git.git
18 years agoMerge branch 'np/delta' into next
Junio C Hamano [Tue, 25 Apr 2006 05:31:15 +0000 (22:31 -0700)]
Merge branch 'np/delta' into next

* np/delta:
  split the diff-delta interface
  Document the configuration file
  Document git-var -l listing also configuration variables
  rev-parse: better error message for ambiguous arguments

18 years agosplit the diff-delta interface
Nicolas Pitre [Tue, 25 Apr 2006 03:07:47 +0000 (23:07 -0400)]
split the diff-delta interface

This patch splits the diff-delta interface into index creation and delta
generation.  A wrapper is provided to preserve the diff-delta() call.

This will allow for an optimization in pack-objects.c where the source
object could be fixed and a full window of objects tentatively tried
against
that same source object without recomputing the source index each time.

This patch only restructure things, plus a couple cleanups for good
measure. There is no performance change yet.

Signed-off-by: Nicolas Pitre <nico@cam.org>
18 years agoMerge branch 'fix'
Junio C Hamano [Tue, 25 Apr 2006 05:26:57 +0000 (22:26 -0700)]
Merge branch 'fix'

* fix:
  Document the configuration file
  Document git-var -l listing also configuration variables
  rev-parse: better error message for ambiguous arguments

18 years agoDocument the configuration file
Petr Baudis [Mon, 24 Apr 2006 22:59:33 +0000 (00:59 +0200)]
Document the configuration file

This patch adds a Documentation/config.txt file included by git-repo-config
and currently aggregating hopefully all the available git plumbing / core
porcelain configuration variables, as well as briefly describing the format.

It also updates an outdated bit of the example in git-repo-config(1).

Signed-off-by: Petr Baudis <pasky@suse.cz>
18 years agoDocument git-var -l listing also configuration variables
Petr Baudis [Mon, 24 Apr 2006 22:59:28 +0000 (00:59 +0200)]
Document git-var -l listing also configuration variables

Signed-off-by: Petr Baudis <pasky@suse.cz>
18 years agorev-parse: better error message for ambiguous arguments
Paul Mackerras [Tue, 25 Apr 2006 00:00:03 +0000 (10:00 +1000)]
rev-parse: better error message for ambiguous arguments

Currently, if git-rev-parse encounters an argument that is neither a
recognizable revision name nor the name of an existing file or
directory, and it hasn't encountered a "--" argument, it prints an
error message saying "No such file or directory".  This can be
confusing for users, including users of programs such as gitk that
use git-rev-parse, who may then think that they can't ask about the
history of files that no longer exist.

This makes it print a better error message, one that points out the
ambiguity and tells the user what to do to fix it.

Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agoMerge branch 'jc/cache-tree' into next
Junio C Hamano [Tue, 25 Apr 2006 04:28:07 +0000 (21:28 -0700)]
Merge branch 'jc/cache-tree' into next

* jc/cache-tree:
  index: make the index file format extensible.
  cache-tree: protect against "git prune".
  Add test-dump-cache-tree

18 years agoindex: make the index file format extensible.
Junio C Hamano [Tue, 25 Apr 2006 04:18:58 +0000 (21:18 -0700)]
index: make the index file format extensible.

... and move the cache-tree data into it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocache-tree: protect against "git prune".
Junio C Hamano [Mon, 24 Apr 2006 22:12:42 +0000 (15:12 -0700)]
cache-tree: protect against "git prune".

We reused the cache-tree data without verifying the tree object
still exists.  Recompute in cache_tree_update() an otherwise
valid cache-tree entry when the tree object disappeared.

This is not usually a problem, but theoretically without this
fix things can break when the user does something like this:

- read-index from a side branch
- write-tree the result
- remove the side branch with "git branch -D"
- remove the unreachable objects with "git prune"
- write-tree what is in the index.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Mon, 24 Apr 2006 09:18:33 +0000 (02:18 -0700)]
Merge branch 'master' into next

* master:
  gitk: Let git-rev-list do the argument list parsing

18 years agoMerge branch 'ar/chmod-series'
Junio C Hamano [Mon, 24 Apr 2006 09:08:35 +0000 (02:08 -0700)]
Merge branch 'ar/chmod-series'

* ar/chmod-series:
  make update-index --chmod work with multiple files and --stdin

18 years agoMerge branch 'sh/daemon'
Junio C Hamano [Mon, 24 Apr 2006 09:08:28 +0000 (02:08 -0700)]
Merge branch 'sh/daemon'

* sh/daemon:
  socksetup: don't return on set_reuse_addr() error

18 years agoMerge branch 'jc/cc-stat'
Junio C Hamano [Mon, 24 Apr 2006 09:06:40 +0000 (02:06 -0700)]
Merge branch 'jc/cc-stat'

* jc/cc-stat:
  Fix "git show --stat"

18 years agoMerge branch 'jc/unresolve'
Junio C Hamano [Mon, 24 Apr 2006 09:05:55 +0000 (02:05 -0700)]
Merge branch 'jc/unresolve'

* jc/unresolve:
  git-update-index --unresolve
  Add git-unresolve <paths>...

