git/git.git
15 years agoMerge branch 'fixes'
Junio C Hamano [Sun, 25 Sep 2005 06:50:36 +0000 (23:50 -0700)]
Merge branch 'fixes'

15 years agoClean generated files a bit more.
Junio C Hamano [Sun, 25 Sep 2005 06:48:29 +0000 (23:48 -0700)]
Clean generated files a bit more.

Now we conditionally compile things in compat/, so we should remove
object files there.  Python execution can leave *.pyc and *.pyo, which
need to be cleaned as well.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoStop installing the backward compatible symlinks.
Junio C Hamano [Sun, 25 Sep 2005 06:42:34 +0000 (23:42 -0700)]
Stop installing the backward compatible symlinks.

Also cmd-renames.sh can now be used to remove the backward compatible
symlinks -- this is not used by default in any way.

As discussed on the list with Pasky, git-ssh-push and git-ssh-pull will
keep calling each other for a while longer.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoshow-branch: fix commit naming breakage.
Junio C Hamano [Sun, 25 Sep 2005 06:33:02 +0000 (23:33 -0700)]
show-branch: fix commit naming breakage.

It was ignoring the generation number of the commit when naming 2nd
and later parents, showing "(linus^n)^2" for any <n> incorrectly as
"linus^2".

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Finish documenting trivial merge rules
Daniel Barkalow [Sun, 25 Sep 2005 03:50:04 +0000 (23:50 -0400)]
[PATCH] Finish documenting trivial merge rules

Fix missing symbol explanations, a few incorrect cases, and add
two-way merge rules.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agogit-grep: fix 'git grep -e $pattern' handling
Junio C Hamano [Sat, 24 Sep 2005 22:09:48 +0000 (15:09 -0700)]
git-grep: fix 'git grep -e $pattern' handling

People typically say 'grep -e $pattern' because $pattern has a leading
dash which would be mistaken as a grep flag.  Make sure we pass -e in
front of $pattern when we invoke grep.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoRemove use of 'xargs -0' from git-reset.
Junio C Hamano [Sat, 24 Sep 2005 22:02:35 +0000 (15:02 -0700)]
Remove use of 'xargs -0' from git-reset.

Even without the trouble it causes to people without GNU xargs,
it was not really necessary to print from Perl and then remove it
outside.  Just unlink it inside Perl.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoUse 0.99.7.GIT as version between 0.99.7 and 0.99.8
Junio C Hamano [Sat, 24 Sep 2005 21:58:57 +0000 (14:58 -0700)]
Use 0.99.7.GIT as version between 0.99.7 and 0.99.8

Pasky taught me how he does his versioning for ELinks.  This will sort
after 0.99.7 and interim fixes 0.99.7a, and before 0.99.8.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoSolaris: give a bit more built-in defaults.
Junio C Hamano [Fri, 23 Sep 2005 17:41:40 +0000 (10:41 -0700)]
Solaris: give a bit more built-in defaults.

Taking the make command line Peter Eriksen uses, give defaults
to SHELL_PATH, TAR, CURLDIR, NO_STRCASESTR, and INSTALL.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] CURLDIR in Makefile
Patrick Mauritz [Mon, 19 Sep 2005 14:11:19 +0000 (16:11 +0200)]
[PATCH] CURLDIR in Makefile

Support systems that do not install curl headers and libraries
in /usr/{include,lib}.

Signed-off-by: Patrick Mauritz <oxygene@studentenbude.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Rename daemon.c's lognotice() to loginfo()
Petr Baudis [Sat, 24 Sep 2005 14:13:03 +0000 (16:13 +0200)]
[PATCH] Rename daemon.c's lognotice() to loginfo()

The syslog code logs with severity LOG_INFO in the loginfo() function, so make
things less confusing.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] git-daemon --syslog to log through syslog
Petr Baudis [Sat, 24 Sep 2005 14:13:01 +0000 (16:13 +0200)]
[PATCH] git-daemon --syslog to log through syslog

Well, this makes it even more clear that we need the packet reader and
friends to use the daemon logging code. :/  Therefore, we at least indicate
in the "Disconnect" log message if the child process exitted with an error
code or not.

Idea by Linus.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Update git-daemon documentation wrt. the --verbose parameter
Petr Baudis [Sat, 24 Sep 2005 14:12:58 +0000 (16:12 +0200)]
[PATCH] Update git-daemon documentation wrt. the --verbose parameter

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoReally ignore generated distribution material.
Junio C Hamano [Sat, 24 Sep 2005 18:19:07 +0000 (11:19 -0700)]
Really ignore generated distribution material.
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agodaemon.c: pid_t is not int.
Junio C Hamano [Sat, 24 Sep 2005 06:26:55 +0000 (23:26 -0700)]
daemon.c: pid_t is not int.

Reported by Morten Welinder <mwelinder@gmail.com>.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge 'fixes' branch.
Junio C Hamano [Sat, 24 Sep 2005 01:46:34 +0000 (18:46 -0700)]
Merge 'fixes' branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFurther clarify licensing status of compat/subprocess.py.
Junio C Hamano [Sat, 24 Sep 2005 01:43:53 +0000 (18:43 -0700)]
Further clarify licensing status of compat/subprocess.py.

