Code

git.git
17 years agocvsserver: Actually implement --export-all
Frank Lichtenheld [Fri, 15 Jun 2007 01:01:53 +0000 (03:01 +0200)]
cvsserver: Actually implement --export-all

Embarrassing bug number two in my options patch.

Also enforce that --export-all is only ever used together with an
explicit whitelist. Otherwise people might export every git repository
on the whole system without realising.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agocvsserver: Let --base-path and pserver get along just fine
Frank Lichtenheld [Fri, 15 Jun 2007 01:01:52 +0000 (03:01 +0200)]
cvsserver: Let --base-path and pserver get along just fine

Embarassing bug number one in my options patch.

Since the code for --base-path support rewrote
the cvsroot value after comparing it with a possible
existing value (i.e. from pserver authentication)
the check always failed.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agocvsserver: Add some useful commandline options
Frank Lichtenheld [Thu, 7 Jun 2007 14:57:01 +0000 (16:57 +0200)]
cvsserver: Add some useful commandline options

Make git-cvsserver understand some options inspired by
git-daemon, namely --base-path, --export-all, --strict-paths.

Also allow the caller to specify a whitelist of allowed
directories, again similar to git-daemon.

While already adding option parsing also support the common
--help and --version options.

Rationale:

While the gitcvs.enabled configuration option already
offers means to limit git-cvsserver access to a repository,
there are some use cases where other methods of access
control prove to be more useful.

E.g. if setting up a pserver for a collection of public
repositories one might want limit the exported repositories
to exactly the directory this collection is located whithout
having to worry about other repositories that might lie around
with the configuration variable set (never trust your users ;)

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMerge branch 'maint'
Junio C Hamano [Fri, 8 Jun 2007 09:55:19 +0000 (02:55 -0700)]
Merge branch 'maint'

* maint:
  checkout: do not get confused with ambiguous tag/branch names

17 years agoEven more missing static
Junio C Hamano [Fri, 8 Jun 2007 09:54:57 +0000 (02:54 -0700)]
Even more missing static

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMore missing static
Junio C Hamano [Fri, 8 Jun 2007 09:22:56 +0000 (02:22 -0700)]
More missing static

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMore missing static
Junio C Hamano [Fri, 8 Jun 2007 09:24:58 +0000 (02:24 -0700)]
More missing static

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMissing statics.
Pierre Habouzit [Thu, 7 Jun 2007 20:45:00 +0000 (22:45 +0200)]
Missing statics.

Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoActive_nr is unsigned, hence can't be < 0
Pierre Habouzit [Thu, 7 Jun 2007 20:44:59 +0000 (22:44 +0200)]
Active_nr is unsigned, hence can't be < 0

Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agocvsserver: Make req_Root more critical of its input data
Frank Lichtenheld [Thu, 7 Jun 2007 14:57:00 +0000 (16:57 +0200)]
cvsserver: Make req_Root more critical of its input data

The path submitted with the Root request has to be absolute
(cvs does it this way and it may save us some sanity checks
later)

If multiple roots are specified (e.g. because we use
pserver authentication which will already include the
root), ensure that they say all the same.

Probably neither is a security risk, and neither should ever
be triggered by a sane client, but when validating
input data, it's better to be save than sorry.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agogitview: Define __slots__ for Commit
Michael Ellerman [Wed, 30 May 2007 04:47:09 +0000 (14:47 +1000)]
gitview: Define __slots__ for Commit

Define __slots__ for the Commit class. This reserves space in each Commit
object for only the defined variables. On my system this reduces heap usage
when viewing a kernel repo by 12% ~= 55868 KB.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agogitview: Use new-style classes
Michael Ellerman [Wed, 30 May 2007 04:47:08 +0000 (14:47 +1000)]
gitview: Use new-style classes

This changes the Commit class to use new-style class, which has
been available since Python 2.2 (Dec 2001).  This is a necessary
step in order to use __slots__[] declaration, so that we can
reduce the memory footprint in the next patch.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agocheckout: do not get confused with ambiguous tag/branch names
Junio C Hamano [Fri, 8 Jun 2007 08:19:13 +0000 (01:19 -0700)]
checkout: do not get confused with ambiguous tag/branch names

Although it is not advisable, we have always allowed a branch
and a tag to have the same basename (i.e. it is not illegal to
have refs/heads/frotz and refs/tags/frotz at the same time).
When talking about a specific commit, the interpretation of
'frotz' has always been "use tag and then check branch",
although we warn when ambiguities exist.

However "git checkout $name" is defined to (1) first see if it
matches the branch name, and if so switch to that branch; (2)
otherwise it is an instruction to detach HEAD to point at the
commit named by $name.  We did not follow this definition when
$name appeared under both refs/heads/ and refs/tags/ -- we
switched to the branch but read the tree from the tagged commit,
which was utterly bogus.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoWar on whitespace
Junio C Hamano [Thu, 7 Jun 2007 07:04:01 +0000 (00:04 -0700)]
War on whitespace

This uses "git-apply --whitespace=strip" to fix whitespace errors that have
crept in to our source files over time.  There are a few files that need
to have trailing whitespaces (most notably, test vectors).  The results
still passes the test, and build result in Documentation/ area is unchanged.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMerge branch 'sv/objfixes'
Junio C Hamano [Wed, 6 Jun 2007 22:43:24 +0000 (15:43 -0700)]
Merge branch 'sv/objfixes'

