git.git
18 years agopack-object: slightly more efficient
Nicolas Pitre [Mon, 15 May 2006 17:47:16 +0000 (13:47 -0400)]
pack-object: slightly more efficient

Avoid creating a delta index for objects with maximum depth since they
are not going to be used as delta base anyway.  This also reduce peak
memory usage slightly as the current object's delta index is not useful
until the next object in the loop is considered for deltification. This
saves a bit more than 1% on CPU usage.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosimple euristic for further free packing improvements
Nicolas Pitre [Mon, 15 May 2006 15:40:05 +0000 (11:40 -0400)]
simple euristic for further free packing improvements

Given that the early eviction of objects with maximum delta depth
may exhibit bad packing on its own, why not considering a bias against
deep base objects in try_delta() to mitigate that bad behavior.

This patch adjust the MAX_size allowed for a delta based on the depth of
the base object as well as enabling the early eviction of max depth
objects from the object window.  When used separately, those two things
produce slightly better and much worse results respectively.  But their
combined effect is a surprising significant packing improvement.

With this really simple patch the GIT repo gets nearly 15% smaller, and
the Linux kernel repo about 5% smaller, with no significantly measurable
CPU usage difference.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoread-tree --reset -u fix.
Linus Torvalds [Mon, 15 May 2006 15:09:31 +0000 (08:09 -0700)]
read-tree --reset -u fix.

The previous commit makes -u to mean "I do want to remove the
local changes, just update it from the read tree" only for
one-way merge.  It makes sense to have it depend on the
"--reset" flag instead.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoread-tree -u one-way merge fix to check out locally modified paths.
Junio C Hamano [Mon, 15 May 2006 07:46:05 +0000 (00:46 -0700)]
read-tree -u one-way merge fix to check out locally modified paths.

The "-u" flag means "update the working tree files", but to
other types of merges, it also implies "I want to keep my local
changes" -- because they prevent local changes from getting lost
by using verify_uptodate.  The one-way merge is different from
other merges in that its purpose is opposite of doing something
else while keeping unrelated local changes.  The point of
one-way merge is to nuke local changes.  So while it feels
somewhat wrong that this actively loses local changes, it is the
right thing to do.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSimplify "git reset --hard"
Linus Torvalds [Sun, 14 May 2006 18:20:37 +0000 (11:20 -0700)]
Simplify "git reset --hard"

Now that the one-way merge strategy does the right thing wrt files that do
not exist in the result, just remove all the random crud we did in "git
reset" to do this all by hand.

Instead, just pass in "-u" to git-read-tree when we do a hard reset, and
depend on git-read-tree to update the working tree appropriately.

This basically means that git reset turns into

# Always update the HEAD ref
git update-ref HEAD "$rev"

case "--soft"
# do nothing to index/working tree
case "--hard"
# read index _and_ update working tree
git-read-tree --reset -u "$rev"
case "--mixed"
# update just index, report on working tree differences
git-read-tree --reset "$rev"
git-update-index --refresh

which is what it was always semantically doing, it just did it in a
rather strange way because it was written to not expect git-read-tree to
do anything to the working tree.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAllow one-way tree merge to remove old files
Linus Torvalds [Sun, 14 May 2006 17:43:50 +0000 (10:43 -0700)]
Allow one-way tree merge to remove old files

For some random reason (probably just because nobody noticed), the one-way
merge strategy didn't mark deleted files as deleted, so if you used

git-read-tree -m -u <newtree>

it would update the files that got changed in the index, but it would not
delete the files that got deleted.

This should fix it, and I can't imagine that anybody depends on the old
strange "update only existing files" behaviour.

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 [Sun, 14 May 2006 23:20:15 +0000 (16:20 -0700)]
Merge branch 'fix'

* fix:
  include header to define uint32_t, necessary on Mac OS X

18 years agoinclude header to define uint32_t, necessary on Mac OS X
Ben Clifford [Sun, 14 May 2006 20:34:56 +0000 (21:34 +0100)]
include header to define uint32_t, necessary on Mac OS X

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'ml/cvs'
Junio C Hamano [Sun, 14 May 2006 07:42:16 +0000 (00:42 -0700)]
Merge branch 'ml/cvs'

* ml/cvs:
  Change to allow subdir updates from Eclipse
  Many fixes for most operations in Eclipse.
  Added logged warnings for CVS error returns
  cvsserver: use git-rev-list instead of git-log
  git-cvsexportcommit: Add -f(orce) and -m(essage prefix) flags, small cleanups.

