-off-by: Paul Jakma <paul@quagga.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge part of jc/portable branch
Junio C Hamano [Tue, 21 Feb 2006 08:52:18 +0000 (00:52 -0800)]
Merge part of jc/portable branch

18 years agogit-mktree: reverse of git-ls-tree.
Junio C Hamano [Tue, 21 Feb 2006 05:50:01 +0000 (21:50 -0800)]
git-mktree: reverse of git-ls-tree.

This reads data in the format a (non recursive) ls-tree outputs
and writes a tree object to the object database.  The created
tree object name is output to the standard output.

For convenience, the input data does not need to be sorted; the
command sorts the input lines internally.

By request from Tommi Virtanen.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'lt/merge-tree'
Junio C Hamano [Tue, 21 Feb 2006 08:49:38 +0000 (00:49 -0800)]
Merge branch 'lt/merge-tree'

* lt/merge-tree:
  git-merge-tree: generalize the "traverse <n> trees in sync" functionality
  Handling large files with GIT
  Handling large files with GIT

18 years agoMerge branch 'jc/ident'
Junio C Hamano [Tue, 21 Feb 2006 08:46:07 +0000 (00:46 -0800)]
Merge branch 'jc/ident'

* jc/ident:
  Keep Porcelainish from failing by broken ident after making changes.
  Delay "empty ident" errors until they really matter.
  Make "empty ident" error message a bit more helpful.

18 years agocherry-pick/revert: error-help message rewording.
Junio C Hamano [Tue, 21 Feb 2006 08:28:04 +0000 (00:28 -0800)]
cherry-pick/revert: error-help message rewording.

It said "after fixing up, commit the result using -F .msg", but
it was not clear for new people how "fix up" should be done.
Hint "git-update-index <path>".

We could recommend "git commit -a -F .msg" instead, but I am
hesitant to give that suggestion in the blind -- you could do a
cherry-pick, revert or a merge in general in a dirty working
tree as long as local modifications do not overlap with the
merge, but using "commit -a" would include them in the result.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix fmt-merge-msg counting.
Junio C Hamano [Tue, 21 Feb 2006 03:26:21 +0000 (19:26 -0800)]
Fix fmt-merge-msg counting.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: 0.9.1: add --version and copyright/license (GPL v2+) information
Eric Wong [Mon, 20 Feb 2006 18:57:29 +0000 (10:57 -0800)]
git-svn: 0.9.1: add --version and copyright/license (GPL v2+) information

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn: add Makefile, test, and associated ignores
Eric Wong [Mon, 20 Feb 2006 18:57:29 +0000 (10:57 -0800)]
contrib/git-svn: add Makefile, test, and associated ignores

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: fix several corner-case and rare bugs with 'commit'
Eric Wong [Mon, 20 Feb 2006 18:57:28 +0000 (10:57 -0800)]
git-svn: fix several corner-case and rare bugs with 'commit'

None of these were really show-stoppers (or even triggered)
on most of the trees I've tracked.

* Node change prevention for identically named nodes.  This is
  a limitation of SVN, but we find the error and exit before
  it's passed to SVN so we don't dirty our working tree when our
  commit fails.  git-svn will exit with an error code 1 if any
  of the following conditions are found:

  1.  a directory is removed and a file of the same name of the
      removed directory is created
  1a. a file has its parent directory removed and the file is
      takes the name of the removed parent directory::
          baz/zzz    =>  baz
  2.  a file is removed and a directory of the same name of the
      removed file is created.
  2a. a file is moved into a deeper directory that shares the
      previous name of the file::
          dir/$file  =>  dir/file/$file

  Since SVN cannot handle these cases, the user will have to
  manually split the commit into several parts.

* --rmdir now handles nested/deep removals. If dir/a/b/c/d/e/file
  is removed, and everything else is in the dir/ hierarchy is
  otherwise empty, then dir/ will be deleted when file is deleted
  from svn and --rmdir specified.

* Always assert that we have written the tree we want to write
  on commits.  This helped me find several bugs in the symlink
  handling code (which as been fixed).

* Several symlink handling fixes.  We now refuse to set
  permissions on symlinks.  We also always unlink a file
  if we're going to overwrite it.