PSF license explicitly states the files in Python distribution is
compatible with GPL, and upstream clarified the licensing terms by
shortening its file header.  This version is a verbatim copy from
release24-maint branch form Python CVS.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] rsh.c unterminated string
H. Peter Anvin [Fri, 23 Sep 2005 23:30:50 +0000 (16:30 -0700)]
[PATCH] rsh.c unterminated string

The change I made to rsh.c would leave the string unterminated under
certain conditions, which unfortunately always applied!  This patch
fixes this.  For some reason this never bit on i386 or ppc, but bit me
on x86-64.

Fix situation where the buffer was not properly null-terminated.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] git-local-fetch: Avoid confusing error messages on packed repositories
Sergey Vlasov [Fri, 23 Sep 2005 12:28:38 +0000 (16:28 +0400)]
[PATCH] git-local-fetch: Avoid confusing error messages on packed repositories

If the source repository was packed, and git-local-fetch needed to
fetch a pack file, it spewed a misleading error message about not
being able to find the unpacked object.  Fixed by adding the
warn_if_not_exists argument to copy_file(), which controls printing
of error messages in case the source file does not exist.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Fix "git-local-fetch -s" with packed source repository
Sergey Vlasov [Fri, 23 Sep 2005 12:28:33 +0000 (16:28 +0400)]
[PATCH] Fix "git-local-fetch -s" with packed source repository

"git-local-fetch -s" did not work with a packed repository, because
symlink() happily created a link to a non-existing object file,
therefore fetch_file() always returned success, and fetch_pack() was
not called.  Fixed by calling stat() before symlink() to ensure the
file really exists.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] git-local-fetch: Avoid calling close(-1)
Sergey Vlasov [Fri, 23 Sep 2005 12:28:28 +0000 (16:28 +0400)]
[PATCH] git-local-fetch: Avoid calling close(-1)

After open() failure, copy_file() called close(ifd) with ifd == -1
(harmless, but causes Valgrind noise).  The same thing was possible
for the destination file descriptor.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] git-local-fetch: Fix error checking and leak in setup_indices()
Sergey Vlasov [Fri, 23 Sep 2005 12:28:23 +0000 (16:28 +0400)]
[PATCH] git-local-fetch: Fix error checking and leak in setup_indices()

setup_indices() did not check the return value of opendir(), and
did not have a corresponding closedir() call.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Plug memory leak in process_tree()
Sergey Vlasov [Fri, 23 Sep 2005 12:28:18 +0000 (16:28 +0400)]
[PATCH] fetch.c: Plug memory leak in process_tree()

When freeing a tree entry, must free its name too.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Do not build object ref lists
Sergey Vlasov [Fri, 23 Sep 2005 12:28:13 +0000 (16:28 +0400)]
[PATCH] fetch.c: Do not build object ref lists

The fetch code does not need object ref lists; by disabling them we
can save some time and memory.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] document command to show diff of a commit
Robert Watson [Fri, 23 Sep 2005 10:06:04 +0000 (11:06 +0100)]
[PATCH] document command to show diff of a commit

Document the best way to show the change introduced by a
commit, based on the suggestion by Linus on the list.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoIgnore a bit more generated files.
Junio C Hamano [Fri, 23 Sep 2005 21:02:57 +0000 (14:02 -0700)]
Ignore a bit more generated files.
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch 'fixes'
Junio C Hamano [Fri, 23 Sep 2005 07:53:43 +0000 (00:53 -0700)]
Merge branch 'fixes'

15 years agoMake 'git diff --cached' synonymous to 'git diff --cached HEAD'.
Junio C Hamano [Fri, 23 Sep 2005 07:53:02 +0000 (00:53 -0700)]
Make 'git diff --cached' synonymous to 'git diff --cached HEAD'.

When making changes to different files (i.e. dirty working tree) and
committing logically separate changes in groups, often it is necessary
to run 'git diff --cached HEAD' to make sure that the changes being
committed makes sense.  Saying 'git diff --cached' by mistake gives
rather uninformative error message from git-diff-files complaining it
does not understand --cached flag.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove call to parse_object() from process()
Sergey Vlasov [Wed, 21 Sep 2005 16:34:29 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove call to parse_object() from process()

The call to parse_object() in process() is not actually needed - if
the object type is unknown, parse_object() will be called by loop();
if the type is known, the object will be parsed by the appropriate
process_*() function.

After this change blobs which exist locally are no longer parsed,
which gives about 2x CPU usage improvement; the downside is that there
will be no warnings for existing corrupted blobs, but detecting such
corruption is the job of git-fsck-objects, not the fetch programs.
Newly fetched objects are still checked for corruption in http-fetch.c
and ssh-fetch.c (local-fetch.c does not seem to do it, but the removed
parse_object() call would not be reached for new objects anyway).