18 years agoMerge branch 'jc/color'
Junio C Hamano [Mon, 24 Apr 2006 09:05:07 +0000 (02:05 -0700)]
Merge branch 'jc/color'

* jc/color:
  Add colordiff for git to contrib/colordiff.

18 years agoMerge git://git.kernel.org/pub/scm/gitk/gitk
Junio C Hamano [Mon, 24 Apr 2006 08:39:01 +0000 (01:39 -0700)]
Merge git://git.kernel.org/pub/scm/gitk/gitk

* git://git.kernel.org/pub/scm/gitk/gitk:
  gitk: Let git-rev-list do the argument list parsing

18 years agoMerge part of 'jc/cache-tree'
Junio C Hamano [Mon, 24 Apr 2006 07:33:28 +0000 (00:33 -0700)]
Merge part of 'jc/cache-tree'

18 years agoAdd test-dump-cache-tree
Junio C Hamano [Mon, 24 Apr 2006 03:20:25 +0000 (20:20 -0700)]
Add test-dump-cache-tree

This was useful in diagnosing the corrupt index.aux format
problem.  But do not bother building or installing it by
default.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUse cache-tree in update-index.
Junio C Hamano [Mon, 24 Apr 2006 07:23:54 +0000 (00:23 -0700)]
Use cache-tree in update-index.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoInvalidate cache-tree entries for touched paths in git-apply.
Junio C Hamano [Sun, 23 Apr 2006 23:52:52 +0000 (16:52 -0700)]
Invalidate cache-tree entries for touched paths in git-apply.

This updates git-apply to maintain cache-tree information.  With
this and the previous write-tree patch, repeated "apply --index"
followed by "write-tree" on a huge tree will hopefully become
faster.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUpdate write-tree to use cache-tree.
Junio C Hamano [Sun, 23 Apr 2006 23:52:35 +0000 (16:52 -0700)]
Update write-tree to use cache-tree.

The updated write-tree reads from $GIT_DIR/index.aux to pick up
subtree objects information, updates the cache-tree with the
index, and updates index.aux file after writing a tree out of
the index file.

Until update-index and other programs that modify the index are
updated to maintain index.aux file, the index.aux file written
by the last write-tree will become stale immediately after they
update the index, which will result in the whole tree
recomputation just like the original write-tree.

The idea is to convert those commands to invalidate cache-tree
whenever they touch the index entries, and write updated
index.aux out.  After the index is updated with them, write-tree
will be able to reuse the parts of the cache-tree that have not
been touched.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd cache-tree.
Junio C Hamano [Sun, 23 Apr 2006 23:52:20 +0000 (16:52 -0700)]
Add cache-tree.

The cache_tree data structure is to cache tree object names that
would result from the current index file.

The idea is to have an optional file to record each tree object
name that corresponds to a directory path in the cache when we
run write_cache(), and read it back when we run read_cache().
During various index manupulations, we selectively invalidate
the parts so that the next write-tree can bypass regenerating
tree objects for unchanged parts of the directory hierarchy.

We could perhaps make the cache-tree data an optional part of
the index file, but that would involve the index format updates,
so unless we need it for performance reasons, the current plan
is to use a separate file, $GIT_DIR/index.aux to store this
information and link it with the index file with the checksum
that is already used for index file integrity check.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'ar/chmod-series' into next
Junio C Hamano [Mon, 24 Apr 2006 00:00:31 +0000 (17:00 -0700)]
Merge branch 'ar/chmod-series' into next

* ar/chmod-series:
  make update-index --chmod work with multiple files and --stdin

18 years agoread-cache/write-cache: optionally return cache checksum SHA1.
Junio C Hamano [Sun, 23 Apr 2006 23:52:08 +0000 (16:52 -0700)]
read-cache/write-cache: optionally return cache checksum SHA1.

read_cache_1() and write_cache_1() takes an extra parameter
*sha1 that returns the checksum of the index file when non-NULL.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogitk: Let git-rev-list do the argument list parsing
Paul Mackerras [Sun, 23 Apr 2006 08:00:24 +0000 (18:00 +1000)]
gitk: Let git-rev-list do the argument list parsing

This is a fix for a problem reported by Jim Radford where an argument
list somewhere overflows on repositories with lots of tags.  In fact
it's now unnecessary to use git-rev-parse since git-rev-list can take
all the arguments that git-rev-parse can.  This is inspired by but not
the same as the solutions suggested by Jim Radford and Linus Torvalds.

Signed-off-by: Paul Mackerras <paulus@samba.org>
18 years agomake update-index --chmod work with multiple files and --stdin
Alex Riesen [Sun, 23 Apr 2006 07:01:29 +0000 (09:01 +0200)]
make update-index --chmod work with multiple files and --stdin

The patch makes "--chmod=-x" and "--chmod=+x" act like "--add"
and "--remove" to affect the behaviour of the command for the
rest of the path parameters, not just the following one.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/diff' into next
Junio C Hamano [Sat, 22 Apr 2006 11:04:46 +0000 (04:04 -0700)]
Merge branch 'jc/diff' into next

* jc/diff:
  Libified diff-index: backward compatibility fix.