* Apply changes in a pre-determined order, so we always have
  rename from locations handy before we delete them.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocontrib/git-svn.txt: add a note about renamed/copied directory support
Eric Wong [Mon, 20 Feb 2006 18:57:28 +0000 (10:57 -0800)]
contrib/git-svn.txt: add a note about renamed/copied directory support

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: change ; to && in addremove()
Eric Wong [Mon, 20 Feb 2006 18:57:28 +0000 (10:57 -0800)]
git-svn: change ; to && in addremove()

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: remove any need for the XML::Simple dependency
Eric Wong [Mon, 20 Feb 2006 18:57:28 +0000 (10:57 -0800)]
git-svn: remove any need for the XML::Simple dependency

XML::Simple was originally required back when I made svn-arch-mirror
because I needed to explictly track renames with Arch.  Then I carried
it over to git-svn because I was afraid somebody could commit an svn
log message that could throw off a non-XML log parser.  Then I noticed
the <n> lines column in the header.  So, no more XML :)

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: Allow for more argument types for commit (from..to)
Eric Wong [Mon, 20 Feb 2006 18:57:26 +0000 (10:57 -0800)]
git-svn: Allow for more argument types for commit (from..to)

Allow 'from..to' notation from the command line.

More liberal sha1 parsing when reading from stdin no longer requires the
sha1 to be the first character, so a leading 'commit ' string is OK.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: allow --find-copies-harder and -l<num> to be passed on commit
Eric Wong [Mon, 20 Feb 2006 18:57:26 +0000 (10:57 -0800)]
git-svn: allow --find-copies-harder and -l<num> to be passed on commit

Both of these options are passed directly to git-diff-tree when
committing to a SVN repository.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: fix a typo in defining the --no-stop-on-copy option
Eric Wong [Mon, 20 Feb 2006 18:57:25 +0000 (10:57 -0800)]
git-svn: fix a typo in defining the --no-stop-on-copy option

Just a typo, I doubt anybody would use (and I highly recommend not
using) this option anyways.  But you never know...

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/merge-msg'
Junio C Hamano [Mon, 20 Feb 2006 05:18:17 +0000 (21:18 -0800)]
Merge branch 'jc/merge-msg'

* jc/merge-msg:
  fmt-merge-msg: do not add excess newline at the end.
  fmt-merge-msg: say which branch things were merged into unless 'master'

18 years agoMerge branch 'jc/mv'
Junio C Hamano [Mon, 20 Feb 2006 05:17:59 +0000 (21:17 -0800)]
Merge branch 'jc/mv'

* jc/mv:
  Allow git-mv to accept ./ in paths.

18 years agofmt-merge-msg: do not add excess newline at the end.
Junio C Hamano [Mon, 20 Feb 2006 05:14:56 +0000 (21:14 -0800)]
fmt-merge-msg: do not add excess newline at the end.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation: fix typo in rev-parse --short option description.
Junio C Hamano [Sun, 19 Feb 2006 18:23:17 +0000 (10:23 -0800)]
Documentation: fix typo in rev-parse --short option description.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAllow git-mv to accept ./ in paths.
Junio C Hamano [Sun, 19 Feb 2006 07:42:03 +0000 (23:42 -0800)]
Allow git-mv to accept ./ in paths.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoOptionally work without python
Johannes Schindelin [Sat, 18 Feb 2006 12:01:18 +0000 (13:01 +0100)]
Optionally work without python

In some setups (notably server setups) you do not need that dependency.
Gracefully handle the absence of python when NO_PYTHON is defined.

Signed-off-by: Johannes E. Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge fixes up to GIT 1.2.2
Junio C Hamano [Sun, 19 Feb 2006 06:55:42 +0000 (22:55 -0800)]
Merge fixes up to GIT 1.2.2

18 years agofmt-merge-msg: say which branch things were merged into unless 'master'
Junio C Hamano [Sun, 19 Feb 2006 06:37:02 +0000 (22:37 -0800)]
fmt-merge-msg: say which branch things were merged into unless 'master'

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoKeep Porcelainish from failing by broken ident after making changes.
Junio C Hamano [Sun, 19 Feb 2006 04:51:26 +0000 (20:51 -0800)]
Keep Porcelainish from failing by broken ident after making changes.

"empty ident not allowed" error makes commit-tree fail, so we
are already safer in that we would not end up with commit
objects that have bogus names on the author or committer fields.
However, before commit-tree is called there are already changes
made to the index file and the working tree.  The operation can
be resumed after fixing the environment problem, but when this
triggers to a newcomer with unusable gecos, the first question
becomes "what did I lose and how would I recover".

This patch modifies some Porcelainish commands to verify
GIT_COMMITTER_IDENT as soon as we know we are going to make some
commits before doing much damage to prevent confusion.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDelay "empty ident" errors until they really matter.
Junio C Hamano [Sun, 19 Feb 2006 04:31:05 +0000 (20:31 -0800)]
Delay "empty ident" errors until they really matter.

Previous one warned people upfront to encourage fixing their
environment early, but some people just use repositories and git
tools read-only without making any changes, and in such a case
there is not much point insisting on them having a usable ident.