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Clean up object flag definitions
Sergey Vlasov [Wed, 21 Sep 2005 16:34:24 +0000 (20:34 +0400)]
[PATCH] fetch.c: Clean up object flag definitions

Remove holes left after deleting flags, and use shifts to emphasize
that flags are single bits.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove redundant test of TO_SCAN in process()
Sergey Vlasov [Wed, 21 Sep 2005 16:34:19 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove redundant test of TO_SCAN in process()

If the SEEN flag was not set, the TO_SCAN flag cannot be set,
therefore testing it is pointless.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove some duplicated code in process()
Sergey Vlasov [Wed, 21 Sep 2005 16:34:14 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove some duplicated code in process()

It does not matter if we call prefetch() or set the TO_SCAN flag before
or after adding the object to process_queue.  However, doing it before
object_list_insert() allows us to kill 3 lines of duplicated code.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove redundant TO_FETCH flag
Sergey Vlasov [Wed, 21 Sep 2005 16:34:09 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove redundant TO_FETCH flag

The TO_FETCH flag also became redundant after adding the SEEN flag -
it was set and checked in process() to prevent adding the same object
to process_queue multiple times, but now SEEN guards against this.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove redundant SCANNED flag
Sergey Vlasov [Wed, 21 Sep 2005 16:34:04 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove redundant SCANNED flag

After adding the SEEN flag, the SCANNED flag became obviously
redundant - each object can get into process_queue through process()
only once, and therefore multiple calls to process_object() for the
same object are not possible.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Make process() look at each object only once
Sergey Vlasov [Wed, 21 Sep 2005 16:33:59 +0000 (20:33 +0400)]
[PATCH] fetch.c: Make process() look at each object only once

The process() function is very often called multiple times for the
same object (because lots of trees refer to the same blobs), but did
not have a fast check for this, therefore a lot of useless calls to
has_sha1_file() and parse_object() were made before discovering that
nothing needs to be done.

This patch adds the SEEN flag which is used in process() to make it
look at each object only once.  When testing git-local-fetch on the
repository of GIT, this gives a 14x improvement in CPU usage (mainly
because the redundant calls to parse_object() are now avoided -
parse_object() always unpacks and parses the object data, even if it
was already parsed before).

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove useless lookup_object_type() call in process()
Sergey Vlasov [Wed, 21 Sep 2005 16:33:54 +0000 (20:33 +0400)]
[PATCH] fetch.c: Remove useless lookup_object_type() call in process()

In all places where process() is called except the one in pull() (which
is executed only once) the pointer to the object is already available,
so pass it as the argument to process() instead of sha1 and avoid an
unneeded call to lookup_object_type().

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoDo not give alarming error message from rsync in fetch and clone.
Junio C Hamano [Tue, 20 Sep 2005 06:52:33 +0000 (23:52 -0700)]
Do not give alarming error message from rsync in fetch and clone.

When we check the optional objects/info/alternates file at the remote
repository, we forgot to really squelch error message from rsync.

Not having that file is not a crime.

Signed-off-by: Junio C Hamano <junkio@cox.net>
(cherry picked from 89d844d084f14bc9506f63cd3c9aa44b21b49067 commit)

15 years ago[PATCH] Retitle 'inspecting what happened' section.
Jon Loeliger [Thu, 22 Sep 2005 15:51:57 +0000 (10:51 -0500)]
[PATCH] Retitle 'inspecting what happened' section.

In the tutorial, there is a section entitled "Checking it out"
that shows how to use diff log and whatchanged to insect some
of the repository state.

As the phrase "checkout" ususally carries some baggage WRT
other revision control mechanism, I suggest that we re-title
this section something like "Inspecting Changes".

Signed-off-by: Jon Loeliger <jdl@freescale.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoPrettyprint octopus merge message.
Junio C Hamano [Thu, 22 Sep 2005 07:55:22 +0000 (00:55 -0700)]
Prettyprint octopus merge message.

Including the current branch in the list of heads being merged
was not a good idea, so drop it.  And shorten the message by
grouping branches and tags together to form a single line.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Verbose git-daemon logging
Petr Baudis [Thu, 22 Sep 2005 09:25:28 +0000 (11:25 +0200)]
[PATCH] Verbose git-daemon logging

This patch makes git-daemon --verbose log some useful things on stderr -
in particular connects, disconnects and upload requests, and in such a
way to be able to trace a particular session. Some more errors are now
also logged (even when --verbose is not passed). It is still not perfect
since messages produced by the non-daemon-specific code are obviously
not formatted properly.

[jc: With minor fix up in the log line truncation, and
 use of write(2) as suggested by Linus.]

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoRetire diff-helper.
Junio C Hamano [Wed, 21 Sep 2005 06:24:39 +0000 (23:24 -0700)]
Retire diff-helper.

The textual diff generation with built-in '-p' in diff-* brothers has
proven to be useful enough that git-diff-helper outlived its usefulness.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoRetire rev-tree.
Junio C Hamano [Sat, 17 Sep 2005 05:30:55 +0000 (22:30 -0700)]
Retire rev-tree.