18 years agoMerge branch 'fix'
Junio C Hamano [Sun, 14 May 2006 05:24:18 +0000 (22:24 -0700)]
Merge branch 'fix'

* fix:
  Fix git-pack-objects for 64-bit platforms

18 years agoFix git-pack-objects for 64-bit platforms
Dennis Stosberg [Thu, 11 May 2006 17:36:32 +0000 (19:36 +0200)]
Fix git-pack-objects for 64-bit platforms

The offset of an object in the pack is recorded as a 4-byte integer
in the index file.  When reading the offset from the mmap'ed index
in prepare_pack_revindex(), the address is dereferenced as a long*.
This works fine as long as the long type is four bytes wide.  On
NetBSD/sparc64, however, a long is 8 bytes wide and so dereferencing
the offset produces garbage.

[jc: taking suggestion by Linus to use uint32_t]

Signed-off-by: Dennis Stosberg <dennis@stosberg.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agofix diff-delta bad memory access
Nicolas Pitre [Wed, 10 May 2006 16:26:08 +0000 (12:26 -0400)]
fix diff-delta bad memory access

It cannot be assumed that the given buffer will never be moved when
shrinking the allocated memory size with realloc().  So let's ignore
that optimization for now.

This patch makes Electric Fence happy on Linux.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Acked-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'fix'
Junio C Hamano [Wed, 10 May 2006 02:24:16 +0000 (19:24 -0700)]
Merge branch 'fix'

* fix:
  checkout: use --aggressive when running a 3-way merge (-m).
  revert/cherry-pick: use aggressive merge.

18 years agocheckout: use --aggressive when running a 3-way merge (-m).
Junio C Hamano [Wed, 10 May 2006 02:23:23 +0000 (19:23 -0700)]
checkout: use --aggressive when running a 3-way merge (-m).

After doing an in-index 3-way merge, we always do the stock
"merge-index merge-one-file" without doing anything fancy;
use of --aggressive helps performance quite a bit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorevert/cherry-pick: use aggressive merge.
Linus Torvalds [Wed, 10 May 2006 02:22:25 +0000 (19:22 -0700)]
revert/cherry-pick: use aggressive merge.

After doing an in-index 3-way merge, we always do the stock
"merge-index merge-one-file" without doing anything fancy;
use of --aggressive helps performance quite a bit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/clean'
Junio C Hamano [Tue, 9 May 2006 23:52:54 +0000 (16:52 -0700)]
Merge branch 'jc/clean'

* jc/clean:
  Teach git-clean optional <paths>... parameters.

18 years agoMerge branch 'mw/alternates'
Junio C Hamano [Tue, 9 May 2006 23:45:45 +0000 (16:45 -0700)]
Merge branch 'mw/alternates'

* mw/alternates:
  clone: don't clone the info/alternates file
  test case for transitive info/alternates
  Transitively read alternatives

18 years agoMerge branch 'jc/xsha1'
Junio C Hamano [Tue, 9 May 2006 23:44:59 +0000 (16:44 -0700)]
Merge branch 'jc/xsha1'

* jc/xsha1:
  get_sha1() - fix infinite loop on nonexistent stage.
  get_sha1(): :path and :[0-3]:path to extract from index.

18 years agoMerge branch 'jc/again'
Junio C Hamano [Tue, 9 May 2006 23:40:53 +0000 (16:40 -0700)]
Merge branch 'jc/again'

* jc/again:
  Fix users of prefix_path() to free() only when necessary
  update-index --again: take optional pathspecs
  update-index --again

18 years agoMerge branch 'np/delta'
Junio C Hamano [Tue, 9 May 2006 23:40:28 +0000 (16:40 -0700)]
Merge branch 'np/delta'

* np/delta:
  improve diff-delta with sparse and/or repetitive data
  tiny optimization to diff-delta
  replace adler32 with Rabin's polynomial in diff-delta
  use delta index data when finding best delta matches
  split the diff-delta interface

18 years agoMerge branch 'jc/bindiff'
Junio C Hamano [Tue, 9 May 2006 21:16:56 +0000 (14:16 -0700)]
Merge branch 'jc/bindiff'

* jc/bindiff:
  improve base85 generated assembly code
  binary diff and apply: testsuite.
  binary diff: further updates.
  binary patch.

18 years agoMerge branch 'fix'
Junio C Hamano [Tue, 9 May 2006 20:54:42 +0000 (13:54 -0700)]
Merge branch 'fix'

* fix:
  read-cache.c: use xcalloc() not calloc()
  apply: fix infinite loop with multiple patches with --index

18 years agoMerge branch 'tojunio' of http://locke.catalyst.net.nz/git/git-martinlanghoff into...
Junio C Hamano [Tue, 9 May 2006 19:25:21 +0000 (12:25 -0700)]
Merge branch 'tojunio' of locke.catalyst.net.nz/git/git-martinlanghoff into ml/cvs

* 'tojunio' of http://locke.catalyst.net.nz/git/git-martinlanghoff:
  Change to allow subdir updates from Eclipse
  Many fixes for most operations in Eclipse.
  Added logged warnings for CVS error returns
  cvsserver: use git-rev-list instead of git-log
  git-cvsexportcommit: Add -f(orce) and -m(essage prefix) flags, small cleanups.