This round attempts to move the error until either "git-var"
asks for the ident explicitly or "commit-tree" wants to use it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix retries in git-cvsimport v1.2.2
Martin Mares [Sat, 18 Feb 2006 20:44:20 +0000 (21:44 +0100)]
Fix retries in git-cvsimport

Fixed a couple of bugs in recovering from broken connections:

The _line() method now returns undef correctly when the connection
is broken instead of falling off the function and returning garbage.

Retries are now reported to stderr and the eventual partially
downloaded file is discarded instead of being appended to.

The "Server gone away" test has been removed, because it was
reachable only if the garbage return bug bit.

Signed-off-by: Martin Mares <mj@ucw.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoarchimport: remove files from the index before adding/updating
Eric Wong [Sat, 18 Feb 2006 11:49:38 +0000 (03:49 -0800)]
archimport: remove files from the index before adding/updating

This fixes a bug when importing where a directory gets removed/renamed
but is immediately replaced by a file of the same name in the same
changeset.

This fix only applies to the accurate (default) strategy the moment.

This patch should also fix the fast strategy if/when it is updated
to handle the cases that would've triggered this bug.

This bug was originally found in git-svn, but I remembered I did the
same thing with archimport as well.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd an Emacs interface in contrib.
Alexandre Julliard [Sat, 18 Feb 2006 16:50:49 +0000 (17:50 +0100)]
Add an Emacs interface in contrib.

This is an Emacs interface for git. The user interface is modeled on
pcl-cvs. It has been developed on Emacs 21 and will probably need some
tweaking to work on XEmacs.

The basic command is 'M-x git-status' which displays a buffer listing
modified files in the selected project tree. In that buffer the
following features are supported:

  - add/remove files
  - list unknown files
  - commit marked files
  - manage .gitignore
  - commit merges based on MERGE_HEAD
  - revert files to the HEAD version
  - resolve conflicts with smerge or ediff
  - diff files against HEAD/base/mine/other or combined diff
  - get a log of the revisions for specified files

There are plenty of unimplemented features too, see the TODO list at
the top of the file...

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake "empty ident" error message a bit more helpful.
Junio C Hamano [Sat, 18 Feb 2006 09:20:06 +0000 (01:20 -0800)]
Make "empty ident" error message a bit more helpful.

It appears that some people who did not care about having bogus
names in their own commit messages are bitten by the recent
change to require a sane environment [*1*].

While it was a good idea to prevent people from using bogus
names to create commits and doing sign-offs, the error message
is not very informative.  This patch attempts to warn things
upfront and hint people how to fix their environments.

[Footnote]

*1* The thread is this one.

    http://marc.theaimsgroup.com/?t=113868084800004

    Especially this message.

    http://marc.theaimsgroup.com/?m=113932830015032

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/topo'
Junio C Hamano [Sat, 18 Feb 2006 09:24:10 +0000 (01:24 -0800)]
Merge branch 'jc/topo'

* jc/topo:
  topo-order: make --date-order optional.

18 years agoMerge branch 'jc/rebase-limit'
Junio C Hamano [Sat, 18 Feb 2006 09:24:01 +0000 (01:24 -0800)]
Merge branch 'jc/rebase-limit'

* jc/rebase-limit:
  rebase: allow rebasing onto different base.

18 years agogitview: typofix
Aneesh Kumar K.V [Sat, 18 Feb 2006 08:24:50 +0000 (13:54 +0530)]
gitview: typofix

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@gmail.com>
18 years agogit-svn: remove files from the index before adding/updating
Eric Wong [Sat, 18 Feb 2006 05:04:47 +0000 (21:04 -0800)]
git-svn: remove files from the index before adding/updating

This fixes a bug when importing where a directory gets removed/renamed
but is immediately replaced by a file of the same name in the same
revision.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
18 years agoMake git-reset delete empty directories
Shawn Pearce [Fri, 17 Feb 2006 07:26:16 +0000 (02:26 -0500)]
Make git-reset delete empty directories

When git-reset --hard is used and a subdirectory becomes
empty (as it contains no tracked files in the target tree)
the empty subdirectory should be removed.  This matches
the behavior of git-checkout-index and git-read-tree -m
which would not have created the subdirectory or would
have deleted it when updating the working directory.

Subdirectories which are not empty will be left behind.
This may happen if the subdirectory still contains object
files from the user's build process (for example).

[jc: simplified the logic a bit, while keeping the test script.]

18 years agoDocument --short and --git-dir in git-rev-parse(1)
Jonas Fonseca [Sat, 18 Feb 2006 01:11:36 +0000 (02:11 +0100)]
Document --short and --git-dir in git-rev-parse(1)

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
18 years agogit-rev-parse: Fix --short= option parsing
Jonas Fonseca [Sat, 18 Feb 2006 01:10:53 +0000 (02:10 +0100)]
git-rev-parse: Fix --short= option parsing

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
18 years agoSupport Irix
Johannes Schindelin [Fri, 17 Feb 2006 14:23:41 +0000 (15:23 +0100)]
Support Irix

Signed-off-by: Johannes E. Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoOptionally support old diffs
Johannes Schindelin [Fri, 17 Feb 2006 14:23:16 +0000 (15:23 +0100)]
Optionally support old diffs

Some versions of diff do not correctly detect a missing new-line at the end
of the file under certain circumstances.