Some old scripts might still use git-rev-tree, but it really is
clearly inferior in every way to git-rev-list that such scripts should
be fixed anyway. Fixing them should be pretty easy.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoRetire git-export.
Junio C Hamano [Sat, 17 Sep 2005 05:21:36 +0000 (22:21 -0700)]
Retire git-export.

git-export was done as a concept example on how easy it is to export
the git data to something else. It's much less powerful than any
number of trivial one-liner scripts now, and real exporters would not
ever use git-export.

It's obviously much less powerful than "git-whatchanged", or just
about any combination of git-rev-list + git-diff-tree.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Fix strange timezone handling
Linus Torvalds [Wed, 21 Sep 2005 22:50:28 +0000 (15:50 -0700)]
[PATCH] Fix strange timezone handling

We generate the ASCII representation of our internal date representation
("seconds since 1970, UTC + timezone information") in two different
places.

One of them uses the stupid and obvious way to make sure that it gets the
sexagecimal representation right for negative timezones even if they might
not be exact hours, and the other one depends on the modulus operator
always matching the sign of argument.

Hey, the clever one works. And C90 even specifies that behaviour. But I
had to think about it for a while when I was re-visiting this area, and
even if I didn't have to, it's kind of strange to have two different ways
to print out the same data format.

So use a common helper for this. And select the stupid and straighforward
way.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoRevert "Use git-merge instead of git-resolve in git-pull."
Junio C Hamano [Wed, 21 Sep 2005 21:01:56 +0000 (14:01 -0700)]
Revert "Use git-merge instead of git-resolve in git-pull."

This reverts f887564ab72e107bcdee3ba83c91e2bb4ae13ca7 commit.

15 years agoRevert "Make Octopus merge message a bit nicer."
Junio C Hamano [Wed, 21 Sep 2005 20:59:54 +0000 (13:59 -0700)]
Revert "Make Octopus merge message a bit nicer."

This reverts 63f1aa6c72c46928f1b6959437aed4becbc42ff3 commit.

15 years ago[PATCH] fetch.c: Remove call to parse_object() from process()
Sergey Vlasov [Wed, 21 Sep 2005 16:34:29 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove call to parse_object() from process()

The call to parse_object() in process() is not actually needed - if
the object type is unknown, parse_object() will be called by loop();
if the type is known, the object will be parsed by the appropriate
process_*() function.

After this change blobs which exist locally are no longer parsed,
which gives about 2x CPU usage improvement; the downside is that there
will be no warnings for existing corrupted blobs, but detecting such
corruption is the job of git-fsck-objects, not the fetch programs.
Newly fetched objects are still checked for corruption in http-fetch.c
and ssh-fetch.c (local-fetch.c does not seem to do it, but the removed
parse_object() call would not be reached for new objects anyway).

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Clean up object flag definitions
Sergey Vlasov [Wed, 21 Sep 2005 16:34:24 +0000 (20:34 +0400)]
[PATCH] fetch.c: Clean up object flag definitions

Remove holes left after deleting flags, and use shifts to emphasize
that flags are single bits.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove redundant test of TO_SCAN in process()
Sergey Vlasov [Wed, 21 Sep 2005 16:34:19 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove redundant test of TO_SCAN in process()

If the SEEN flag was not set, the TO_SCAN flag cannot be set,
therefore testing it is pointless.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove some duplicated code in process()
Sergey Vlasov [Wed, 21 Sep 2005 16:34:14 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove some duplicated code in process()

It does not matter if we call prefetch() or set the TO_SCAN flag before
or after adding the object to process_queue.  However, doing it before
object_list_insert() allows us to kill 3 lines of duplicated code.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove redundant TO_FETCH flag
Sergey Vlasov [Wed, 21 Sep 2005 16:34:09 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove redundant TO_FETCH flag

The TO_FETCH flag also became redundant after adding the SEEN flag -
it was set and checked in process() to prevent adding the same object
to process_queue multiple times, but now SEEN guards against this.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove redundant SCANNED flag
Sergey Vlasov [Wed, 21 Sep 2005 16:34:04 +0000 (20:34 +0400)]
[PATCH] fetch.c: Remove redundant SCANNED flag

After adding the SEEN flag, the SCANNED flag became obviously
redundant - each object can get into process_queue through process()
only once, and therefore multiple calls to process_object() for the
same object are not possible.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Make process() look at each object only once
Sergey Vlasov [Wed, 21 Sep 2005 16:33:59 +0000 (20:33 +0400)]
[PATCH] fetch.c: Make process() look at each object only once

The process() function is very often called multiple times for the
same object (because lots of trees refer to the same blobs), but did
not have a fast check for this, therefore a lot of useless calls to
has_sha1_file() and parse_object() were made before discovering that
nothing needs to be done.

This patch adds the SEEN flag which is used in process() to make it
look at each object only once.  When testing git-local-fetch on the
repository of GIT, this gives a 14x improvement in CPU usage (mainly
because the redundant calls to parse_object() are now avoided -
parse_object() always unpacks and parses the object data, even if it
was already parsed before).