18 years agoread-cache.c: use xcalloc() not calloc()
Yakov Lerner [Tue, 9 May 2006 16:14:00 +0000 (16:14 +0000)]
read-cache.c: use xcalloc() not calloc()

Elsewhere we use xcalloc(); we should consistently do so.

Signed-off-by: Yakov Lerner <iler.ml@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoapply: fix infinite loop with multiple patches with --index
Eric Wong [Tue, 9 May 2006 08:08:23 +0000 (01:08 -0700)]
apply: fix infinite loop with multiple patches with --index

When multiple patches are passed to git-apply, it will attempt
to open multiple file descriptors to an index, which means
multiple entries will be in the circular cache_file_list.

This change makes git-apply only open the index once and
write the index at exit.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTeach git-clean optional <paths>... parameters.
Junio C Hamano [Mon, 8 May 2006 19:02:44 +0000 (12:02 -0700)]
Teach git-clean optional <paths>... parameters.

When optional paths arguments are given, git-clean passes them
to underlying git-ls-files; with this, you can say:

git clean 'temp-*'

to clean only the garbage files whose names begin with 'temp-'.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Pavel Roskin <proski@gnu.org>
18 years agoMerge branch 'fix'
Junio C Hamano [Mon, 8 May 2006 23:40:23 +0000 (16:40 -0700)]
Merge branch 'fix'

* fix:
  Separate object name errors from usage errors
  Documentation: {caret} fixes (git-rev-list.txt)
  Fix "git diff --stat" with long filenames
  Fix repo-config set-multivar error return path.

18 years agoSeparate object name errors from usage errors
Dmitry V. Levin [Mon, 8 May 2006 21:43:38 +0000 (01:43 +0400)]
Separate object name errors from usage errors

Separate object name errors from usage errors.

Signed-off-by: Dmitry V. Levin <ldv@altlinux.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoget_sha1() - fix infinite loop on nonexistent stage.
Junio C Hamano [Mon, 8 May 2006 22:44:06 +0000 (15:44 -0700)]
get_sha1() - fix infinite loop on nonexistent stage.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation: {caret} fixes (git-rev-list.txt)
Junio C Hamano [Mon, 8 May 2006 20:46:53 +0000 (13:46 -0700)]
Documentation: {caret} fixes (git-rev-list.txt)

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoimprove base85 generated assembly code
Nicolas Pitre [Mon, 8 May 2006 15:31:11 +0000 (11:31 -0400)]
improve base85 generated assembly code

This code is arguably pretty hot, if you use binary patches of course.
This patch helps gcc generate both smaller and faster code especially in
the error free path.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix "git diff --stat" with long filenames
Linus Torvalds [Mon, 8 May 2006 16:46:53 +0000 (09:46 -0700)]
Fix "git diff --stat" with long filenames

When we cut off the front of a filename to make it fit on the line, we add
a "..." in front. However, the way the "git diff" code was written, we
will never reset the prefix back to the empty string, so every single
filename afterwards will have the "..." prefix, whether appropriate or
not.

You can see this with "git diff v2.6.16.." on the current kernel tree,
since there are filenames with long names that changed there:

 [ snip snip ]
 Documentation/filesystems/vfs.txt                  |  229
 .../firmware_class/firmware_sample_driver.c        |    3
 .../firmware_sample_firmware_class.c               |    1
 ...Documentation/fujitsu/frv/kernel-ABI.txt           |  192
 ...Documentation/hwmon/w83627hf                       |    4
 [ snip snip ]

notice how the two Documentation/firmware** filenames caused the "..." to
be added, but then the later filenames don't want it, and it also screws
up the alignment of the line numbering afterwards.

Trivially fixed by moving the declaration (and initial setting) of the
"prefix" variable into the for-loop where it is used.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix repo-config set-multivar error return path.
Junio C Hamano [Mon, 8 May 2006 04:27:30 +0000 (21:27 -0700)]
Fix repo-config set-multivar error return path.

This hopefully fixes the problem an earlier commit 5d8ee9ceb attemted
to fix.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoclone: don't clone the info/alternates file
Martin Waitz [Sun, 7 May 2006 18:19:33 +0000 (20:19 +0200)]
clone: don't clone the info/alternates file

Now that the cloned alternates file is parsed, too we don't need to
copy it into our new repository, we just reference it.

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotest case for transitive info/alternates
Martin Waitz [Sun, 7 May 2006 18:19:47 +0000 (20:19 +0200)]
test case for transitive info/alternates

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTransitively read alternatives
Martin Waitz [Sun, 7 May 2006 18:19:21 +0000 (20:19 +0200)]
Transitively read alternatives

When adding an alternate object store then add entries from its
info/alternates files, too.
Relative entries are only allowed in the current repository.
Loops and duplicate alternates through multiple repositories are ignored.
Just to be sure that nothing breaks it is not allow to build deep
nesting levels using info/alternates.