18 years agoLibified diff-index: backward compatibility fix.
Junio C Hamano [Sat, 22 Apr 2006 10:58:04 +0000 (03:58 -0700)]
Libified diff-index: backward compatibility fix.

"diff-index -m" does not mean "do not ignore merges", but means
"pretend missing files match the index".

The previous round tried to address this, but failed because
setup_revisions() ate "-m" flag before the caller had a chance
to intervene.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/diff' into next
Junio C Hamano [Sat, 22 Apr 2006 10:07:08 +0000 (03:07 -0700)]
Merge branch 'jc/diff' into next

* jc/diff:
  Libify diff-index.
  Libify diff-files.

18 years agoMerge branch 'jc/fmt-patch' into next
Junio C Hamano [Sat, 22 Apr 2006 10:06:58 +0000 (03:06 -0700)]
Merge branch 'jc/fmt-patch' into next

* jc/fmt-patch:
  git-fmt-patch: thinkofix to show [PATCH] properly.

18 years agogit-fmt-patch: thinkofix to show [PATCH] properly.
Junio C Hamano [Sat, 22 Apr 2006 10:06:13 +0000 (03:06 -0700)]
git-fmt-patch: thinkofix to show [PATCH] properly.

Updating "subject" variable without changing the hardcoded
number of bytes to memcpy from it would not help much.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoLibify diff-index.
Junio C Hamano [Sat, 22 Apr 2006 09:43:00 +0000 (02:43 -0700)]
Libify diff-index.

The second installment to libify diff brothers.  The pathname
arguments are checked more strictly than before because we now
use the revision.c::setup_revisions() infrastructure.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoLibify diff-files.
Junio C Hamano [Sat, 22 Apr 2006 06:57:45 +0000 (23:57 -0700)]
Libify diff-files.

This is the first installment to libify diff brothers.

The updated diff-files uses revision.c::setup_revisions()
infrastructure to parse its command line arguments, which means
the pathname arguments are checked more strictly than before.
The tests are adjusted to separate possibly missing paths from
the rest of arguments with double-dashes, to show the kosher
way.

As Linus pointed out, renaming diff.c to diff-lib.c was simply
stupid, so I am renaming it back.  The new diff-lib.c is to
contain pieces extracted from diff brothers.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/cc-stat' into next
Junio C Hamano [Sat, 22 Apr 2006 08:46:29 +0000 (01:46 -0700)]
Merge branch 'jc/cc-stat' into next

* jc/cc-stat:
  Fix "git show --stat"

18 years agoMerge branch 'jc/color' into next
Junio C Hamano [Sat, 22 Apr 2006 05:25:16 +0000 (22:25 -0700)]
Merge branch 'jc/color' into next

* jc/color:
  Add colordiff for git to contrib/colordiff.
  Makefile: dependency for builtin-help.o

18 years agoFix "git show --stat"
Junio C Hamano [Sat, 22 Apr 2006 02:39:04 +0000 (19:39 -0700)]
Fix "git show --stat"

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd colordiff for git to contrib/colordiff.
Junio C Hamano [Wed, 12 Apr 2006 06:05:14 +0000 (23:05 -0700)]
Add colordiff for git to contrib/colordiff.

I hacked it up to teach it the git extended diff headers, made
it not to read the whole patch in the array.

Also, the original program, when arguments are given, ran "diff"
with the given arguments and showed the output from it.  Of
course, I changed it to run "git diff" ;-).

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/diff'
Junio C Hamano [Sat, 22 Apr 2006 05:03:07 +0000 (22:03 -0700)]
Merge branch 'jc/diff'

* jc/diff:
  diff --stat: do not drop rename information.

18 years agoMakefile: dependency for builtin-help.o
Junio C Hamano [Sat, 22 Apr 2006 04:56:13 +0000 (21:56 -0700)]
Makefile: dependency for builtin-help.o

builtin-help.c wants to include common-cmds.h which is a
generated file.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/fmt-patch' into next
Junio C Hamano [Fri, 21 Apr 2006 20:32:58 +0000 (13:32 -0700)]
Merge branch 'jc/fmt-patch' into next

* jc/fmt-patch:
  Split up builtin commands into separate files from git.c
  git-log produces no output

18 years agoMerge branch 'master' into jc/fmt-patch
Junio C Hamano [Fri, 21 Apr 2006 20:19:58 +0000 (13:19 -0700)]
Merge branch 'master' into jc/fmt-patch

* master:
  Split up builtin commands into separate files from git.c
  git-log produces no output
  fix pack-object buffer size
  mailinfo: decode underscore used in "Q" encoding properly.
  Reintroduce svn pools to solve the memory leak.
  pack-objects: do not stop at object that is "too small"
  git-commit --amend: two fixes.
  get_tree_entry(): make it available from tree-walk
  sha1_name.c: no need to include diff.h; tree-walk.h will do.
  sha1_name.c: prepare to make get_tree_entry() reusable from others.
  get_sha1() shorthands for blob/tree objects
  pre-commit hook: complain about conflict markers.
  git-merge: a bit more readable user guidance.
  diff: move diff.c to diff-lib.c to make room.
  git log: don't do merge diffs by default
  Allow "git repack" users to specify repacking window/depth
  Document git-clone --reference
  Fix filename scaling for binary files
  Fix uninteresting tags in new revision parsing

Conflicts:

    Adjusted the addition of fmt-patch to match the recent split
    from git.c to builtin.log.c.