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] fetch.c: Remove useless lookup_object_type() call in process()
Sergey Vlasov [Wed, 21 Sep 2005 16:33:54 +0000 (20:33 +0400)]
[PATCH] fetch.c: Remove useless lookup_object_type() call in process()

In all places where process() is called except the one in pull() (which
is executed only once) the pointer to the object is already available,
so pass it as the argument to process() instead of sha1 and avoid an
unneeded call to lookup_object_type().

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch 'fixes'
Junio C Hamano [Wed, 21 Sep 2005 19:31:33 +0000 (12:31 -0700)]
Merge branch 'fixes'

15 years agoFix documentation dependency.
Junio C Hamano [Wed, 21 Sep 2005 19:29:59 +0000 (12:29 -0700)]
Fix documentation dependency.

Randal L. Schwartz noticed that 'make install' does not rebuild what
is installed.  Make the 'install' rule depend on 'man'.

I noticed also 'touch' of the source files were used to express include
dependencies, which is a no-no.  Rewrite it to do dependencies properly,
and add missing include dependencies while we are at it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMake object creation in http fetch a bit safer.
Junio C Hamano [Wed, 21 Sep 2005 09:46:32 +0000 (02:46 -0700)]
Make object creation in http fetch a bit safer.

Unlike write_sha1_file() that tries to create the object file in a
temporary location and then move it to the final location, fetch_object
could have been interrupted in the middle, leaving a corrupt file.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoClarify dual license status of subprocess.py file.
Junio C Hamano [Wed, 21 Sep 2005 07:58:32 +0000 (00:58 -0700)]
Clarify dual license status of subprocess.py file.

The author of the file we stole from Python 2.4 distribution, Peter
Astrand <astrand@lysator.liu.se>, OK'ed to add this at the end of the
licensing terms section of the file:

    Use of this file within git is permitted under GPLv2.

Signed-off-by: Junio C Hamano <junkio@cox.net>
(cherry picked from 089f20dce19711d34f5383ee289a9b1fbd3f3307 commit)

15 years agoUpdate tutorial with Octopus usage.
Junio C Hamano [Wed, 21 Sep 2005 01:21:10 +0000 (18:21 -0700)]
Update tutorial with Octopus usage.

Making an Octopus is simply a natural extension of merging just one
branch into the current branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMake Octopus merge message a bit nicer.
Junio C Hamano [Mon, 19 Sep 2005 07:03:47 +0000 (00:03 -0700)]
Make Octopus merge message a bit nicer.

Linus says that 'of .' to mean the commits came from the local repository
was too confusing and ugly -- I tend to agree with him.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoUse git-merge instead of git-resolve in git-pull.
Junio C Hamano [Sat, 17 Sep 2005 06:38:59 +0000 (23:38 -0700)]
Use git-merge instead of git-resolve in git-pull.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMake time-based commit filtering work with topological ordering.
Linus Torvalds [Wed, 21 Sep 2005 00:55:46 +0000 (17:55 -0700)]
Make time-based commit filtering work with topological ordering.

The trick is to consider the time-based filtering a limiter, the same way
we do for release ranges.

That means that the time-based filtering runs _before_ the topological
sorting, which makes it meaningful again. It also simplifies the code
logic.

This makes "gitk" useful with time ranges.

[ Second version: --merge-order now unaffected by the re-org ]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Teach "git-rev-parse" about date-based cut-offs
Linus Torvalds [Tue, 20 Sep 2005 21:13:24 +0000 (14:13 -0700)]
[PATCH] Teach "git-rev-parse" about date-based cut-offs

This adds the options "--since=date" and "--before=date" to git-rev-parse,
which knows how to translate them into seconds since the epoch for
git-rev-list.

With this, you can do

git log --since="2 weeks ago"

or

git log --until=yesterday

to show the commits that have happened in the last two weeks or are
older than 24 hours, respectively.

The flags "--after=" and "--before" are synonyms for --since and --until,
and you can combine them, so

git log --after="Aug 5" --before="Aug 10"

is a valid (but strange) thing to do.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] ARM optimized SHA1 implementation
Nicolas Pitre [Tue, 20 Sep 2005 16:27:13 +0000 (12:27 -0400)]
[PATCH] ARM optimized SHA1 implementation

This is my ARM assembly SHA1 implementation for GIT. It is approximately
50% faster than the generic C version. On an XScale processor running at
400MHz:

generic C version: 9.8 MB/s
my version: 14.5 MB/s

It's not that I expect a lot of big GIT users on ARM, but I stillknow
about one important ARM user that might benefit from it, and writing
that code was fun.

I also reworked the makefile a bit so any optimized SHA1 implementations
is used regardless of whether NO_OPENSSL is defined or not.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Make the git-fsck-objects diagnostics more useful
Petr Baudis [Tue, 20 Sep 2005 18:56:05 +0000 (20:56 +0200)]
[PATCH] Make the git-fsck-objects diagnostics more useful

Actually report what exactly is wrong with the object, instead of an
ambiguous 'bad sha1 file' or such. In places where we already do, unify
the format and clean the messages up.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoDo not give alarming error message from rsync in fetch and clone.
Junio C Hamano [Tue, 20 Sep 2005 06:52:33 +0000 (23:52 -0700)]
Do not give alarming error message from rsync in fetch and clone.