* sv/objfixes:
  Don't assume tree entries that are not dirs are blobs
  git-cvsimport: Make sure to use $git_dir always instead of .git sometimes
  fix documentation of unpack-objects -n
  Accept dates before 2000/01/01 when specified as seconds since the epoch

17 years agoDon't assume tree entries that are not dirs are blobs
Sam Vilain [Wed, 6 Jun 2007 10:25:17 +0000 (22:25 +1200)]
Don't assume tree entries that are not dirs are blobs

When scanning the trees in track_tree_refs() there is a "lazy" test
that assumes that entries are either directories or files.  Don't do
that.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agogit-cvsimport: Make sure to use $git_dir always instead of .git sometimes
Michael Milligan [Tue, 5 Jun 2007 06:06:30 +0000 (00:06 -0600)]
git-cvsimport: Make sure to use $git_dir always instead of .git sometimes

CVS import was failing on a couple repos I was trying to import.
I was setting GIT_DIR=newproj.git and using the -i flag, but this bug
was thwarting the effort...  evil CVS.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agofix documentation of unpack-objects -n
Sam Vilain [Wed, 6 Jun 2007 21:23:16 +0000 (09:23 +1200)]
fix documentation of unpack-objects -n

unpack-objects -n didn't print the object list as promised on the
manual page, so alter the documentation to reflect the behaviour

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoAccept dates before 2000/01/01 when specified as seconds since the epoch
Johannes Sixt [Wed, 6 Jun 2007 08:11:55 +0000 (10:11 +0200)]
Accept dates before 2000/01/01 when specified as seconds since the epoch

Tests with git-filter-branch on a repository that was converted from
CVS and that has commits reaching back to 1999 revealed that it is
necessary to parse dates before 2000/01/01 when they are specified
as seconds since 1970/01/01. There is now still a limit, 100000000,
which is 1973/03/03 09:46:40 UTC, in order to allow that dates are
represented as 8 digits.

Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMakefile: Remove git-merge-base from PROGRAMS.
Johannes Sixt [Mon, 4 Jun 2007 11:53:05 +0000 (13:53 +0200)]
Makefile: Remove git-merge-base from PROGRAMS.

git-merge-base is a builtin.

Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agot5000: skip ZIP tests if unzip was not found
Johannes Schindelin [Wed, 6 Jun 2007 18:57:40 +0000 (19:57 +0100)]
t5000: skip ZIP tests if unzip was not found

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMerge branch 'mm/tag'
Junio C Hamano [Wed, 6 Jun 2007 09:29:41 +0000 (02:29 -0700)]
Merge branch 'mm/tag'

* mm/tag:
  Teach git-tag about showing tag annotations.

17 years agogit-branch --track: fix tracking branch computation.
Junio C Hamano [Wed, 6 Jun 2007 08:10:14 +0000 (01:10 -0700)]
git-branch --track: fix tracking branch computation.

The original code did not take hierarchical branch names into account at all.

Tested-by: Gerrit Pape <pape@smarden.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoFix typo in git-mergetool
Josh Triplett [Wed, 6 Jun 2007 04:24:19 +0000 (21:24 -0700)]
Fix typo in git-mergetool

Signed-off-by: Josh Triplett <josh@freedesktop.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoAdd the --numbered-files option to git-format-patch.
Jon Loeliger [Tue, 5 Jun 2007 20:06:53 +0000 (15:06 -0500)]
Add the --numbered-files option to git-format-patch.

With this option, git-format-patch will generate simple
numbered files as output instead of the default using
with the first commit line appended.

This simplifies the ability to generate an MH-style
drafts folder with each message to be sent.

Signed-off-by: Jon Loeliger <jdl@freescale.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years ago$EMAIL is a last resort fallback, as it's system-wide.
Pierre Habouzit [Tue, 5 Jun 2007 16:40:41 +0000 (18:40 +0200)]
$EMAIL is a last resort fallback, as it's system-wide.

$EMAIL is a system-wide setup that is used for many many many
applications. If the git user chose a specific user.email setup,
then _this_ should be honoured rather than $EMAIL.

Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agomake clean should remove all the test programs too
Matthias Lederhofer [Tue, 5 Jun 2007 13:43:17 +0000 (15:43 +0200)]
make clean should remove all the test programs too

Signed-off-by: Matthias Lederhofer <matled@gmx.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoadd git-filter-branch to .gitignore
Matthias Lederhofer [Tue, 5 Jun 2007 13:26:12 +0000 (15:26 +0200)]
add git-filter-branch to .gitignore

Signed-off-by: Matthias Lederhofer <matled@gmx.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMerge branch 'gb/idx'
Junio C Hamano [Wed, 6 Jun 2007 04:36:51 +0000 (21:36 -0700)]
Merge branch 'gb/idx'

* gb/idx:
  Unify write_index_file functions

17 years agomerge-recursive: refuse to merge binary files
Johannes Schindelin [Tue, 5 Jun 2007 02:36:49 +0000 (03:36 +0100)]
merge-recursive: refuse to merge binary files

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoMove buffer_is_binary() to xdiff-interface.h
Johannes Schindelin [Tue, 5 Jun 2007 02:36:11 +0000 (03:36 +0100)]
Move buffer_is_binary() to xdiff-interface.h

We already have two instances where we want to determine if a buffer
contains binary data as opposed to text.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agogit-fsck: learn about --verbose
Johannes Schindelin [Tue, 5 Jun 2007 02:44:00 +0000 (03:44 +0100)]
git-fsck: learn about --verbose

With --verbose, it gets really chatty now.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agogitweb: Handle non UTF-8 text better
Martin Koegler [Sun, 3 Jun 2007 15:42:44 +0000 (17:42 +0200)]
gitweb: Handle non UTF-8 text better

gitweb assumes that everything is in UTF-8. If a text contains invalid
UTF-8 character sequences, the text must be in a different encoding.