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoChange to allow subdir updates from Eclipse
Martyn Smith [Tue, 28 Mar 2006 01:24:27 +0000 (13:24 +1200)]
Change to allow subdir updates from Eclipse

(Now you can rightclick any directory and select team-update/team-commit) and it should work

18 years agoMerge branch 'fix'
Junio C Hamano [Sun, 7 May 2006 22:36:39 +0000 (15:36 -0700)]
Merge branch 'fix'

* fix:
  repack: honor -d even when no new pack was created
  clone: keep --reference even with -l -s
  repo-config: document what value_regexp does a bit more clearly.
  Release config lock if the regex is invalid
  core-tutorial.txt: escape asterisk

18 years agoSparse fix for builtin-diff
Peter Hagervall [Sun, 7 May 2006 14:50:47 +0000 (16:50 +0200)]
Sparse fix for builtin-diff

You gotta love sparse:

builtin-diff.c:88:4: error: Just how const do you want this type to be?

Signed-off-by: Peter Hagervall <hager@cs.umu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorepack: honor -d even when no new pack was created
Martin Waitz [Sun, 7 May 2006 18:18:53 +0000 (20:18 +0200)]
repack: honor -d even when no new pack was created

If all objects are reachable via an alternate object store then we
still have to remove all obsolete local packs.

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoclone: keep --reference even with -l -s
Martin Waitz [Sun, 7 May 2006 18:19:09 +0000 (20:19 +0200)]
clone: keep --reference even with -l -s

Both -l -s and --reference update objects/info/alternates and used
to write over each other.

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMany fixes for most operations in Eclipse.
Martyn Smith [Mon, 27 Mar 2006 03:51:42 +0000 (15:51 +1200)]
Many fixes for most operations in Eclipse.

* Implemented global -n option
* Implemented "Questionable"
* Fixed Directory method, I _believe_ it's now correct in both cmdline and Eclipse.
* Directory method Now looks for localdir of "." and compares the repo dir, uses THIS as a basis for all directory level calculations.
* Added extra parameter to filenamesplit() to force stripping of "prepended" directory name. This ensures commits/updates etc work from any directory in the source tree.
* Modified argsfromdir() so it is "always" called. This means that when the client specifies a directory, the method can detect this and behave accordingly (this is currently only implemented for the '.' directory)
* Fixed "commit" method to correctly work from in a subdir

18 years agorepo-config: document what value_regexp does a bit more clearly.
Junio C Hamano [Sun, 7 May 2006 22:32:51 +0000 (15:32 -0700)]
repo-config: document what value_regexp does a bit more clearly.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRelease config lock if the regex is invalid
Pavel Roskin [Sun, 7 May 2006 21:36:12 +0000 (17:36 -0400)]
Release config lock if the regex is invalid

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdded logged warnings for CVS error returns
Martyn Smith [Fri, 17 Mar 2006 00:33:19 +0000 (13:33 +1300)]
Added logged warnings for CVS error returns

18 years agoMerge with git://kernel.org/pub/scm/git/git.git
Martin Langhoff [Sun, 7 May 2006 22:03:37 +0000 (10:03 +1200)]
Merge with git://kernel.org/pub/scm/git/git.git

18 years agocore-tutorial.txt: escape asterisk
Matthias Lederhofer [Sun, 7 May 2006 17:32:53 +0000 (19:32 +0200)]
core-tutorial.txt: escape asterisk

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix users of prefix_path() to free() only when necessary
Johannes Schindelin [Sat, 6 May 2006 22:02:53 +0000 (00:02 +0200)]
Fix users of prefix_path() to free() only when necessary

Unfortunately, prefix_path() sometimes returns a newly xmalloc()ed buffer,
and in other cases it returns a substring!

For example, when calling

git update-index ./hello.txt

prefix_path() returns "hello.txt", but does not allocate a new buffer. The
original code only checked if the result of prefix_path() was different from
what was passed in, and thusly trigger a segmentation fault.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix users of prefix_path() to free() only when necessary
Johannes Schindelin [Sat, 6 May 2006 22:02:53 +0000 (00:02 +0200)]
Fix users of prefix_path() to free() only when necessary

Unfortunately, prefix_path() sometimes returns a newly xmalloc()ed buffer,
and in other cases it returns a substring!

For example, when calling

git update-index ./hello.txt

prefix_path() returns "hello.txt", but does not allocate a new buffer. The
original code only checked if the result of prefix_path() was different from
what was passed in, and thusly trigger a segmentation fault.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot1300-repo-config: two new config parsing tests.
sean [Sat, 6 May 2006 19:43:43 +0000 (15:43 -0400)]
t1300-repo-config: two new config parsing tests.