18 years agoSplit up builtin commands into separate files from git.c
Linus Torvalds [Fri, 21 Apr 2006 17:27:34 +0000 (10:27 -0700)]
Split up builtin commands into separate files from git.c

Right now it split it into "builtin-log.c" for log-related commands
("log", "show" and "whatchanged"), and "builtin-help.c" for the
informational commands (usage printing and "help" and "version").

This just makes things easier to read, I find.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'fix'
Junio C Hamano [Fri, 21 Apr 2006 20:14:30 +0000 (13:14 -0700)]
Merge branch 'fix'

* fix:
  git-log produces no output

18 years agogit-log produces no output
Linus Torvalds [Fri, 21 Apr 2006 19:25:13 +0000 (12:25 -0700)]
git-log produces no output

When $PAGER is set to 'less -i', we used to fail because we
assumed the $PAGER is a command and simply exec'ed it.

Try exec first, and then run it through shell if it fails.  This
allows even funkier PAGERs like these ;-):

PAGER='sed -e "s/^/`date`: /" | more'
PAGER='contrib/colordiff.perl | less -RS'

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Fri, 21 Apr 2006 07:45:55 +0000 (00:45 -0700)]
Merge branch 'master' into next

* master:
  fix pack-object buffer size
  mailinfo: decode underscore used in "Q" encoding properly.
  Reintroduce svn pools to solve the memory leak.
  pack-objects: do not stop at object that is "too small"

18 years agoMerge branch 'fix'
Junio C Hamano [Fri, 21 Apr 2006 07:45:40 +0000 (00:45 -0700)]
Merge branch 'fix'

* fix:
  fix pack-object buffer size
  mailinfo: decode underscore used in "Q" encoding properly.
  Reintroduce svn pools to solve the memory leak.
  pack-objects: do not stop at object that is "too small"

18 years agofix pack-object buffer size
Nicolas Pitre [Thu, 20 Apr 2006 21:25:37 +0000 (17:25 -0400)]
fix pack-object buffer size

The input line has 40 _chars_ of sha1 and no 20 _bytes_. It should also
account for the space before the pathname, and the terminating \n and \0.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agomailinfo: decode underscore used in "Q" encoding properly.
Junio C Hamano [Fri, 21 Apr 2006 07:06:58 +0000 (00:06 -0700)]
mailinfo: decode underscore used in "Q" encoding properly.

Quoted-Printable (RFC 2045) and the "Q" encoding (RFC 2047) are
subtly different; the latter is used on the mail header and an
underscore needs to be decoded to 0x20.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoReintroduce svn pools to solve the memory leak.
Santi_Béjar [Mon, 27 Mar 2006 11:26:01 +0000 (13:26 +0200)]
Reintroduce svn pools to solve the memory leak.

Introduced in 4802426.

Signed-off-by: Santi Béjar <sbejar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agopack-objects: do not stop at object that is "too small"
Junio C Hamano [Fri, 21 Apr 2006 06:36:22 +0000 (23:36 -0700)]
pack-objects: do not stop at object that is "too small"

Because we sort the delta window by name-hash and then size,
hitting an object that is too small to consider as a delta base
for the current object does not mean we do not have better
candidate in the window beyond it.

Noticed by Shawn Pearce, analyzed by Nico, Linus and me.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/fmt-patch' into next
Junio C Hamano [Thu, 20 Apr 2006 19:45:16 +0000 (12:45 -0700)]
Merge branch 'jc/fmt-patch' into next

* jc/fmt-patch:
  rename internal format-patch wip

18 years agorename internal format-patch wip
Junio C Hamano [Thu, 20 Apr 2006 19:45:02 +0000 (12:45 -0700)]
rename internal format-patch wip

Otherwise "git format-patch" would invoke unfinished internal one that
does only --stdout

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Thu, 20 Apr 2006 09:53:03 +0000 (02:53 -0700)]
Merge branch 'master' into next

* master:
  git-commit --amend: two fixes.

18 years agoMerge branch 'fix'
Junio C Hamano [Thu, 20 Apr 2006 09:52:04 +0000 (02:52 -0700)]
Merge branch 'fix'

* fix:
  git-commit --amend: two fixes.

18 years agogit-commit --amend: two fixes.
Junio C Hamano [Thu, 20 Apr 2006 08:20:56 +0000 (01:20 -0700)]
git-commit --amend: two fixes.

When running "git commit --amend" only to fix the commit log
message without any content change, we mistakenly showed the
git-status output that says "nothing to commit" without
commenting it out.