When defining NO_ACCURATE_DIFF, work around this bug.

Signed-off-by: Johannes E. Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix cpio call
Johannes Schindelin [Fri, 17 Feb 2006 14:22:45 +0000 (15:22 +0100)]
Fix cpio call

To some cpio's, -a and -m options are mutually exclusive. Use only -m.

Signed-off-by: Johannes E. Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoPrevent git-upload-pack segfault if object cannot be found
Carl Worth [Sat, 18 Feb 2006 00:14:52 +0000 (16:14 -0800)]
Prevent git-upload-pack segfault if object cannot be found

Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAbstract test_create_repo out for use in tests.
Carl Worth [Fri, 17 Feb 2006 21:33:26 +0000 (13:33 -0800)]
Abstract test_create_repo out for use in tests.

Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTrap exit to clean up created directory if clone fails.
Carl Worth [Fri, 17 Feb 2006 21:33:24 +0000 (13:33 -0800)]
Trap exit to clean up created directory if clone fails.

Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSubmittingPatches: note on whitespaces
Junio C Hamano [Sat, 18 Feb 2006 00:15:26 +0000 (16:15 -0800)]
SubmittingPatches: note on whitespaces

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd a README for gitview
Aneesh Kumar K.V [Fri, 17 Feb 2006 12:48:36 +0000 (18:18 +0530)]
Add a README for gitview

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd contrib/README.
Junio C Hamano [Fri, 17 Feb 2006 21:33:14 +0000 (13:33 -0800)]
Add contrib/README.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-tag: -l to list tags (usability).
Junio C Hamano [Fri, 17 Feb 2006 12:04:39 +0000 (04:04 -0800)]
git-tag: -l to list tags (usability).

git-tag -l lists all tags, and git-tag -l <pattern> filters the
result with <pattern>.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd contrib/gitview from Aneesh.
Aneesh Kumar [Fri, 17 Feb 2006 10:10:31 +0000 (02:10 -0800)]
Add contrib/gitview from Aneesh.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: ensure fetch always works chronologically.
Eric Wong [Fri, 17 Feb 2006 02:13:32 +0000 (18:13 -0800)]
git-svn: ensure fetch always works chronologically.

We run svn log against a URL without a working copy for the first fetch,
so we end up a log that's sorted from highest to lowest.  That's bad, we
always want lowest to highest.  Just default to --revision 0:HEAD now if
-r isn't specified for the first fetch.

Also sort the revisions after we get them just in case somebody
accidentally reverses the argument to --revision for whatever reason.

Thanks again to Emmanuel Guerin for helping me find this.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: fix revision order when XML::Simple is not loaded
Eric Wong [Thu, 16 Feb 2006 19:47:51 +0000 (11:47 -0800)]
git-svn: fix revision order when XML::Simple is not loaded

Thanks to Emmanuel Guerin for finding the bug.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoIntroducing contrib/git-svn.
Eric Wong [Thu, 16 Feb 2006 09:24:16 +0000 (01:24 -0800)]
Introducing contrib/git-svn.

18 years agoAllow building Git in systems without iconv
Fernando J. Pereda [Thu, 16 Feb 2006 08:38:01 +0000 (09:38 +0100)]
Allow building Git in systems without iconv

Systems using some uClibc versions do not properly support
iconv stuff. This patch allows Git to be built on those
systems by passing NO_ICONV=YesPlease to make. The only
drawback is mailinfo won't do charset conversion in those
systems.

Signed-off-by: Fernando J. Pereda <ferdy@gentoo.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-merge-tree: generalize the "traverse <n> trees in sync" functionality
Linus Torvalds [Thu, 16 Feb 2006 03:25:32 +0000 (19:25 -0800)]
git-merge-tree: generalize the "traverse <n> trees in sync" functionality

It's actually very useful for other things too. Notably, we could do the
combined diff a lot more efficiently with this.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoHandling large files with GIT
Linus Torvalds [Wed, 15 Feb 2006 02:33:02 +0000 (18:33 -0800)]
Handling large files with GIT

On Tue, 14 Feb 2006, Linus Torvalds wrote:
>
> Here, btw, is the trivial diff to turn my previous "tree-resolve" into a
> "resolve tree relative to the current branch".

Gaah. It was trivial, and it happened to work fine for my test-case, but
when I started looking at not doing that extremely aggressive subdirectory
merging, that showed a few other issues...