- correctly insert a new variable into a section that only
  contains a single (different) variable.

- correctly insert a new section that matches the initial
  substring of an existing section.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAnother config file parsing fix.
sean [Sat, 6 May 2006 18:14:02 +0000 (14:14 -0400)]
Another config file parsing fix.

If the variable we need to store should go into a section
that currently only has a single variable (not matching
the one we're trying to insert), we will already be into
the next section before we notice we've bypassed the correct
location to insert the variable.

To handle this case we store the current location as soon
as we find a variable matching the section of our new
variable.

This breakage was brought up by Linus.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobinary diff and apply: testsuite.
Junio C Hamano [Sat, 6 May 2006 07:15:54 +0000 (00:15 -0700)]
binary diff and apply: testsuite.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoupdate-index --again: take optional pathspecs
Junio C Hamano [Sat, 6 May 2006 06:09:05 +0000 (23:09 -0700)]
update-index --again: take optional pathspecs

When pathspecs are given, update-index --again further limits
the set of paths to be updated to those that match them.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoupdate-index --again
Junio C Hamano [Sat, 6 May 2006 00:40:47 +0000 (17:40 -0700)]
update-index --again

After running 'git-update-index' for some paths, you may want to
do the update on the same set of paths again.

The new flag --again checks the paths whose index entries are
are different from the HEAD commit and updates them from the
working tree contents.

This was brought up by Carl Worth on #git.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoupdate-index: plug memory leak from prefix_path()
Junio C Hamano [Sat, 6 May 2006 05:53:56 +0000 (22:53 -0700)]
update-index: plug memory leak from prefix_path()

prefix_path() sometimes allocates new memory and returns it, and
other times returns the incoming path argument intact.  The
callers need to be a bit careful not to leak memory.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocheckout-index: plug memory leak from prefix_path()
Junio C Hamano [Sat, 6 May 2006 05:38:06 +0000 (22:38 -0700)]
checkout-index: plug memory leak from prefix_path()

prefix_path() sometimes allocates new memory and returns it, and
other times returns the incoming path argument intact.  The
callers need to be a bit careful not to leak memory.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoupdate-index --unresolve: work from a subdirectory.
Junio C Hamano [Sat, 6 May 2006 00:50:06 +0000 (17:50 -0700)]
update-index --unresolve: work from a subdirectory.

It completely forgot to take the prefix into account, so you
had to feed the full path even when you start from a
subdirectory, which was nonsensical.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobinary diff: further updates.
Junio C Hamano [Fri, 5 May 2006 09:41:53 +0000 (02:41 -0700)]
binary diff: further updates.

This updates the user interface and generated diff data format.

 * "diff --binary" is used to signal that we want an e-mailable
   binary patch.  It implies --full-index and -p.

 * "apply --allow-binary-replacement" acquired a short synonym
   "apply --binary".

 * After the "GIT binary patch\n" header line there is a token
   to record which binary patch mechanism was used, so that we
   can extend it later.  Currently there are two mechanisms
   defined: "literal" and "delta".  The former records the
   deflated postimage and the latter records the deflated delta
   from the preimage to postimage.

   For purely implementation convenience, I added the deflated
   length after these "literal/delta" tokens (otherwise the
   decoding side needs to guess and reallocate the buffer while
   inflating).  Improvement patches are very welcomed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobinary patch.
Junio C Hamano [Thu, 4 May 2006 23:51:44 +0000 (16:51 -0700)]
binary patch.

This adds "binary patch" to the diff output and teaches apply
what to do with them.

On the diff generation side, traditionally, we said "Binary
files differ\n" without giving anything other than the preimage
and postimage object name on the index line.  This was good
enough for applying a patch generated from your own repository
(very useful while rebasing), because the postimage would be
available in such a case.  However, this was not useful when the
recipient of such a patch via e-mail were to apply it, even if
the preimage was available.

This patch allows the diff to generate "binary" patch when
operating under --full-index option.  The binary patch follows
the usual extended git diff headers, and looks like this:

"GIT binary patch\n"
<length byte><data>"\n"
...
"\n"

Each line is prefixed with a "length-byte", whose value is upper
or lowercase alphabet that encodes number of bytes that the data
on the line decodes to (1..52 -- 'A' means 1, 'B' means 2, ...,
'Z' means 26, 'a' means 27, ...).  <data> is 1 or more groups of
5-byte sequence, each of which encodes up to 4 bytes in base85
encoding.  Because 52 / 4 * 5 = 65 and we have the length byte,
an output line is capped to 66 characters.  The payload is the
same diff-delta as we use in the packfiles.