If you have already run update-index but you want to amend the
top commit, "git commit --amend --only" without any paths should
have worked, because --only means "starting from the base
commit, update-index these paths only to prepare the index to
commit, and perform the commit".  However, we refused -o without
paths.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoStop carrying Geert's similarity estimator.
Junio C Hamano [Thu, 20 Apr 2006 07:33:37 +0000 (00:33 -0700)]
Stop carrying Geert's similarity estimator.

I tried the code with pack-objects.c::try_delta(), and was
somewhat dissapointed.  The current type-path based heuristics
already limits the delta attempts to similar objects anyway, so
it is not a good place to apply it.

The Net never forgets, so we can resurrect it if we wanted to
later.

18 years agoMerge branch 'jc/unresolve' into next
Junio C Hamano [Thu, 20 Apr 2006 06:52:30 +0000 (23:52 -0700)]
Merge branch 'jc/unresolve' into next

* jc/unresolve:
  git-update-index --unresolve

18 years agogit-update-index --unresolve
Junio C Hamano [Thu, 20 Apr 2006 06:52:05 +0000 (23:52 -0700)]
git-update-index --unresolve

Retire git-unresolve and make it into "git-update-index --unresolve".
It processes all paths that follow.

During a merge, you would mark a path that is dealt with with:

$ git update-index hello

and you would "undo" it with:

$ git update-index --unresolve hello

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/diff' into next
Junio C Hamano [Thu, 20 Apr 2006 06:22:40 +0000 (23:22 -0700)]
Merge branch 'jc/diff' into next

* jc/diff:
  diff --stat: do not drop rename information.

18 years agodiff --stat: do not drop rename information.
Junio C Hamano [Thu, 20 Apr 2006 03:41:14 +0000 (20:41 -0700)]
diff --stat: do not drop rename information.

When a verbatim rename or copy is detected, we did not show
anything on the "diff --stat" for the filepair.  This makes it
to show the rename information.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/fmt-patch' into next
Junio C Hamano [Thu, 20 Apr 2006 03:04:09 +0000 (20:04 -0700)]
Merge branch 'jc/fmt-patch' into next

* jc/fmt-patch:
  Minor tweak on subject line in --pretty=email

18 years agoMerge branch 'jc/unresolve' into next
Junio C Hamano [Wed, 19 Apr 2006 23:12:57 +0000 (16:12 -0700)]
Merge branch 'jc/unresolve' into next

* jc/unresolve:
  Add git-unresolve <paths>...
  get_tree_entry(): make it available from tree-walk
  sha1_name.c: no need to include diff.h; tree-walk.h will do.
  sha1_name.c: prepare to make get_tree_entry() reusable from others.
  pre-commit hook: complain about conflict markers.
  git-merge: a bit more readable user guidance.
  diff: move diff.c to diff-lib.c to make room.
  git log: don't do merge diffs by default
  Allow "git repack" users to specify repacking window/depth

18 years agoAdd git-unresolve <paths>...
Junio C Hamano [Wed, 19 Apr 2006 19:51:29 +0000 (12:51 -0700)]
Add git-unresolve <paths>...

This is an attempt to address the issue raised on #git channel
recently by Carl Worth.

After a conflicted automerge, "git diff" shows a combined diff
to give you how the tentative automerge result differs from
what came from each branch.  During a complex merge, it is
tempting to be able to resolve a few paths at a time, mark
them "I've dealt with them" with git-update-index to unclutter
the next "git diff" output, and keep going.  However, when the
final result does not compile or otherwise found to be a
mismerge, the workflow to fix the mismerged paths suddenly
changes to "git diff HEAD -- path" (to get a diff from our
HEAD before merging) and "git diff MERGE_HEAD -- path" (to get
a diff from theirs), and it cannot show the combined anymore.

With git-unresolve <paths>..., the versions from our branch and
their branch for specified blobs are placed in stage #2 and
stage #3, without touching the working tree files.  This gives
you the combined diff back for easier review, along with
"diff --ours" and "diff --theirs".

One thing it does not do is to place the base in stage #1; this
means "diff --base" would behave differently between the run
immediately after a conflicted three-way merge, and the run
after an update-index by mistake followed by a git-unresolve.

We could theoretically run merge-base between HEAD and
MERGE_HEAD to find which tree to place in stage #1, but
reviewing "diff --base" is not that useful so....

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'lt/xsha1'
Junio C Hamano [Wed, 19 Apr 2006 22:57:45 +0000 (15:57 -0700)]
Merge branch 'lt/xsha1'

* lt/xsha1:
  get_tree_entry(): make it available from tree-walk
  sha1_name.c: no need to include diff.h; tree-walk.h will do.
  sha1_name.c: prepare to make get_tree_entry() reusable from others.
  get_sha1() shorthands for blob/tree objects

18 years agoMerge branch 'maint'
Junio C Hamano [Wed, 19 Apr 2006 22:41:55 +0000 (15:41 -0700)]
Merge branch 'maint'

* maint:
  pre-commit hook: complain about conflict markers.
  git-merge: a bit more readable user guidance.