This commit introduces $fallback_encoding which would be used as input
encoding if gitweb encounters text with is not valid UTF-8.

Add basic test for this in t/t9500-gitweb-standalone-no-errors.sh

Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Tested-by: Alexandre Julliard <julliard@winehq.org>
Tested-by: Ismail Dönmez <ismail@pardus.org.tr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoAdd test-sha1 to .gitignore.
Randal L. Schwartz [Sun, 3 Jun 2007 15:27:52 +0000 (08:27 -0700)]
Add test-sha1 to .gitignore.

Signed-off-by: Junio C Hamano <gitster@pobox.com>
17 years agoTeach git-tag about showing tag annotations.
Matthijs Melchior [Sun, 3 Jun 2007 00:05:39 +0000 (02:05 +0200)]
Teach git-tag about showing tag annotations.

The <pattern> for -l is now a shell pattern, not a list of grep parameters.
Option -l may be repeated with another <pattern>.

The new -n [<num>] option specifies how many lines from
the annotation are to be printed.
Not specifieing -n or -n 0 will just produce the tag names
Just -n or -n 1 will show the first line of the annotation on
the tag line.
Other valuse for -n will show that number of lines from the annotation.

The exit code used to indicate if any tag was found.
This is changed due to a different implementation.

A good way to test a tag for existence is to use:
git show-ref --quiet --verify refs/tags/$TAGNAME

Signed-off-by: Matthijs Melchior <mmelchior@xs4all.nl>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-apply: what is detected and fixed is not just trailing spaces.
Junio C Hamano [Sun, 3 Jun 2007 02:55:54 +0000 (19:55 -0700)]
git-apply: what is detected and fixed is not just trailing spaces.

But we kept saying "trailing whitespace" all the same.  Reword the
error messages a bit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUpdate to SubmittingPatches
Johannes Schindelin [Sun, 3 Jun 2007 00:46:47 +0000 (01:46 +0100)]
Update to SubmittingPatches

Make people aware of our testsuite, and of non-ASCII encodings.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint' to sync with 1.5.2.1
Junio C Hamano [Sun, 3 Jun 2007 02:36:20 +0000 (19:36 -0700)]
Merge branch 'maint' to sync with 1.5.2.1

17 years agoRelease Notes: start preparing for 1.5.3
Junio C Hamano [Sun, 3 Jun 2007 02:32:48 +0000 (19:32 -0700)]
Release Notes: start preparing for 1.5.3

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branches 'lh/submodules' and 'pb/am'
Junio C Hamano [Sun, 3 Jun 2007 02:04:54 +0000 (19:04 -0700)]
Merge branches 'lh/submodules' and 'pb/am'

* lh/submodules:
  Add basic test-script for git-submodule
  Add git-submodule command

* pb/am:
  Remove git-applypatch
  git-applymbox: Remove command

17 years agoGIT 1.5.2.1 v1.5.2.1
Junio C Hamano [Sat, 2 Jun 2007 22:01:33 +0000 (15:01 -0700)]
GIT 1.5.2.1

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUnify write_index_file functions
Geert Bosch [Fri, 1 Jun 2007 19:18:05 +0000 (15:18 -0400)]
Unify write_index_file functions

This patch unifies the write_index_file functions in
builtin-pack-objects.c and index-pack.c.  As the name
"index" is overloaded in git, move in the direction of
using "idx" and "pack idx" when refering to the pack index.
There should be no change in functionality.

Signed-off-by: Geert Bosch <bosch@gnat.com>
Acked-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAdd basic test-script for git-submodule
Lars Hjemli [Sat, 2 Jun 2007 01:27:42 +0000 (03:27 +0200)]
Add basic test-script for git-submodule

This test tries to verify basic sanity of git-submodule, i.e. that it is
able to clone and update a submodule repository, that its status output is
sane, and that it barfs when the submodule path is occupied during init.

Signed-off-by: Lars Hjemli <hjemli@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'np/pack'
Junio C Hamano [Sat, 2 Jun 2007 19:18:56 +0000 (12:18 -0700)]
Merge branch 'np/pack'

* np/pack:
  fix repack with --max-pack-size
  builtin-pack-object: cache small deltas
  git-pack-objects: cache small deltas between big objects
  builtin-pack-objects: don't fail, if delta is not possible

17 years agoMerge branch 'sp/pack'
Junio C Hamano [Sat, 2 Jun 2007 19:18:51 +0000 (12:18 -0700)]
Merge branch 'sp/pack'

* sp/pack:
  Style nit - don't put space after function names
  Ensure the pack index is opened before access
  Simplify index access condition in count-objects, pack-redundant
  Test for recent rev-parse $abbrev_sha1 regression
  rev-parse: Identify short sha1 sums correctly.
  Attempt to delay prepare_alt_odb during get_sha1
  Micro-optimize prepare_alt_odb
  Lazily open pack index files on demand

17 years agogit-rebase: suggest to use git-add instead of git-update-index
Jonas Fonseca [Sat, 2 Jun 2007 17:59:49 +0000 (19:59 +0200)]
git-rebase: suggest to use git-add instead of git-update-index

The command is part of the main porcelain making git-add more
appropriate.