On the consumption side, git-apply now can decode and apply the
binary patch when --allow-binary-replacement is given, the diff
was generated with --full-index, and the receiving repository
has the preimage blob, which is the same condition as it always
required when accepting an "Binary files differ\n" patch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agopack-object: squelch eye-candy on non-tty
Junio C Hamano [Fri, 5 May 2006 10:20:44 +0000 (03:20 -0700)]
pack-object: squelch eye-candy on non-tty

One of my post-update scripts runs a git-fetch into a separate
repository and sends the results back to me (2>&1); I end up
getting this in the mail:

    Generating pack...
    Done counting 180 objects.
    Result has 131 objects.
    Deltifying 131 objects.
       0% (0/131) done^M   1% (2/131) done^M...

This defaults not to do the progress report when not on a tty.

You could give --progress to force the progress report, but
let's not bother even documenting it nor mentioning it in the
usage string.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'fix'
Junio C Hamano [Fri, 5 May 2006 21:49:49 +0000 (14:49 -0700)]
Merge branch 'fix'

* fix:
  core.prefersymlinkrefs: use symlinks for .git/HEAD
  repo-config: trim white-space before comment
  Fix for config file section parsing.

18 years agocore.prefersymlinkrefs: use symlinks for .git/HEAD
Junio C Hamano [Tue, 2 May 2006 07:40:24 +0000 (00:40 -0700)]
core.prefersymlinkrefs: use symlinks for .git/HEAD

When inspecting a project whose build infrastructure used to
assume that .git/HEAD is a symlink ref, core.prefersymlinkrefs
in the config file of such a project would help to bisect its
history.

Signed-off-by: Junio C Hamano <junkio@cox.net>
(cherry picked from 9f0bb90d161edf8c43f5261d12bf83f14eb02ff4 commit)

18 years agorepo-config: trim white-space before comment
Johannes Schindelin [Tue, 2 May 2006 14:58:37 +0000 (16:58 +0200)]
repo-config: trim white-space before comment

Earlier, calling

git-repo-config core.hello

on a .git/config like this:

[core]
hello = world ; a comment

would yield "world " (i.e. with a trailing space).

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
(cherry picked from c1aee1fd8d94da9b3c5d2dc1d4264f7e73a58f80 commit)

18 years agoFix for config file section parsing.
sean [Fri, 5 May 2006 13:49:15 +0000 (09:49 -0400)]
Fix for config file section parsing.

Currently, if the target key has a section that matches
the initial substring of another section we mistakenly
believe we've found the correct section.  To avoid this
problem, ensure that the section lengths are identical
before comparison.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoClarify git-cherry documentation.
sean [Fri, 5 May 2006 19:06:07 +0000 (15:06 -0400)]
Clarify git-cherry documentation.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoUpdate git-unpack-objects documentation.
sean [Fri, 5 May 2006 19:05:36 +0000 (15:05 -0400)]
Update git-unpack-objects documentation.

Document that git-unpack-objects will not produce any
results when used on a pack that exists in a repository;
move it first.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix up docs where "--" isn't displayed correctly.
sean [Fri, 5 May 2006 19:05:24 +0000 (15:05 -0400)]
Fix up docs where "--" isn't displayed correctly.

A bare "--" doesn't show up in man or html pages correctly
as two individual dashes unless backslashed as \--
in the asciidoc source.  Note, no backslash is needed
inside a literal block.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSeveral trivial documentation touch ups.
sean [Fri, 5 May 2006 19:05:10 +0000 (15:05 -0400)]
Several trivial documentation touch ups.

  Move incorrect asciidoc level 2 titles back to level 1.

  Show output of git-name-rev in man page example.

  Reword sentences that begin with a period (.) in asciidoc
  numbered lists to work around conversion to man page bug.

  Mention that git-repack now calls git-prune-packed
  when the -d option is passed to it.

  [imap] section headers in the config file example need to be
  contained in a literal block.  imap.pass is the proper config
  file variable to use, not imap.password.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn 1.0.0
Eric Wong [Fri, 5 May 2006 19:35:40 +0000 (12:35 -0700)]
git-svn 1.0.0

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: documentation updates
Eric Wong [Fri, 5 May 2006 19:35:39 +0000 (12:35 -0700)]
git-svn: documentation updates

* Clarify that 'init' requires an argument
* Remove instances of 'SVN_URL' in the manpage, it's not an
  environment variable.
* Refer to 'Additional Fetch Arguments' when documenting 'fetch'
* document --authors-file / -A option

Thanks to Pavel Roskin and Seth Falcon for bringing these issues
to my attention.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agodelta: stricter constness
Junio C Hamano [Fri, 5 May 2006 09:35:58 +0000 (02:35 -0700)]
delta: stricter constness

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMakefile: do not link rev-list any specially.
Junio C Hamano [Fri, 5 May 2006 00:38:41 +0000 (17:38 -0700)]
Makefile: do not link rev-list any specially.