18 years agodiff: move diff.c to diff-lib.c to make room.
Junio C Hamano [Wed, 19 Apr 2006 22:12:02 +0000 (15:12 -0700)]
diff: move diff.c to diff-lib.c to make room.

Now I am not doing any real "git-diff in C" yet, but this would
help before doing so.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agopre-commit hook: complain about conflict markers.
Junio C Hamano [Wed, 19 Apr 2006 21:58:24 +0000 (14:58 -0700)]
pre-commit hook: complain about conflict markers.

Several <<< or === or >>> characters at the beginning of a line
is very likely to be leftover conflict markers from a failed
automerge the user resolved incorrectly, so detect them.

As usual, this can be defeated with "git commit --no-verify" if
you really do want to have those files, just like changes that
introduce trailing whitespaces.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-merge: a bit more readable user guidance.
Junio C Hamano [Wed, 19 Apr 2006 21:54:27 +0000 (14:54 -0700)]
git-merge: a bit more readable user guidance.

We said "fix up by hand" after failed automerge, which was a big
"Huh?  Now what?".  Be a bit more explicit without being too
verbose. Suggested by Carl Worth.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMinor tweak on subject line in --pretty=email
Junio C Hamano [Wed, 19 Apr 2006 22:16:08 +0000 (15:16 -0700)]
Minor tweak on subject line in --pretty=email

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoget_tree_entry(): make it available from tree-walk
Junio C Hamano [Wed, 19 Apr 2006 21:05:47 +0000 (14:05 -0700)]
get_tree_entry(): make it available from tree-walk

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosha1_name.c: no need to include diff.h; tree-walk.h will do.
Junio C Hamano [Wed, 19 Apr 2006 18:56:53 +0000 (11:56 -0700)]
sha1_name.c: no need to include diff.h; tree-walk.h will do.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosha1_name.c: prepare to make get_tree_entry() reusable from others.
Junio C Hamano [Wed, 19 Apr 2006 18:56:07 +0000 (11:56 -0700)]
sha1_name.c: prepare to make get_tree_entry() reusable from others.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit log: don't do merge diffs by default
Linus Torvalds [Wed, 19 Apr 2006 17:20:49 +0000 (10:20 -0700)]
git log: don't do merge diffs by default

I personally prefer "ignore_merges" to be on by default, because quite
often the merge diff is distracting and not interesting. That's true both
with "-p" and with "--stat" output.

If you want output from merges, you can trivially use the "-m", "-c" or
"--cc" flags to tell that you're interested in merges, which also tells
the diff generator what kind of diff to do (for --stat, any of the three
will do, of course, but they differ for plain patches or for
--patch-with-stat).

This trivial patch just removes the two lines that tells "git log" not to
ignore merges. It will still show the commit log message, of course, due
to the "always_show_header" part.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAllow "git repack" users to specify repacking window/depth
Linus Torvalds [Wed, 19 Apr 2006 17:05:12 +0000 (10:05 -0700)]
Allow "git repack" users to specify repacking window/depth

.. but don't even bother documenting it. I don't think any normal person
is supposed to ever really care, but it simplifies testing when you want
to use the "git repack" wrapper rather than forcing you to use the core
programs (which already do support the window/depth arguments, of course).

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Wed, 19 Apr 2006 09:25:36 +0000 (02:25 -0700)]
Merge branch 'master' into next

* master:
  Document git-clone --reference
  Fix filename scaling for binary files

18 years agoMerge branch 'fix'
Junio C Hamano [Wed, 19 Apr 2006 09:25:29 +0000 (02:25 -0700)]
Merge branch 'fix'

* fix:
  Document git-clone --reference
  Fix filename scaling for binary files

18 years agoMerge branch 'lt/xsha1' into next
Junio C Hamano [Wed, 19 Apr 2006 05:02:04 +0000 (22:02 -0700)]
Merge branch 'lt/xsha1' into next

* lt/xsha1:
  get_sha1() shorthands for blob/tree objects
  Fix uninteresting tags in new revision parsing

18 years agoget_sha1() shorthands for blob/tree objects
Linus Torvalds [Tue, 18 Apr 2006 23:45:16 +0000 (16:45 -0700)]
get_sha1() shorthands for blob/tree objects

This is a fairly straightforward patch to allow "get_sha1()" to also have
shorthands for tree and blob objects.

The syntax is very simple and intuitive: you can specify a tree or a blob
by simply specifying <revision>:<path>, and get_sha1() will do the SHA1
lookup from the tree for you.

You can currently do it with "git ls-tree <rev> <path>" and parsing the
output, but that's actually pretty awkward.

With this, you can do something like

git cat-file blob v1.2.4:Makefile

to get the contents of "Makefile" at revision v1.2.4.