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint'
Junio C Hamano [Sat, 2 Jun 2007 19:05:08 +0000 (12:05 -0700)]
Merge branch 'maint'

* maint:
  Use =20 when rfc2047 encoding spaces.
  Create a new manpage for the gitignore format, and reference it elsewhere
  Documentation: robustify asciidoc GIT_VERSION replacement

17 years agoUse =20 when rfc2047 encoding spaces.
Kristian Høgsberg [Fri, 1 Jun 2007 21:08:12 +0000 (17:08 -0400)]
Use =20 when rfc2047 encoding spaces.

Encode ' ' using '=20' even though rfc2047 allows using '_' for
readability.  Unfortunately, many programs do not understand this and
just leave the underscore in place.  Using '=20' seems to work better.

[jc: with adjustment to t3901]

Signed-off-by: Kristian Høgsberg <hoegsberg@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoCreate a new manpage for the gitignore format, and reference it elsewhere
Josh Triplett [Sat, 2 Jun 2007 17:08:54 +0000 (10:08 -0700)]
Create a new manpage for the gitignore format, and reference it elsewhere

Only git-ls-files(1) describes the gitignore format in detail, and it does so
with reference to git-ls-files options.  Most users don't use the plumbing
command git-ls-files directly, and shouldn't have to look in its manpage for
information on the gitignore format.

Create a new manpage gitignore(5) (Documentation/gitignore.txt), and factor
out the gitignore documentation into that file, changing it to refer to
.gitignore and $GIT_DIR/info/exclude as used by porcelain commands.  Reference
gitignore(5) from other relevant manpages and documentation.  Remove
now-redundant information on exclude patterns from git-ls-files(1), leaving
only information on how git-ls-files options specify exclude patterns and what
precedence they have.

Signed-off-by: Josh Triplett <josh@freedesktop.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: robustify asciidoc GIT_VERSION replacement
Jeff King [Fri, 1 Jun 2007 09:10:30 +0000 (05:10 -0400)]
Documentation: robustify asciidoc GIT_VERSION replacement

Instead of using sed on the resulting file, we now have a
git_version asciidoc attribute. This means that we don't
pipe the output of asciidoc, which means we can detect build
failures.

Problem reported by Scott Lamb, solution suggested by Jonas Fonseca.

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix git-am(1) synopsis formatting
Jonas Fonseca [Sat, 2 Jun 2007 17:56:44 +0000 (19:56 +0200)]
Fix git-am(1) synopsis formatting

Signed-off-by: Jonas Fonseca <fonseca@diku.dk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix minor grammatical typos in the git-gc man page
Theodore Ts'o [Thu, 31 May 2007 23:00:48 +0000 (19:00 -0400)]
Fix minor grammatical typos in the git-gc man page

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agodiff-delta: use realloc instead of xrealloc
Martin Koegler [Tue, 29 May 2007 19:08:35 +0000 (21:08 +0200)]
diff-delta: use realloc instead of xrealloc

Commit 83572c1a914d3f7a8dd66d954c11bbc665b7b923 changed many
realloc to xrealloc. This change was made in diff-delta.c too,
although the code can handle an out of memory failure.

This patch reverts this change in diff-delta.c.

Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint'
Junio C Hamano [Thu, 31 May 2007 07:15:14 +0000 (00:15 -0700)]
Merge branch 'maint'

* maint:
  git-config: Improve documentation of git-config file handling
  git-config: Various small fixes to asciidoc documentation
  decode_85(): fix missing return.
  fix signed range problems with hex conversions

17 years agoMerge branch 'maint-1.5.1' into maint
Junio C Hamano [Thu, 31 May 2007 07:09:26 +0000 (00:09 -0700)]
Merge branch 'maint-1.5.1' into maint

* maint-1.5.1:
  git-config: Improve documentation of git-config file handling
  git-config: Various small fixes to asciidoc documentation
  decode_85(): fix missing return.
  fix signed range problems with hex conversions

17 years agogit-config: Improve documentation of git-config file handling
Frank Lichtenheld [Thu, 31 May 2007 00:35:37 +0000 (02:35 +0200)]
git-config: Improve documentation of git-config file handling

The description which files git-config uses and how the various
command line options and environment variables affect its
behaviour was incomplete, outdated and confusing.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-config: Various small fixes to asciidoc documentation
Frank Lichtenheld [Thu, 31 May 2007 00:35:36 +0000 (02:35 +0200)]
git-config: Various small fixes to asciidoc documentation

Add '' around the only mentioned commandline option that didn't
have it.