So in case people want to try, here's a third patch. Oh, and it's against
my _original_ path, not incremental to the middle one (ie both patches two
and three are against patch #1, it's not a nice series).

Now I'm really done, and won't be sending out any more patches today.
Sorry for the noise.

Linus

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoHandling large files with GIT
Linus Torvalds [Wed, 15 Feb 2006 02:05:30 +0000 (18:05 -0800)]
Handling large files with GIT

On Tue, 14 Feb 2006, Junio C Hamano wrote:

> Linus Torvalds <torvalds@osdl.org> writes:
>
> > If somebody is interested in making the "lots of filename changes" case go
> > fast, I'd be more than happy to walk them through what they'd need to
> > change. I'm just not horribly motivated to do it myself. Hint, hint.
>
> In case anybody is wondering, I share the same feeling.  I
> cannot say I'd be "more than happy to" clean up potential
> breakages during the development of such changes, but if the
> change eventually would help certain use cases, I can be
> persuaded to help debugging such a mess ;-).

Actually, I got interested in seeing how hard this is, and wrote a simple
first cut at doing a tree-optimized merger.

Let me shout a bit first:

  THIS IS WORKING CODE, BUT BE CAREFUL: IT'S A TECHNOLOGY DEMONSTRATION
  RATHER THAN THE FINAL PRODUCT!

With that out of the way, let me descibe what this does (and then describe
the missing parts).

This is basically a three-way merge that works entirely on the "tree"
level, rather than on the index. A lot of the _concepts_ are the same,
though, and if you're familiar with the results of an index merge, some of
the output will make more sense.

You give it three trees: the base tree (tree 0), and the two branches to
be merged (tree 1 and tree 2 respectively). It will then walk these three
trees, and resolve them as it goes along.

The interesting part is:
 - it can resolve whole sub-directories in one go, without actually even
   looking recursively at them. A whole subdirectory will resolve the same
   way as any individual files will (although that may need some
   modification, see later).
 - if it has a "content conflict", for subdirectories that means "try to
   do a recursive tree merge", while for non-subdirectories it's just a
   content conflict and we'll output the stage 1/2/3 information.
 - a successful merge will output a single stage 0 ("merged") entry,
   potentially for a whole subdirectory.
 - it outputs all the resolve information on stdout, so something like the
   recursive resolver can pretty easily parse it all.

Now, the caveats:
 - we probably need to be more careful about subdirectory resolves. The
   trivial case (both branches have the exact same subdirectory) is a
   trivial resolve, but the other cases ("branch1 matches base, branch2 is
   different" probably can't be silently just resolved to the "branch2"
   subdirectory state, since it might involve renames into - or out of -
   that subdirectory)
 - we do not track the current index file at all, so this does not do the
   "check that index matches branch1" logic that the three-way merge in
   git-read-tree does. The theory is that we'd do a full three-way merge
   (ignoring the index and working directory), and then to update the
   working tree, we'd do a two-way "git-read-tree branch1->result"
 - I didn't actually make it do all the trivial resolve cases that
   git-read-tree does. It's a technology demonstration.

Finally (a more serious caveat):
 - doing things through stdout may end up being so expensive that we'd
   need to do something else. In particular, it's likely that I should
   not actually output the "merge results", but instead output a "merge
   results as they _differ_ from branch1"

However, I think this patch is already interesting enough that people who
are interested in merging trees might want to look at it. Please keep in
mind that tech _demo_ part, and in particular, keep in mind the final
"serious caveat" part.

In many ways, the really _interesting_ part of a merge is not the result,
but how it _changes_ the branch we're merging into. That's particularly
important as it should hopefully also mean that the output size for any
reasonable case is minimal (and tracks what we actually need to do to the
current state to create the final result).

The code very much is organized so that doing the result as a "diff
against branch1" should be quite easy/possible. I was actually going to do
it, but I decided that it probably makes the output harder to read. I
dunno.

Anyway, let's think about this kind of approach.. Note how the code itself
is actually quite small and short, although it's prbably pretty "dense".

As an interesting test-case, I'd suggest this merge in the kernel:

git-merge-tree $(git-merge-base 4cbf876 7d2babc4cbf876 7d2babc

which resolves beautifully (there are no actual file-level conflicts), and
you can look at the output of that command to start thinking about what
it does.

The interesting part (perhaps) is that timing that command for me shows
that it takes all of 0.004 seconds.. (the git-merge-base thing takes
considerably more ;)

The point is, we _can_ do the actual merge part really really quickly.

Linus

PS. Final note: when I say that it is "WORKING CODE", that is obviously by
my standards. IOW, I tested it once and it gave reasonable results - so it
must be perfect.