Now, this isn't necessarily something you really need all that often, but
the concept itself is actually pretty powerful. We could, for example,
allow things like

git diff v0.99.6:git-commit-script..v1.3.0:git-commit.sh

to see the difference between two arbitrary files in two arbitrary
revisions. To do that, the only thing we'd have to do is to make
git-diff-tree accept two blobs to diff, in addition to the two trees it
now expects.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix uninteresting tags in new revision parsing
Linus Torvalds [Wed, 19 Apr 2006 03:31:41 +0000 (20:31 -0700)]
Fix uninteresting tags in new revision parsing

When I unified the revision argument parsing, I introduced a simple bug
wrt tags that had been marked uninteresting. When it was preparing for the
revision walk, it would mark all the parent commits of an uninteresting
tag correctly uninteresting, but it would forget about the commit itself.

This means that when I just did my 2.6.17-rc2 release, and my scripts
generated the log for "v2.6.17-rc1..v2.6.17-rc2", everything was fine,
except the commit pointed to by 2.6.17-rc1 (which shouldn't have been
there) was included. Even though it should obviously have been marked as
being uninteresting.

Not a huge deal, and the fix is trivial.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocument git-clone --reference
Shawn Pearce [Wed, 19 Apr 2006 00:19:48 +0000 (20:19 -0400)]
Document git-clone --reference

The new --reference flag introduced to git-clone in
GIT 1.3.0 was not documented but is rather handy.
So document it.

Also corrected a minor issue with the documentation for the
-s flag; the info/alternates file name was spelled wrong.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/fmt-patch' into next
Junio C Hamano [Tue, 18 Apr 2006 23:45:55 +0000 (16:45 -0700)]
Merge branch 'jc/fmt-patch' into next

* jc/fmt-patch:
  Tentative built-in format-patch.

18 years agoTentative built-in format-patch.
Junio C Hamano [Tue, 18 Apr 2006 23:45:27 +0000 (16:45 -0700)]
Tentative built-in format-patch.

This only does --stdout right now.  To write into separate files
with pretty-printed filenames like the real thing does, it needs
a bit mroe work.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix filename scaling for binary files
Jonas Fonseca [Tue, 18 Apr 2006 21:26:43 +0000 (23:26 +0200)]
Fix filename scaling for binary files

Set maximum filename length for binary files so that scaling won't be
triggered and result in invalid string access.

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'sh/daemon' into next
Junio C Hamano [Tue, 18 Apr 2006 21:41:57 +0000 (14:41 -0700)]
Merge branch 'sh/daemon' into next

* sh/daemon:
  socksetup: don't return on set_reuse_addr() error

18 years agosocksetup: don't return on set_reuse_addr() error
Serge E. Hallyn [Tue, 18 Apr 2006 13:11:06 +0000 (08:11 -0500)]
socksetup: don't return on set_reuse_addr() error

The set_reuse_addr() error case was the only error case in
socklist() where we returned rather than continued.  Not sure
why.  Either we must free the socklist, or continue.  This patch
continues on error.

Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/pager-cat'
Junio C Hamano [Tue, 18 Apr 2006 20:57:22 +0000 (13:57 -0700)]
Merge branch 'jc/pager-cat'

* jc/pager-cat:
  Do not fork PAGER=cat

18 years agoMerge branch 'lt/logopt'
Junio C Hamano [Tue, 18 Apr 2006 20:56:36 +0000 (13:56 -0700)]
Merge branch 'lt/logopt'

* lt/logopt:
  Fix "git log --stat": make sure to set recursive with --stat.
  combine-diff: show diffstat with the first parent.
  git.c: LOGSIZE is unused after log printing cleanup.
  Log message printout cleanups (#3): fix --pretty=oneline
  Log message printout cleanups (#2)
  Log message printout cleanups
  rev-list --header: output format fix
  Fixes for option parsing
  log/whatchanged/show - log formatting cleanup.
  Simplify common default options setup for built-in log family.
  Tentative built-in "git show"
  Built-in git-whatchanged.
  rev-list option parser fix.
  Split init_revisions() out of setup_revisions()
  Fix up rev-list option parsing.
  Fix up default abbrev in setup_revisions() argument parser.
  Common option parsing for "git log --diff" and friends

18 years agoMerge branch 'master' into next
Junio C Hamano [Tue, 18 Apr 2006 20:52:46 +0000 (13:52 -0700)]
Merge branch 'master' into next

* master:
  GIT 1.3.0
  Add git-annotate(1) and git-blame(1)
  diff --stat: make sure to set recursive.
  git-svnimport symlink support

18 years agoMerge branch 'lt/logopt' into next
Junio C Hamano [Tue, 18 Apr 2006 20:18:21 +0000 (13:18 -0700)]
Merge branch 'lt/logopt' into next

* lt/logopt:
  Fix "git log --stat": make sure to set recursive with --stat.

18 years agoGIT 1.3.0 v1.3.0
Junio C Hamano [Tue, 18 Apr 2006 20:15:18 +0000 (13:15 -0700)]
GIT 1.3.0

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd git-annotate(1) and git-blame(1)
Jonas Fonseca [Mon, 6 Mar 2006 05:24:44 +0000 (06:24 +0100)]
Add git-annotate(1) and git-blame(1)

[jc: with entries in git.txt]

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix "git log --stat": make sure to set recursive with --stat.
Junio C Hamano [Tue, 18 Apr 2006 18:41:28 +0000 (11:41 -0700)]
Fix "git log --stat": make sure to set recursive with --stat.