We used to depend on bignum from openssl for rev-list to compute
merge-order, but there is no reason to use different build
recipe from other programs anymore.  Just build it with git-%$X
rule like everybody else.

Noticed by Alexey Dobriyan.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobuiltin-push: --all and --tags _are_ explicit refspecs
Johannes Schindelin [Thu, 4 May 2006 21:18:14 +0000 (23:18 +0200)]
builtin-push: --all and --tags _are_ explicit refspecs

... so do not get refspecs from remotes/* or the config if one of them
was specified.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/logs'
Junio C Hamano [Thu, 4 May 2006 07:04:32 +0000 (00:04 -0700)]
Merge branch 'jc/logs'

* jc/logs:
  builtin-log/whatchanged/show: make them official.

18 years agoMerge branch 'jc/show-branch-dense'
Junio C Hamano [Thu, 4 May 2006 06:58:35 +0000 (23:58 -0700)]
Merge branch 'jc/show-branch-dense'

* jc/show-branch-dense:
  show-branch: omit uninteresting merges.

18 years agoMerge branch 'jc/symref'
Junio C Hamano [Thu, 4 May 2006 06:54:55 +0000 (23:54 -0700)]
Merge branch 'jc/symref'

* jc/symref:
  core.prefersymlinkrefs: use symlinks for .git/HEAD

18 years agoMerge branch 'jc/diff'
Junio C Hamano [Thu, 4 May 2006 06:54:34 +0000 (23:54 -0700)]
Merge branch 'jc/diff'

* jc/diff:
  builtin-diff: call it "git-diff", really.
  builtin-diff.c: die() formatting type fix.
  built-in diff: assorted updates.
  built-in diff.

18 years agoMerge branch 'js/repoconfig'
Junio C Hamano [Thu, 4 May 2006 06:41:28 +0000 (23:41 -0700)]
Merge branch 'js/repoconfig'

* js/repoconfig:
  repo-config: deconvolute logics
  repo-config: readability fixups.
  repo-config: support --get-regexp

18 years agoMerge branch 'jc/count'
Junio C Hamano [Thu, 4 May 2006 06:40:39 +0000 (23:40 -0700)]
Merge branch 'jc/count'

* jc/count:
  builtin-count-objects: open packs when running -v
  builtin-count-objects: make it official.
  built-in count-objects.

18 years agoMerge branch 'js/remoteconfig'
Junio C Hamano [Thu, 4 May 2006 06:28:09 +0000 (23:28 -0700)]
Merge branch 'js/remoteconfig'

* js/remoteconfig:
  Revert "fetch, pull: ask config for remote information"
  fetch, pull: ask config for remote information
  builtin-push: also ask config for remote information
  builtin-push: make it official.
  Fix builtin-push to honor Push: lines in remotes file.
  builtin-push: resurrect parsing of Push: lines
  git builtin "push"

18 years agoMerge branch 'maint'
Junio C Hamano [Thu, 4 May 2006 06:00:02 +0000 (23:00 -0700)]
Merge branch 'maint'

* maint:
  Add a few more words to the glossary.
  Added definitions for a few words:
  Alphabetize the glossary.

18 years agoAdd a few more words to the glossary. v1.3.2
Jon Loeliger [Thu, 4 May 2006 04:19:54 +0000 (23:19 -0500)]
Add a few more words to the glossary.

Clean up a few entries and fix typos.

    bare repository
    cherry-picking
    hook
    topic branch

[jc: removing questionable "symbolic ref -- see 'ref'" for now.]

Signed-off-by: Jon Loeliger <jdl@jdl.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdded definitions for a few words:
Jon Loeliger [Thu, 4 May 2006 04:18:59 +0000 (23:18 -0500)]
Added definitions for a few words:

    fast forward
    pickaxe
    refspec
    tracking branch

Wild hack allows "link:git-" prefix to reference commands too.

Signed-off-by: Jon Loeliger <jdl@jdl.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAlphabetize the glossary.
Jon Loeliger [Thu, 4 May 2006 04:15:46 +0000 (23:15 -0500)]
Alphabetize the glossary.

Signed-off-by: Jon Loeliger <jdl@jdl.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agosha1_to_hex() usage cleanup
Linus Torvalds [Thu, 4 May 2006 00:21:08 +0000 (17:21 -0700)]
sha1_to_hex() usage cleanup

Somebody on the #git channel complained that the sha1_to_hex() thing uses
a static buffer which caused an error message to show the same hex output
twice instead of showing two different ones.

That's pretty easily rectified by making it uses a simple LRU of a few
buffers, which also allows some other users (that were aware of the buffer
re-use) to be written in a more straightforward manner.