When we check the optional objects/info/alternates file at the remote
repository, we forgot to really squelch error message from rsync.

Not having that file is not a crime.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Return proper error valud from "parse_date()"
Linus Torvalds [Mon, 19 Sep 2005 22:53:50 +0000 (15:53 -0700)]
[PATCH] Return proper error valud from "parse_date()"

Right now we don't return any error value at all from parse_date(), and if
we can't parse it, we just silently leave the result buffer unchanged.

That's fine for the current user, which will always default to the current
date, but it's a crappy interface, and we might well be better off with an
error message rather than just the default date.

So let's change the thing to return a negative value if an error occurs,
and the length of the result otherwise (snprintf behaviour: if the buffer
is too small, it returns how big it _would_ have been).

[ I started looking at this in case we could support date-based revision
  names. Looks ugly. Would have to parse relative dates.. ]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoShow modified files in git-ls-files
Junio C Hamano [Mon, 19 Sep 2005 22:11:15 +0000 (15:11 -0700)]
Show modified files in git-ls-files

Add -m/--modified to show files that have been modified wrt. the index.

[jc: The original came from Brian Gerst on Sep 1st but it only checked
if the paths were cache dirty without actually checking the files were
modified.  I also added the usage string and a new test.]

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFast-path 'update-index --refresh' a bit.
Junio C Hamano [Mon, 19 Sep 2005 21:49:50 +0000 (14:49 -0700)]
Fast-path 'update-index --refresh' a bit.

If the length in the stat information does not match what is recorded
in the index, there is no point rehashing the contents to see if the
index entry can be refreshed.

We need to be a bit careful.  Immediately after read-tree or
checkout-index without -u, ce_size is set to zero and does not match
the length of the blob that is recorded, and we need to actually look
at the contents to see if it has been changed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoUse GECOS field a bit better to produce default human readable name.
Junio C Hamano [Mon, 19 Sep 2005 23:06:56 +0000 (16:06 -0700)]
Use GECOS field a bit better to produce default human readable name.

This updates the default human readable name we generate from GECOS
field.  We assume the "full-name, followed by additional information
separated by commas" format, with an & expanding to the capitalized
login name.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoDocument -i flag to git-read-tree
Junio C Hamano [Mon, 19 Sep 2005 18:33:14 +0000 (11:33 -0700)]
Document -i flag to git-read-tree

Somehow I missed it when we updated read-tree to support the recursive
merge strategy.  Also -i should require -m as well, which the command
did not check.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Documentation: Update all files to use the new gitlink: macro
Sergey Vlasov [Mon, 19 Sep 2005 10:10:51 +0000 (14:10 +0400)]
[PATCH] Documentation: Update all files to use the new gitlink: macro

The replacement was performed automatically by these commands:

perl -pi -e 's/link:(git.+)\.html\[\1\]/gitlink:$1\[1\]/g' \
README Documentation/*.txt
perl -pi -e 's/link:git\.html\[git\]/gitlink:git\[7\]/g' \
README Documentation/*.txt

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Documentation: Add asciidoc.conf file and gitlink: macro
Sergey Vlasov [Mon, 19 Sep 2005 10:10:14 +0000 (14:10 +0400)]
[PATCH] Documentation: Add asciidoc.conf file and gitlink: macro

Introduce an asciidoc.conf file with the purpose of adding a gitlink:
macro which will improve the manpage output.

Original cogito patch by Jonas Fonseca <fonseca@diku.dk>;
asciidoc.conf from that patch was further enhanced to use the proper
DocBook tag <citerefentry> for references to man pages.

Signed-off-by: Sergey Vlasov <vsu@altlinux.ru>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFix extended short SHA1 name completion
Linus Torvalds [Mon, 19 Sep 2005 22:16:03 +0000 (15:16 -0700)]
Fix extended short SHA1 name completion

get_sha1() would not do sha1 completion of short SHA1's when they were
part of a more complex expression.  So doing

git-rev-parse 727132834e6be48a93c1bd6458a29d474ce7d5d5^

would work, and return 87c6aeb4efdd4355918d127a91bd0adc5a02f8ff. But using
the shorthand version

git-rev-list 72713^

wouldn't work.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoShip our own copy of subprocess.py
Junio C Hamano [Tue, 20 Sep 2005 02:47:54 +0000 (19:47 -0700)]
Ship our own copy of subprocess.py

so people without the latest Python could run merge-recursive.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoshow-branch: Be nicer when running in a corrupt repository.
Junio C Hamano [Mon, 19 Sep 2005 08:12:18 +0000 (01:12 -0700)]
show-branch: Be nicer when running in a corrupt repository.

We may end up trying to print a commit we do not actually have but we
know about its existence only because another commit we do have refers
to it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
(cherry picked from b204feab9371040982d2c60611925e7693106c84 commit)

15 years ago[PATCH] Fix git-init-db creating crap directories.
Petr Baudis [Tue, 20 Sep 2005 00:19:50 +0000 (02:19 +0200)]
[PATCH] Fix git-init-db creating crap directories.