Just like "patch" format always needs recursive, "diffstat"
format does not make sense without setting recursive.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodiff --stat: make sure to set recursive.
Junio C Hamano [Tue, 18 Apr 2006 18:29:33 +0000 (11:29 -0700)]
diff --stat: make sure to set recursive.

Just like "patch" format always needs recursive, "diffstat"
format does not make sense without setting recursive.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svnimport symlink support
Herbert Valerio Riedel [Mon, 17 Apr 2006 10:58:39 +0000 (06:58 -0400)]
git-svnimport symlink support

added svn:special symlink support for access methods other than
direct-http

Signed-off-by: Herbert Valerio Riedel <hvr@gnu.org>
Acked-by: Matthias Urlichs <smurf@smurf.noris.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'lt/logopt' into next
Junio C Hamano [Tue, 18 Apr 2006 06:15:41 +0000 (23:15 -0700)]
Merge branch 'lt/logopt' into next

* lt/logopt:
  combine-diff: show diffstat with the first parent.
  git.c: LOGSIZE is unused after log printing cleanup.
  Log message printout cleanups (#3): fix --pretty=oneline
  Log message printout cleanups (#2)
  Log message printout cleanups
  rev-list --header: output format fix

18 years agoMerge branch 'master' into next
Junio C Hamano [Tue, 18 Apr 2006 06:15:20 +0000 (23:15 -0700)]
Merge branch 'master' into next

* master:
  packed_object_info_detail(): check for corrupt packfile.
  cleanups: remove unused variable from exec_cmd.c
  cleanups: prevent leak of two strduped strings in config.c
  cleanups: Remove impossible case in quote.c
  cleanups: Remove unused vars from combine-diff.c
  cleanups: Fix potential bugs in connect.c
  Allow empty lines in info/grafts

18 years agocombine-diff: show diffstat with the first parent.
Junio C Hamano [Tue, 18 Apr 2006 05:53:03 +0000 (22:53 -0700)]
combine-diff: show diffstat with the first parent.

Asking for stat (either with --stat or --patch-with-stat) gives
you diffstat for the first parent, even under combine-diff.

While the combined patch is useful to highlight the complexity
and interaction of the parts touched by all branches when
reviewing a merge commit, diffstat is a tool to assess the
extent of damage the merge brings in, and showing stat with the
first parent is more sensible than clever per-parent diffstat.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit.c: LOGSIZE is unused after log printing cleanup.
Junio C Hamano [Tue, 18 Apr 2006 04:47:35 +0000 (21:47 -0700)]
git.c: LOGSIZE is unused after log printing cleanup.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoLog message printout cleanups (#3): fix --pretty=oneline
Linus Torvalds [Tue, 18 Apr 2006 00:43:40 +0000 (17:43 -0700)]
Log message printout cleanups (#3): fix --pretty=oneline

This option is very special, since pretty_print_commit() will _remove_
the newline at the end of it, so we want to have an extra separator
between the things.

I added a honking big comment this time, so that (a) I don't forget this
_again_ (I broke "oneline" several times during this printout cleanup),
and so that people can understand _why_ the code does what it does.

Now, arguably the alternate fix is to always have the '\n' at the end in
pretty-print-commit, but git-rev-list depends on the current behaviour
(but we could have git-rev-list remove it, whatever).

With the big comment, the code hopefully doesn't get broken again. And now
things like

git log --pretty=oneline --cc --patch-with-stat

works (even if that is admittedly a totally insane combination: if you
want the patch, having the "oneline" log format is just crazy, but hey,
it _works_. Even insane people are people).

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoLog message printout cleanups (#2)
Linus Torvalds [Mon, 17 Apr 2006 23:59:42 +0000 (16:59 -0700)]
Log message printout cleanups (#2)

Here's a further patch on top of the previous one with cosmetic
improvements (no "real" code changes, just trivial updates):

 - it gets the "---" before a diffstat right, including for the combined
   merge case. Righ now the logic is that we always use "---" when we have
   a diffstat, and an empty line otherwise. That's how I visually prefer
   it, but hey, it can be tweaked later.

 - I made "diff --cc/combined" add the "---/+++" header lines too. The
   thing won't be mistaken for a valid diff, since the "@@" lines have too
   many "@" characters (three or more), but it just makes it visually
   match a real diff, which at least to me makes a big difference in
   readability. Without them, it just looks very "wrong".

   I guess I should have taken the filename from each individual entry
   (and had one "---" file per parent), but I didn't even bother to try to
   see how that works, so this was the simple thing.

With this, doing a

git log --cc --patch-with-stat

looks quite readable, I think. The only nagging issue - as far as I'm
concerned - is that diffstats for merges are pretty questionable the way
they are done now. I suspect it would be better to just have the _first_
diffstat, and always make the merge diffstat be the one for "result
against first parent".

Signed-off-by: Junio C Hamano <junkio@cox.net>