Make reference to section EXAMPLE a link and rename it to
EXAMPLES because it actually contains a lot of examples.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoalways start looking up objects in the last used pack first
Nicolas Pitre [Thu, 31 May 2007 02:48:13 +0000 (22:48 -0400)]
always start looking up objects in the last used pack first

Jon Smirl said:

| Once an object reference hits a pack file it is very likely that
| following references will hit the same pack file. So first place to
| look for an object is the same place the previous object was found.

This is indeed a good heuristic so here it is.  The search always start
with the pack where the last object lookup succeeded.  If the wanted
object is not available there then the search continues with the normal
pack ordering.

To test this I split the Linux repository into 66 packs and performed a
"time git-rev-list --objects --all > /dev/null".  Best results are as
follows:

Pack Sort w/o this patch w/ this patch
-------------------------------------------------------------
recent objects last 26.4s 20.9s
recent objects first 24.9s 18.4s

This shows that the pack order based on object age has some influence,
but that the last-used-pack heuristic is even more significant in
reducing object lookup.

Signed-off-by: Nicolas Pitre <nico@cam.org> --- Note: the
--max-pack-size to git-repack currently produces packs with old objects
after those containing recent objects.  The pack sort based on
filesystem timestamp is therefore backward for those.  This needs to be
fixed of course, but at least it made me think about this variable for
the test.
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agofix repack with --max-pack-size
Nicolas Pitre [Thu, 31 May 2007 01:43:12 +0000 (21:43 -0400)]
fix repack with --max-pack-size

Two issues here:

1) git-repack -a --max-pack-size=10 on the GIT repo dies pretty quick.
   There is a lot of confusion about deltas that were suposed to be
   reused from another pack but that get stored undeltified due to pack
   limit and object size doesn't match entry->size anymore.  This test
   is not really worth the complexity for determining when it is valid
   so get rid of it.

2) If pack limit is reached, the object buffer is freed, including when
   it comes from a cached delta data.  In practice the object will be
   stored in a subsequent pack undeltified, but let's make sure no
   pointer to freed data subsists by clearing entry->delta_data.

I also reorganized that code a bit to make it more readable.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMakefile: Use generic rule to build test programs
Julian Phillips [Wed, 30 May 2007 23:18:24 +0000 (00:18 +0100)]
Makefile: Use generic rule to build test programs

Use a generic make rule to build all the test programs, rather than
specifically mentioning each one.

Signed-off-by: Julian Phillips <julian@quantumfyre.co.uk>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agodecode_85(): fix missing return.
Jerald Fitzjerald [Wed, 30 May 2007 12:39:58 +0000 (05:39 -0700)]
decode_85(): fix missing return.

When the function detected an invalid base85 sequence, it issued
an error message but forgot to return error status at that point
and kept going.

Signed-off-by: Jerald Fitzjerald <jfj@freemail.gr>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agofix signed range problems with hex conversions
Linus Torvalds [Wed, 30 May 2007 17:32:19 +0000 (10:32 -0700)]
fix signed range problems with hex conversions

Make hexval_table[] "const".  Also make sure that the accessor
function hexval() does not access the table with out-of-range
values by declaring its parameter "unsigned char", instead of
"unsigned int".

With this, gcc can just generate:

movzbl  (%rdi), %eax
movsbl  hexval_table(%rax),%edx
movzbl  1(%rdi), %eax
movsbl  hexval_table(%rax),%eax
sall    $4, %edx
orl     %eax, %edx

for the code to generate a byte from two hex characters.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAdd DLH to .mailmap
Junio C Hamano [Wed, 30 May 2007 00:55:47 +0000 (17:55 -0700)]
Add DLH to .mailmap

... and make the entries sorted.

17 years agoStyle nit - don't put space after function names
Shawn O. Pearce [Wed, 30 May 2007 06:13:42 +0000 (02:13 -0400)]
Style nit - don't put space after function names

Our style is to not put a space after a function name.  I did here,
and Junio applied the patch with the incorrect formatting.  So I'm
cleaning up after myself since I noticed it upon review.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoEnsure the pack index is opened before access
Shawn O. Pearce [Wed, 30 May 2007 06:13:14 +0000 (02:13 -0400)]
Ensure the pack index is opened before access

In this particular location of fsck the index should have already
been opened by verify_pack, which is called just before we get
here and loop through the object names.  However, just in case a
future version of that function does not use the index file we'll
double-check its open before we access the num_objects field.

Better safe now than sorry later.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoSimplify index access condition in count-objects, pack-redundant
Shawn O. Pearce [Wed, 30 May 2007 06:12:28 +0000 (02:12 -0400)]
Simplify index access condition in count-objects, pack-redundant

My earlier lazy index opening patch changed this condition to check
index_data and call open_pack_index if it was NULL. In truth we only
care about num_objects.  Since open_pack_index does no harm if the
index is already open, and all indexes are likely to be closed in
this application, the "performance optimization" of inlining the
index_data check here was wrong.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTest for recent rev-parse $abbrev_sha1 regression
Shawn O. Pearce [Wed, 30 May 2007 04:50:26 +0000 (00:50 -0400)]
Test for recent rev-parse $abbrev_sha1 regression

My recent patch "Lazily open pack index files on demand" caused a
regression in the case of parsing abbreviated SHA-1 object names.
Git was unable to translate the abbreviated name into the full name
if the object was packed, as the pack .idx files were not opened
before being accessed.