Whether it works for anybody else, or indeed for any other test-case, is
not my problem ;)

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotopo-order: make --date-order optional.
Junio C Hamano [Thu, 16 Feb 2006 06:05:33 +0000 (22:05 -0800)]
topo-order: make --date-order optional.

This adds --date-order to rev-list; it is similar to topo order
in the sense that no parent comes before all of its children,
but otherwise things are still ordered in the commit timestamp
order.

The same flag is also added to show-branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/add'
Junio C Hamano [Thu, 16 Feb 2006 03:42:15 +0000 (19:42 -0800)]
Merge branch 'jc/add'

* jc/add:
  Detect misspelled pathspec to git-add

18 years agoMerge fixes up to 1.2.1
Junio C Hamano [Thu, 16 Feb 2006 03:39:21 +0000 (19:39 -0800)]
Merge fixes up to 1.2.1

18 years agoMore useful/hinting error messages in git-checkout v1.2.1
Josef Weidendorfer [Wed, 15 Feb 2006 19:22:11 +0000 (20:22 +0100)]
More useful/hinting error messages in git-checkout

Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoPrint an error if cloning a http repo and NO_CURL is set
Fernando J. Pereda [Wed, 15 Feb 2006 11:37:30 +0000 (12:37 +0100)]
Print an error if cloning a http repo and NO_CURL is set

If Git is compiled with NO_CURL=YesPlease and one tries to
clone a http repository, git-clone tries to call the curl
binary. This trivial patch prints an error instead in such
situation.

Signed-off-by: Fernando J. Pereda <ferdy@gentoo.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agopacked objects: minor cleanup
Junio C Hamano [Wed, 15 Feb 2006 20:47:43 +0000 (12:47 -0800)]
packed objects: minor cleanup

The delta depth is unsigned.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDetect misspelled pathspec to git-add
Junio C Hamano [Wed, 15 Feb 2006 09:05:59 +0000 (01:05 -0800)]
Detect misspelled pathspec to git-add

This is in the same spirit as an earlier patch for git-commit.
It does an extra ls-files to avoid complaining when a fully
tracked directory name is given on the command line (otherwise
--others restriction would say the pathspec does not match).

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agols-files --error-unmatch pathspec error reporting fix.
Junio C Hamano [Wed, 15 Feb 2006 09:10:13 +0000 (01:10 -0800)]
ls-files --error-unmatch pathspec error reporting fix.

Earlier patch mistakenly used prefix_len when it meant
prefix_offset.  The latter is to strip the leading directories
when run from a subdirectory.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'kh/svn'
Junio C Hamano [Wed, 15 Feb 2006 01:51:50 +0000 (17:51 -0800)]
Merge branch 'kh/svn'

* kh/svn:
  git-svnimport: -r adds svn revision number to commit messages

18 years agoMerge branch 'jc/commit'
Junio C Hamano [Wed, 15 Feb 2006 01:51:02 +0000 (17:51 -0800)]
Merge branch 'jc/commit'

* jc/commit:
  commit: detect misspelled pathspec while making a partial commit.
  combine-diff: diff-files fix (#2)
  combine-diff: diff-files fix.

18 years agoMerge branch 'jc/rebase'
Junio C Hamano [Wed, 15 Feb 2006 01:49:00 +0000 (17:49 -0800)]
Merge branch 'jc/rebase'

* jc/rebase:
  rebase: allow a hook to refuse rebasing.

18 years agoMerge branch 'ra/email'
Junio C Hamano [Wed, 15 Feb 2006 01:46:41 +0000 (17:46 -0800)]
Merge branch 'ra/email'

* ra/email:
  send-email: Add --cc
  send-email: Add some options for controlling how addresses are automatically added to the cc: list.

18 years agorebase: allow rebasing onto different base.
Junio C Hamano [Tue, 14 Feb 2006 22:42:05 +0000 (14:42 -0800)]
rebase: allow rebasing onto different base.

This allows you to rewrite history a bit more flexibly, by
separating the other branch name and new branch point.  By
default, the new branch point is the same as the tip of the
other branch as before, but you can specify where you graft the
rebased branch onto.

When you have this ancestry graph:

          A---B---C topic
         /
    D---E---F---G master

$ git rebase --onto master~1 master topic

would rewrite the history to look like this:

      A'\''--B'\''--C'\'' topic
     /
    D---E---F---G master

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocheckout: fix dirty-file display.
Junio C Hamano [Wed, 15 Feb 2006 00:05:57 +0000 (16:05 -0800)]
checkout: fix dirty-file display.

When we refused to switch branches, we incorrectly showed
differences from the branch we would have switched to.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocommit: detect misspelled pathspec while making a partial commit.
Junio C Hamano [Tue, 14 Feb 2006 20:40:20 +0000 (12:40 -0800)]
commit: detect misspelled pathspec while making a partial commit.