The base target directory for the templates copying was initialized
to git_dir, but git_dir[len] is not zero but / at the time we do the
initialization. This is not what we want for our target directory string
since we pass it to mkdir(), so make it zero-terminated manually.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Do not install compatibility symlink for what we do not install
Patrick Mauritz [Mon, 19 Sep 2005 14:11:19 +0000 (16:11 +0200)]
[PATCH] Do not install compatibility symlink for what we do not install

We sometimes do not install git-send-email nor git-http-pull; do not
unconditionally create symlinks to them.

Signed-off-by: Patrick Mauritz <oxygene@studentenbude.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Remove total confusion from "git checkout"
Linus Torvalds [Mon, 19 Sep 2005 17:11:18 +0000 (10:11 -0700)]
[PATCH] Remove total confusion from "git checkout"

The target to check out does not need to be a branch. The _result_ of the
checkout needs to be a branch. Don't confuse the two, and then insult the
user.

Insulting is ok, but I personally get really pissed off is a tool is both
confused and insulting. At least be _correct_ and insulting.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] strcasestr compatibility replacement
Linus Torvalds [Mon, 19 Sep 2005 01:30:50 +0000 (18:30 -0700)]
[PATCH] strcasestr compatibility replacement

Some C libraries lack strcasestr(); add a stupid replacement
to help folks with such.

[jc: original Linus posting, updated with his "also need <ctype.h>",
 updated further with a fix from Joachim B Haga <cjhaga@fys.uio.no>"]

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoGIT 0.99.7 v0.99.7
Junio C Hamano [Sun, 18 Sep 2005 22:45:59 +0000 (15:45 -0700)]
GIT 0.99.7

15 years agoArrgh -- another asciidoc caret workaround.
Junio C Hamano [Sun, 18 Sep 2005 22:45:09 +0000 (15:45 -0700)]
Arrgh -- another asciidoc caret workaround.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Improve git-update-index error reporting
Petr Baudis [Sun, 18 Sep 2005 19:09:22 +0000 (21:09 +0200)]
[PATCH] Improve git-update-index error reporting

This makes git-update-index error reporting much less confusing. The
user will know what went wrong with better precision, and will be given
a hopefully less confusing advice.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Improved "git add"
Linus Torvalds [Sun, 18 Sep 2005 18:27:45 +0000 (11:27 -0700)]
[PATCH] Improved "git add"

This fixes everybodys favourite complaint about "git add", namely that it
doesn't take directories.

We use "git-ls-files --others" to generate an arbitrary list of filenames,
and thus also automatically honor ignore-files while we're at it.

Side note: there's a lot of room for improvement here. In particular, if
we have a long list of filenames (importing a big archive), this will just
do a big stupid for-loop and add them one at a time. Maybe it should use

generate-list | xargs -0 git-update-idex --add --

instead.

Also, I think we should have a default ignore list if we don't find a
.git/info/exclude file. Ignoring "*.o" and ".*" by default would probably
be the right thing to do.

But I think this is a good first step.

Use the "-n" flag to just show the list of files to be added without
adding them.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Add "--git-dir" flag to git-rev-parse
Linus Torvalds [Sun, 18 Sep 2005 18:18:30 +0000 (11:18 -0700)]
[PATCH] Add "--git-dir" flag to git-rev-parse

Especially when you're deep inside the git repository, it's not all that
trivial for scripts to figure out where GIT_DIR is if it isn't set.

So add a flag to git-rev-parse to show where it is, since it will have
figured it out anyway.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Support alternates and http-alternates in http-fetch
Daniel Barkalow [Sun, 18 Sep 2005 18:14:19 +0000 (14:14 -0400)]
[PATCH] Support alternates and http-alternates in http-fetch

This allows the remote repository to refer to additional repositories
in a file objects/info/http-alternates or
objects/info/alternates. Each line may be:

 a relative path, starting with ../, to get from the objects directory
  of the starting repository to the objects directory of the added
  repository.

 an absolute path of the objects directory of the added repository (on
  the same server).

 (only in http-alternates) a full URL of the objects directory of the
  added repository.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoDocument extended SHA1 used by git-rev-parse.
Junio C Hamano [Sun, 18 Sep 2005 18:13:58 +0000 (11:13 -0700)]
Document extended SHA1 used by git-rev-parse.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agofetch() assumes we do not have the object.
Junio C Hamano [Sun, 18 Sep 2005 21:11:53 +0000 (14:11 -0700)]
fetch() assumes we do not have the object.

Bugfix for the previous one.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoMerge branch 'master' of .
Junio C Hamano [Sun, 18 Sep 2005 08:18:15 +0000 (01:18 -0700)]
Merge branch 'master' of .

15 years agoImprove the safety check used in fetch.c
Junio C Hamano [Sun, 18 Sep 2005 08:01:07 +0000 (01:01 -0700)]
Improve the safety check used in fetch.c

The recent safety check to trust only the commits we have made
things impossibly slow and turn out to waste a lot of memory.