This is a simple test to repack a repository then test for an
abbreviated SHA-1 within the packfile.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorev-parse: Identify short sha1 sums correctly.
James Bowes [Tue, 29 May 2007 23:29:51 +0000 (19:29 -0400)]
rev-parse: Identify short sha1 sums correctly.

find_short_packed_object was not loading the pack index files.
Teach it to do so.

Signed-off-by: James Bowes <jbowes@dangerouslyinc.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agobuiltin-pack-object: cache small deltas
Martin Koegler [Mon, 28 May 2007 21:20:59 +0000 (23:20 +0200)]
builtin-pack-object: cache small deltas

Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-pack-objects: cache small deltas between big objects
Martin Koegler [Mon, 28 May 2007 21:20:58 +0000 (23:20 +0200)]
git-pack-objects: cache small deltas between big objects

Creating deltas between big blobs is a CPU and memory intensive task.
In the writing phase, all (not reused) deltas are redone.

This patch adds support for caching deltas from the deltifing phase, so
that that the writing phase is faster.

The caching is limited to small deltas to avoid increasing memory usage very much.
The implemented limit is (memory needed to create the delta)/1024.

Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agobuiltin-pack-objects: don't fail, if delta is not possible
Martin Koegler [Mon, 28 May 2007 21:20:57 +0000 (23:20 +0200)]
builtin-pack-objects: don't fail, if delta is not possible

If builtin-pack-objects runs out of memory while finding
the best deltas, it bails out with an error.

If the delta index creation fails (because there is not enough memory),
we can downgrade the error message to a warning and continue with the
next object.

Signed-off-by: Martin Koegler <mkoegler@auto.tuwien.ac.at>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'db/remote'
Junio C Hamano [Tue, 29 May 2007 08:24:20 +0000 (01:24 -0700)]
Merge branch 'db/remote'

* db/remote:
  Move refspec pattern matching to match_refs().
  Update local tracking refs when pushing
  Add handlers for fetch-side configuration of remotes.
  Move refspec parser from connect.c and cache.h to remote.{c,h}
  Move remote parsing into a library file out of builtin-push.

17 years agoMerge branch 'dh/repack' (early part)
Junio C Hamano [Tue, 29 May 2007 08:16:28 +0000 (01:16 -0700)]
Merge branch 'dh/repack' (early part)

* 'dh/repack' (early part):
  Ensure git-repack -a -d --max-pack-size=N deletes correct packs
  pack-objects: clarification & option checks for --max-pack-size
  git-repack --max-pack-size: add option parsing to enable feature
  git-repack --max-pack-size: split packs as asked by write_{object,one}()
  git-repack --max-pack-size: write_{object,one}() respect pack limit
  git-repack --max-pack-size: new file statics and code restructuring
  Alter sha1close() 3rd argument to request flush only

17 years agoMerge branch 'np/delta'
Junio C Hamano [Tue, 29 May 2007 07:49:23 +0000 (00:49 -0700)]
Merge branch 'np/delta'

* np/delta:
  update diff-delta.c copyright
  improve delta long block matching with big files

17 years agoMerge branch 'jc/nodelta'
Junio C Hamano [Tue, 29 May 2007 07:41:50 +0000 (00:41 -0700)]
Merge branch 'jc/nodelta'

* jc/nodelta:
  builtin-pack-objects: remove unnecessary code for no-delta
  Teach "delta" attribute to pack-objects.
  pack-objects: pass fullname down to add_object_entry()

17 years agoMerge branch 'ar/verbose'
Junio C Hamano [Tue, 29 May 2007 07:41:36 +0000 (00:41 -0700)]
Merge branch 'ar/verbose'

* ar/verbose:
  Add another verbosity level to git-fetch
  Verbose connect messages to show the IP addresses used

17 years agoMerge branch 'ar/run'
Junio C Hamano [Tue, 29 May 2007 07:41:22 +0000 (00:41 -0700)]
Merge branch 'ar/run'

* ar/run:
  Allow environment variables to be unset in the processes started by run_command
  Add ability to specify environment extension to run_command
  Add run_command_v_opt_cd: chdir into a directory before exec

17 years agoMerge branch 'ar/mergestat'
Junio C Hamano [Tue, 29 May 2007 07:38:52 +0000 (00:38 -0700)]
Merge branch 'ar/mergestat'

* ar/mergestat:
  Add a configuration option to control diffstat after merge

17 years agoMerge branch 'rr/cvsexport'
Junio C Hamano [Tue, 29 May 2007 07:37:23 +0000 (00:37 -0700)]
Merge branch 'rr/cvsexport'

* rr/cvsexport:
  Add option to cvs update before export

17 years agoMerge branch 'maint'
Junio C Hamano [Tue, 29 May 2007 07:27:24 +0000 (00:27 -0700)]
Merge branch 'maint'

* maint:
  cvsserver: Fix some typos in asciidoc documentation
  cvsserver: Note that CVS_SERVER can also be specified as method variable
  cvsserver: Correct inetd.conf example in asciidoc documentation
  user-manual: fixed typo in example
  Add test case for $Id$ expanded in the repository
  git-svn: avoid md5 calculation entirely if SVN doesn't provide one
  Makefile: Remove git-fsck and git-verify-pack from PROGRAMS
  Fix stupid typo in lookup_tag()
  git-gui: Guess our share/git-gui/lib path at runtime if possible
  Correct key bindings to Control-<foo>
  git-gui: Tighten internal pattern match for lib/ directory