When you say "git commit Documentaiton" to make partial commit
for the files only in that directory, we did not detect that as
a misspelled pathname and attempted to commit index without
change.  If nothing matched, there is no harm done, but if the
index gets modified otherwise by having another valid pathspec
or after an explicit update-index, a user will not notice
without paying attention to the "git status" preview.

This introduces --error-unmatch option to ls-files, and uses it
to detect this common user error.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svnimport: -r adds svn revision number to commit messages
Karl Hasselström [Tue, 14 Feb 2006 02:43:34 +0000 (03:43 +0100)]
git-svnimport: -r adds svn revision number to commit messages

New -r flag for prepending the corresponding Subversion revision
number to each commit message.

Signed-off-by: Karl Hasselström <kha@treskal.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocombine-diff: diff-files fix (#2)
Junio C Hamano [Tue, 14 Feb 2006 09:11:42 +0000 (01:11 -0800)]
combine-diff: diff-files fix (#2)

The raw format "git-diff-files -c" to show unmerged state forgot
to initialize the status fields from parents, causing NUL
characters to be emitted.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge some proposed fixes
Junio C Hamano [Tue, 14 Feb 2006 07:34:58 +0000 (23:34 -0800)]
Merge some proposed fixes

Conflicts:

Documentation/git-commit.txt - taking the post 1.2.0 semantics.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'pb/bisect'
Junio C Hamano [Tue, 14 Feb 2006 07:26:53 +0000 (23:26 -0800)]
Merge branch 'pb/bisect'

* pb/bisect:
  Properly git-bisect reset after bisecting from non-master head

18 years agocombine-diff: diff-files fix.
Junio C Hamano [Tue, 14 Feb 2006 07:07:04 +0000 (23:07 -0800)]
combine-diff: diff-files fix.

When showing a conflicted merge from index stages and working
tree file, we did not fetch the mode from the working tree,
and mistook that as a deleted file.  Also if the manual
resolution (or automated resolution by git rerere) ended up
taking either parent's version, we did not show _anything_ for
that path.  Either was quite bad and confusing.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agos/SHELL/SHELL_PATH/ in Makefile
Fredrik Kuivinen [Mon, 13 Feb 2006 23:15:14 +0000 (00:15 +0100)]
s/SHELL/SHELL_PATH/ in Makefile

With the current Makefile we don't use the shell chosen by the
platform specific defines when we invoke GIT-VERSION-GEN.

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobisect: remove BISECT_NAMES after done.
Junio C Hamano [Tue, 14 Feb 2006 05:25:38 +0000 (21:25 -0800)]
bisect: remove BISECT_NAMES after done.

I noticed that we forgot to clean this file and kept it that
way, while trying to help with Andrew's bisect problem.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation: git-ls-files asciidocco.
Junio C Hamano [Tue, 14 Feb 2006 05:52:10 +0000 (21:52 -0800)]
Documentation: git-ls-files asciidocco.

Noticed by Jon Nelson.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosend-email: Add --cc
Ryan Anderson [Mon, 13 Feb 2006 08:05:15 +0000 (03:05 -0500)]
send-email: Add --cc

Since Junio used this in an example, and I've personally tried to use it, I
suppose the option should actually exist.

Signed-off-by: Ryan Anderson <ryan@michonline.com>
18 years agoDocumentation: git-commit in 1.2.X series defaults to --include.
Junio C Hamano [Mon, 13 Feb 2006 08:26:14 +0000 (00:26 -0800)]
Documentation: git-commit in 1.2.X series defaults to --include.

The documentation was mistakenly describing the --only semantics to
be default.  The 1.2.0 release and its maintenance series 1.2.X will
keep the traditional --include semantics as the default.  Clarify the
situation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosend-email: Add some options for controlling how addresses are automatically added...
Ryan Anderson [Mon, 13 Feb 2006 07:57:09 +0000 (02:57 -0500)]
send-email: Add some options for controlling how addresses are automatically added to the cc: list.

Signed-off-by: Ryan Anderson <ryan@michonline.com>
18 years agorebase: allow a hook to refuse rebasing.
Junio C Hamano [Mon, 13 Feb 2006 07:17:04 +0000 (23:17 -0800)]
rebase: allow a hook to refuse rebasing.

This lets a hook to interfere a rebase and help prevent certain
branches from being rebased by mistake.  A sample hook to show
how to prevent a topic branch that has already been merged into
publish branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit: Now --only semantics is the default.
Junio C Hamano [Mon, 13 Feb 2006 07:55:07 +0000 (23:55 -0800)]
git-commit: Now --only semantics is the default.