This commit fixes it with the following improvements:

 - mark already scanned objects and avoid rescanning the same
   object again;

 - free the tree entries when we have scanned the tree entries;
   this is the same as b0d8923ec01fd91b75ab079034f89ced91500157
   which reduced memory usage by rev-list;

 - plug memory leak from the object_list dequeuing code;

 - use the process_queue not just for fetching but for scanning,
   to make things tail recursive to avoid deep recursion; the
   deep recursion was especially prominent when we cloned a big
   pack.

 - avoid has_sha1_file() call when we already know we do not have
   that object.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoArchive-destroying "git repack -a -d" bug.
Junio C Hamano [Sun, 18 Sep 2005 06:35:49 +0000 (23:35 -0700)]
Archive-destroying "git repack -a -d" bug.

Using "git repack -a -d" can destroy your git archive if you use it
twice in succession, because the new pack can be called the same as
the old pack.  Found by Linus.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoDo not fail after calling bisect_auto_next()
Junio C Hamano [Sat, 17 Sep 2005 20:51:03 +0000 (13:51 -0700)]
Do not fail after calling bisect_auto_next()

As a convenience measure, 'bisect bad' or 'bisect good' automatically
does 'bisect next' when it knows it can, but the result of that test
to see if it can was leaking through as the exit code from the whole
thing, which was bad.  Noticed by Anton Blanchard.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Add support for alternates in HTTP
Daniel Barkalow [Thu, 15 Sep 2005 03:26:08 +0000 (23:26 -0400)]
[PATCH] Add support for alternates in HTTP

This tries .../objects/info/http-alternates and then
.../objects/info/alternates, looking for a file which specifies where
else to download objects and packs from.

It currently only supports absolute paths, and doesn't support full URLs.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoFix typo in test comment.
Junio C Hamano [Sat, 17 Sep 2005 06:27:52 +0000 (23:27 -0700)]
Fix typo in test comment.

I do not know why it was spelled git-rev-tree when I meant to say
git-read-tree, but the typo was left since day one.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] getdomainname should be usable on SunOS with -lnsl
Junio C Hamano [Mon, 12 Sep 2005 05:25:49 +0000 (22:25 -0700)]
[PATCH] getdomainname should be usable on SunOS with -lnsl

Jason Riedy suggests that we should be able to use getdomainname
if we properly specify which libraries to link.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Make "git shortlog" understand raw logs
Linus Torvalds [Sat, 17 Sep 2005 01:02:46 +0000 (18:02 -0700)]
[PATCH] Make "git shortlog" understand raw logs

This is a nicer fix for git-shortlog being unable to handle the raw log
format. Just use a more permissive regexp instead of doing two nearly
identical ones.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Fix "git-rev-list" revision range parsing
Linus Torvalds [Sat, 17 Sep 2005 00:53:19 +0000 (17:53 -0700)]
[PATCH] Fix "git-rev-list" revision range parsing

There were two bugs in there:
 - if the range didn't end up working, we restored the '.' character in
   the wrong place.
 - an empty end-of-range should be interpreted as HEAD.

See rev-parse.c for the reference implementation of this.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoAdd git-send-email to .gitignore
Junio C Hamano [Sat, 17 Sep 2005 03:40:12 +0000 (20:40 -0700)]
Add git-send-email to .gitignore

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years agoTeach rsync transport about alternates.
Junio C Hamano [Sat, 17 Sep 2005 18:56:41 +0000 (11:56 -0700)]
Teach rsync transport about alternates.

For local operations and downloading and uploading via git aware protocols,
use of $GIT_OBJECT_DIRECTORY/info/alternates is recommended on the server
side for big projects that are derived from another one (like Linux kernel).

However, dumb protocols and rsync transport needs to resolve this on the
client end, which we did not bother doing until this week.

I noticed we use "rsync -z" but most of our payload is already compressed,
which was not quite right.  This commit also fixes it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
15 years ago[PATCH] Avoid building object ref lists when not needed
Linus Torvalds [Fri, 16 Sep 2005 21:55:33 +0000 (14:55 -0700)]
[PATCH] Avoid building object ref lists when not needed

The object parsing code builds a generic "this object references that
object" because doing a full connectivity check for fsck requires it.

However, nothing else really needs it, and it's quite expensive for
git-rev-list that can have tons of objects in flight.

So, exactly like the commit buffer save thing, add a global flag to
disable it, and use it in git-rev-list.

Before:

$ /usr/bin/time git-rev-list --objects v2.6.12..HEAD | wc -l
12.28user 0.29system 0:12.57elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+26718minor)pagefaults 0swaps
59124

After this change:

$ /usr/bin/time git-rev-list --objects v2.6.12..HEAD | wc -l
10.33user 0.18system 0:10.54elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+18509minor)pagefaults 0swaps
59124

and note how the number of pages touched by git-rev-list for this
particular object list has shrunk from 26,718 (104 MB) to 18,509 (72 MB).

Calculating the total object difference between two git revisions is still
clearly the most expensive git operation (both in memory and CPU time),
but it's now less than 40% of what it used to be.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>