17 years agocvsserver: Handle 'cvs login'
Frank Lichtenheld [Sun, 27 May 2007 12:33:10 +0000 (14:33 +0200)]
cvsserver: Handle 'cvs login'

Since this is a trivial variation of the general pserver
authentication, there is really no reason not to support
it.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot9400: Work around CVS' deficiencies
Frank Lichtenheld [Sun, 27 May 2007 12:33:09 +0000 (14:33 +0200)]
t9400: Work around CVS' deficiencies

If we are too fast with our changes, the file in
the working copy might still have the same mtime
as noted in the CVS/Entries. This will cause CVS
to happily report to the server that the file is
unmodified which can lead to data loss (and in
our case test failure).

CVS sucks!

Work around that by sleeping for a second.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAllow contrib new-workdir to link into bare repositories
Shawn O. Pearce [Sun, 27 May 2007 03:09:27 +0000 (23:09 -0400)]
Allow contrib new-workdir to link into bare repositories

On one particular system I like to keep a cluster of bare Git
repositories and spawn new-workdirs off of them.  Since the bare
repositories don't have working directories associated with them
they don't have a .git/ subdirectory that hosts the repository we
are linking to.

Using a bare repository as the backing repository for a workdir
created by this script does require that the user delete core.bare
from the repository's configuration file, so that Git auto-senses
the bareness of a repository based on pathname information, and
not based on the config file.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agomailsplit: fix for more than one input files
Junio C Hamano [Mon, 28 May 2007 22:48:07 +0000 (15:48 -0700)]
mailsplit: fix for more than one input files

Earlier commit d63bd9a broke the case where more than one input
files are fed to mailsplit by not incrementing the base counter
when splitting second and subsequent input files.  This should
fix it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocvsserver: Fix some typos in asciidoc documentation
Frank Lichtenheld [Sun, 27 May 2007 12:33:08 +0000 (14:33 +0200)]
cvsserver: Fix some typos in asciidoc documentation

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocvsserver: Note that CVS_SERVER can also be specified as method variable
Frank Lichtenheld [Sun, 27 May 2007 12:33:07 +0000 (14:33 +0200)]
cvsserver: Note that CVS_SERVER can also be specified as method variable

Reasonably new versions of the cvs CLI client allow one to
specifiy CVS_SERVER as a method variable directly in
CVSROOT. This is way more convinient than using an
environment variable since it gets saved in CVS/Root.

Since I only discovered this by accident I guess there
might be others out there that learnt CVS on the 1.11
series (or even earlier) and profit from such a note
about cvs improvements in the last couple years.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocvsserver: Correct inetd.conf example in asciidoc documentation
Frank Lichtenheld [Sun, 27 May 2007 12:33:06 +0000 (14:33 +0200)]
cvsserver: Correct inetd.conf example in asciidoc documentation

While the given example worked, it made us look rather
incompetent. Give the correct reason why one needs the
more complex syntax and change the example to reflect
that.

Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agouser-manual: fixed typo in example
Steffen Prohaska [Sat, 26 May 2007 19:16:27 +0000 (21:16 +0200)]
user-manual: fixed typo in example

Signed-off-by: Steffen Prohaska <prohaska@zib.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAdd test case for $Id$ expanded in the repository
Andy Parkins [Sun, 27 May 2007 10:52:11 +0000 (11:52 +0100)]
Add test case for $Id$ expanded in the repository

This test case would have caught the bug fixed by revision
c23290d5.

It puts various forms of $Id$ into a file in the repository,
without allowing git to collapse them to uniformity.  Then enables the
$Id$ expansion on checkout, and checks that what is checked out has
coped with the various forms.

Signed-off-by: Andy Parkins <andyparkins@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint-1.5.1' into maint
Junio C Hamano [Tue, 29 May 2007 06:54:26 +0000 (23:54 -0700)]
Merge branch 'maint-1.5.1' into maint

* maint-1.5.1:
  git-svn: avoid md5 calculation entirely if SVN doesn't provide one
  Fix stupid typo in lookup_tag()

17 years agogit-svn: avoid md5 calculation entirely if SVN doesn't provide one
Eric Wong [Sun, 27 May 2007 22:59:01 +0000 (15:59 -0700)]
git-svn: avoid md5 calculation entirely if SVN doesn't provide one

There's no point in calculating an MD5 if we're not going to use
it.  We'll also avoid the possibility of there being a bug in the
Perl MD5 library not being able to handle zero-sized files.