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 [Thu, 4 May 2006 00:15:06 +0000 (17:15 -0700)]
Merge branch 'fix'

* fix:
  fix various typos in documentation

18 years agoblame: Fix path pruning
Fredrik Kuivinen [Wed, 3 May 2006 21:28:46 +0000 (23:28 +0200)]
blame: Fix path pruning

This makes git-blame useable again, it has been totally broken for
some time on larger repositories.

Signed-off-by: Fredrik Kuivinen <freku045@student.liu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: use git-rev-list instead of git-log
Martin Langhoff [Wed, 3 May 2006 22:53:37 +0000 (10:53 +1200)]
cvsserver: use git-rev-list instead of git-log

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorepo-config: deconvolute logics
Johannes Schindelin [Wed, 3 May 2006 12:41:03 +0000 (14:41 +0200)]
repo-config: deconvolute logics

It was rightly noticed that the logic is quite convoluted. Fix that.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsserver: use git-rev-list instead of git-log
Martin Langhoff [Wed, 3 May 2006 22:51:46 +0000 (10:51 +1200)]
cvsserver: use git-rev-list instead of git-log

On 5/4/06, Linus Torvalds <torvalds@osdl.org> wrote:
> No it wasn't. "git log --parents" was definitely supposed to still work.
>
> That said, I suspect a git-cvsserver kind of usage is better off using
> "git-rev-list --parents HEAD" instead, which didn't break in the first
> place.

18 years agoFix "git-log --parents" breakage post v1.3.0
Linus Torvalds [Wed, 3 May 2006 14:59:00 +0000 (07:59 -0700)]
Fix "git-log --parents" breakage post v1.3.0

Post 1.3.0 "git log" forgets to list parent commits on the first line
when --parents is given.  git-cvsserver relied on it.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoadd documentation for update-index --unresolve
Matthias Kestenholz [Wed, 3 May 2006 10:53:19 +0000 (12:53 +0200)]
add documentation for update-index --unresolve

Signed-off-by: Matthias Kestenholz <matthias@spinlock.ch>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agofix various typos in documentation
Matthias Kestenholz [Wed, 3 May 2006 10:51:40 +0000 (12:51 +0200)]
fix various typos in documentation

Signed-off-by: Matthias Kestenholz <matthias@spinlock.ch>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge with git://kernel.org/pub/scm/git/git.git
Martin Langhoff [Wed, 3 May 2006 10:27:33 +0000 (22:27 +1200)]
Merge with git://kernel.org/pub/scm/git/git.git

18 years agobuiltin-count-objects: open packs when running -v
Junio C Hamano [Wed, 3 May 2006 06:03:15 +0000 (23:03 -0700)]
builtin-count-objects: open packs when running -v

Otherwise we would report absolutely no objects in a fully
packed repository.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoimprove diff-delta with sparse and/or repetitive data
Nicolas Pitre [Wed, 3 May 2006 03:31:00 +0000 (23:31 -0400)]
improve diff-delta with sparse and/or repetitive data

It is useless to preserve multiple hash entries for consecutive blocks
with the same hash.  Keeping only the first one will allow for matching
the longest string of identical bytes while subsequent blocks will only
allow for shorter matches.  The backward matching code will match the
end of it as necessary.

This improves both performances (no repeated string compare with long
successions of identical bytes, or even small group of bytes), as well
as compression (less likely to need random hash bucket entry culling),
especially with sparse files.

With well behaved data sets this patch doesn't change much.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotiny optimization to diff-delta
Nicolas Pitre [Wed, 3 May 2006 03:46:51 +0000 (23:46 -0400)]
tiny optimization to diff-delta

This is my assembly freak side looking at generated code again.  And
since create_delta() is certainly pretty high on the radar every bits
count.  In this case shorter code is generated if hash_mask is not
copied to a local variable.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorepo-config: readability fixups.
Junio C Hamano [Wed, 3 May 2006 04:06:56 +0000 (21:06 -0700)]
repo-config: readability fixups.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocore.prefersymlinkrefs: use symlinks for .git/HEAD
Junio C Hamano [Tue, 2 May 2006 07:40:24 +0000 (00:40 -0700)]
core.prefersymlinkrefs: use symlinks for .git/HEAD

When inspecting a project whose build infrastructure used to
assume that .git/HEAD is a symlink ref, core.prefersymlinkrefs
in the config file of such a project would help to bisect its
history.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agorepo-config: support --get-regexp
Johannes Schindelin [Tue, 2 May 2006 12:22:48 +0000 (14:22 +0200)]
repo-config: support --get-regexp

With --get-regexp, output all key/value pairs where the key matches a
regexp. Example:

git-repo-config --get-regexp remote.*.url

will output something like

remote.junio.url git://git.kernel.org/pub/scm/git/git.git
remote.gitk.url git://git.kernel.org/pub/scm/gitk/gitk.git

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>