This changes the "git commit paths..." to default to --only
semantics from traditional --include semantics, as agreed on the
list.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoGIT 1.2.0 v1.2.0
Junio C Hamano [Sun, 12 Feb 2006 21:14:53 +0000 (13:14 -0800)]
GIT 1.2.0

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix "test: unexpected operator" on bsd
Junio C Hamano [Sun, 12 Feb 2006 21:13:12 +0000 (13:13 -0800)]
Fix "test: unexpected operator" on bsd

This fixes the same issue as a previous fix by Alex Riesen does.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoProperly git-bisect reset after bisecting from non-master head
Petr Baudis [Sun, 12 Feb 2006 16:06:14 +0000 (17:06 +0100)]
Properly git-bisect reset after bisecting from non-master head

git-bisect reset without an argument would return to master even
if the bisecting started at a non-master branch. This patch makes
it save the original branch name to .git/head-name and restore it
afterwards.

This is also compatible with Cogito and cg-seek, so cg-status will
show that we are seeked on the bisect branch and cg-reset will
properly restore the original branch.

git-bisect start will refuse to work if it is not on a bisect but
.git/head-name exists; this is to protect against conflicts with
other seeking tools.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-commit: show dirtiness including index.
Junio C Hamano [Sun, 12 Feb 2006 21:05:53 +0000 (13:05 -0800)]
git-commit: show dirtiness including index.

Earlier, when we switched a branch we used diff-files to show
paths that are dirty in the working tree.  But we allow switching
branches with updated index ("read-tree -m -u $old $new" works that
way), and only showing paths that have differences in the working
tree but not paths that are different in index was confusing.

This shows both as modified from the top commit of the branch we
just have switched to.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake pack-objects chattier.
Junio C Hamano [Sun, 12 Feb 2006 21:01:54 +0000 (13:01 -0800)]
Make pack-objects chattier.

You could give -q to squelch it, but currently no tool does it.
This would make 'git clone host:repo here' over ssh not silent
again.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoavoid echo -e, there are systems where it does not work
Alex Riesen [Sun, 12 Feb 2006 18:05:34 +0000 (19:05 +0100)]
avoid echo -e, there are systems where it does not work

FreeBSD 4.11 being one example: the built-in echo doesn't have -e,
and the installed /bin/echo does not do "-e" as well.
"printf" works, laking just "\e" and "\xAB'.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agofix "test: 2: unexpected operator" on bsd
Alex Riesen [Sun, 12 Feb 2006 18:03:16 +0000 (19:03 +0100)]
fix "test: 2: unexpected operator" on bsd

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix object re-hashing
Linus Torvalds [Sun, 12 Feb 2006 19:24:50 +0000 (11:24 -0800)]
Fix object re-hashing

The hashed object lookup had a subtle bug in re-hashing: it did

for (i = 0; i < count; i++)
if (objs[i]) {
.. rehash ..

where "count" was the old hash couny. Oon the face of it is obvious, since
it clearly re-hashes all the old objects.

However, it's wrong.

If the last old hash entry before re-hashing was in use (or became in use
by the re-hashing), then when re-hashing could have inserted an object
into the hash entries with idx >= count due to overflow. When we then
rehash the last old entry, that old entry might become empty, which means
that the overflow entries should be re-hashed again.

In other words, the loop has to be fixed to either traverse the whole
array, rather than just the old count.

(There's room for a slight optimization: instead of counting all the way
up, we can break when we see the first empty slot that is above the old
"count". At that point we know we don't have any collissions that we might
have to fix up any more. This patch only does the trivial fix)

[jc: with trivial fix on trivial fix]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agohashtable-based objects: minimum fixups.
Junio C Hamano [Sun, 12 Feb 2006 02:51:19 +0000 (18:51 -0800)]
hashtable-based objects: minimum fixups.

Calling hashtable_index from find_object before objs is created
would result in division by zero failure.  Avoid it.

Also the given object name may not be aligned suitably for
unsigned int; avoid dereferencing casted pointer.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUse a hashtable for objects instead of a sorted list
Johannes Schindelin [Sun, 12 Feb 2006 01:57:57 +0000 (02:57 +0100)]
Use a hashtable for objects instead of a sorted list

In a simple test, this brings down the CPU time from 47 sec to 22 sec.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd howto about separating topics.
kent@lysator.liu.se [Sun, 12 Feb 2006 12:00:52 +0000 (13:00 +0100)]
Add howto about separating topics.

This howto consists of a footnote from an email by JC to the git
mailing list (<7vfyms0x4p.fsf@assigned-by-dhcp.cox.net>).

Signed-off-by: Kent Engstrom <kent@lysator.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'pb/repo'
Junio C Hamano [Sun, 12 Feb 2006 13:02:30 +0000 (05:02 -0800)]
Merge branch 'pb/repo'

* pb/repo:
  Add support for explicit type specifiers when calling git-repo-config