This is a followup to 20b3d206acbbb042c7ad5f42d36ff8d036a538c5,
which allows us to track repositories that do not provide MD5
checksums.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint' of git://repo.or.cz/git-gui into maint
Junio C Hamano [Tue, 29 May 2007 03:23:10 +0000 (20:23 -0700)]
Merge branch 'maint' of git://repo.or.cz/git-gui into maint

* 'maint' of git://repo.or.cz/git-gui:
  git-gui: Guess our share/git-gui/lib path at runtime if possible
  Correct key bindings to Control-<foo>
  git-gui: Tighten internal pattern match for lib/ directory

17 years agoMakefile: Remove git-fsck and git-verify-pack from PROGRAMS
Nguyen Thai Ngoc Duy [Tue, 29 May 2007 02:05:43 +0000 (22:05 -0400)]
Makefile: Remove git-fsck and git-verify-pack from PROGRAMS

Those are builtins. Remove them from PROGRAMS variable

Signed-off-by: Nguyễn Thái Ngọc Duy <pclouds@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix stupid typo in lookup_tag()
Johan Herland [Mon, 28 May 2007 23:21:25 +0000 (01:21 +0200)]
Fix stupid typo in lookup_tag()

Signed-off-by: Johan Herland <johan@herland.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-gui: Guess our share/git-gui/lib path at runtime if possible gitgui-0.7.2
Shawn O. Pearce [Sun, 27 May 2007 04:03:37 +0000 (00:03 -0400)]
git-gui: Guess our share/git-gui/lib path at runtime if possible

Johannes Sixt asked me to try to avoid embedding the runtime location
of git-gui's library directory in the executable script.  Not embedding
it helps the MinGW to be relocatable to another directory should a user
wish to install the programs in a directory other than the location the
packager wanted them to be installed into.

Most of this is a hack.  We try to determine if the path of our master
git-gui script will be able to locate the lib by ../share/git-gui/lib.
This should be true if $(gitexecdir) and $(libdir) have the same prefix.
If they do then we defer the assignment of $(libdir) until runtime, and
we get it from $argv0 rather than embedding it into the script itself.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
17 years agoupdate diff-delta.c copyright
Nicolas Pitre [Sat, 26 May 2007 02:16:27 +0000 (22:16 -0400)]
update diff-delta.c copyright

There is actually nothing left from the original LibXDiff code I used
over 2 years ago, and even the GIT implementation has diverged quite a
bit from LibXDiff's at this point.  Let's update the copyright notice
to better reflect that fact.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoimprove delta long block matching with big files
Nicolas Pitre [Sat, 26 May 2007 01:38:58 +0000 (21:38 -0400)]
improve delta long block matching with big files

Martin Koegler noted that create_delta() performs a new hash lookup
after every block copy encoding which are currently limited to 64KB.

In case of larger identical blocks, the next hash lookup would normally
point to the next 64KB block in the reference buffer and multiple block
copy operations will be consecutively encoded.

It is however possible that the reference buffer be sparsely indexed if
hash buckets have been trimmed down in create_delta_index() when hashing
of the reference buffer isn't well balanced.  In that case the hash
lookup following a block copy might fail to match anything and the fact
that the reference buffer still matches beyond the previous 64KB block
will be missed.

Let's rework the code so that buffer comparison isn't bounded to 64KB
anymore.  The match size should be as large as possible up front and
only then should multiple block copy be encoded to cover it all.
Also, fewer hash lookups will be performed in the end.

According to Martin, this patch should reduce his 92MB pack down to 75MB
with the dataset he has.

Tests performed on the Linux kernel repo show a slightly smaller pack and
a slightly faster repack.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAttempt to delay prepare_alt_odb during get_sha1
Shawn O. Pearce [Sat, 26 May 2007 05:25:11 +0000 (01:25 -0400)]
Attempt to delay prepare_alt_odb during get_sha1

Not every input value passed to get_sha1 is an abbreviated SHA-1.
Its actually quite common for refs to be passed and for those
refs to resolve to full SHA-1s, in which case we may not need to
initialize the alternate object database list in this process.

I'm relocating the call to prepare_alt_odb closer to the code
that actually needs it to maintain the fix first introduced by
Junio in 99a19b43 (to avoid ambiguous SHA-1 abbreviations from
being accepted).  This allows us to avoid the alt_odb list setup
if we won't actually need it.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMicro-optimize prepare_alt_odb
Shawn O. Pearce [Sat, 26 May 2007 05:24:40 +0000 (01:24 -0400)]
Micro-optimize prepare_alt_odb

Calling getenv() is not that expensive, but its also not free,
and its certainly not cheaper than testing to see if alt_odb_tail
is not null.

Because we are calling prepare_alt_odb() from within find_sha1_file
every time we cannot find an object file locally we want to skip out
of prepare_alt_odb() as early as possible once we have initialized
our alternate list.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoLazily open pack index files on demand
Shawn O. Pearce [Sat, 26 May 2007 05:24:19 +0000 (01:24 -0400)]
Lazily open pack index files on demand

In some repository configurations the user may have many packfiles,
but all of the recent commits/trees/tags/blobs are likely to
be in the most recent packfile (the one with the newest mtime).
It is therefore common to be able to complete an entire operation
by accessing only one packfile, even if there are 25 packfiles
available to the repository.

Rather than opening and mmaping the corresponding .idx file for
every pack found, we now only open and map the .idx when we suspect
there might be an object of interest in there.

Of course we cannot known in advance which packfile contains an
object, so we still need to scan the entire packed_git list to
locate anything.  But odds are users want to access objects in the
most recently created packfiles first, and that may be all they
ever need for the current operation.

Junio observed in b867092f that placing recent packfiles before
older ones can slightly improve access times for recent objects,
without degrading it for historical object access.

This change improves upon Junio's observations by trying even harder
to avoid the .idx files that we won't need.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>