git/git.git
14 years agoExplain "Not a git repository: '.git'".
Junio C Hamano [Fri, 12 Jan 2007 20:24:16 +0000 (12:24 -0800)]
Explain "Not a git repository: '.git'".

Andy Parkins noticed that the error message some "whole tree"
oriented commands emit is stated misleadingly when they refused
to run from a subdirectory.

We could probably allow some of them to work from a subdirectory
but that is a semantic change that could have unintended side
effects, so let's start at first by rewording the error message
to be easier to read without doing anything else to be safe.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agomerge-recursive: do not report the resulting tree object name
Junio C Hamano [Fri, 12 Jan 2007 20:05:58 +0000 (12:05 -0800)]
merge-recursive: do not report the resulting tree object name

It is not available in the outermost merge, and it is only
useful for debugging merge-recursive in the inner merges.

Sergey Vlasov noticed that the old code accesses an
uninitialized location.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-revert: Fix die before git-sh-setup defines it.
Bob Proulx [Fri, 12 Jan 2007 06:45:38 +0000 (23:45 -0700)]
git-revert: Fix die before git-sh-setup defines it.

The code previously checked it's own name and called 'die' upon
an error.  However 'die' was not yet defined because git-sh-setup
had not been sourced yet.  Instead simply write the error message
to stderr and exit with an error as was originally desired.

Signed-off-by: Bob Proulx <bob@proulx.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agofix documentation for git-commit --no-verify
Michael S. Tsirkin [Fri, 12 Jan 2007 07:49:35 +0000 (09:49 +0200)]
fix documentation for git-commit --no-verify

Despite what the documentation claims, git-commit does not check commit
for suspicious lines: all hooks are disabled by default,
and the pre-comit hook could be changed to do something else.

Signed-off-by: Michael S. Tsirkin <mst@mellanox.co.il>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoFix up totally buggered read_or_die()
Linus Torvalds [Fri, 12 Jan 2007 04:37:38 +0000 (20:37 -0800)]
Fix up totally buggered read_or_die()

The "read_or_die()" function would silently NOT die for a partial read,
and since it was of type "void" it obviously couldn't even return the
partial number of bytes read.

IOW, it was totally broken. This hopefully fixes it up.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoClean up write_in_full() users
Linus Torvalds [Fri, 12 Jan 2007 04:23:00 +0000 (20:23 -0800)]
Clean up write_in_full() users

With the new-and-improved write_in_full() semantics, where a partial write
simply always returns a real error (and always sets 'errno' when that
happens, including for the disk full case), a lot of the callers of
write_in_full() were just unnecessarily complex.

In particular, there's no reason to ever check for a zero length or
return: if the length was zero, we'll return zero, otherwise, if a disk
full resulted in the actual write() system call returning zero the
write_in_full() logic would have correctly turned that into a negative
return value, with 'errno' set to ENOSPC.

I really wish every "write_in_full()" user would just check against "<0"
now, but this fixes the nasty and stupid ones.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoreflog-expire: brown paper bag fix.
Junio C Hamano [Fri, 12 Jan 2007 03:56:43 +0000 (19:56 -0800)]
reflog-expire: brown paper bag fix.

When --stale-fix is not passed, the code did not initialize the
two commit objects properly.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoGIT v1.5.0-rc1 v1.5.0-rc1
Junio C Hamano [Fri, 12 Jan 2007 02:22:48 +0000 (18:22 -0800)]
GIT v1.5.0-rc1

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoplug a few leaks in revision walking used in describe.
Junio C Hamano [Thu, 11 Jan 2007 06:36:16 +0000 (22:36 -0800)]
plug a few leaks in revision walking used in describe.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoChose better tag names in git-describe after merges.
Shawn O. Pearce [Wed, 10 Jan 2007 11:39:47 +0000 (06:39 -0500)]
Chose better tag names in git-describe after merges.

Recently git.git itself encountered a situation on its master and
next branches where git-describe stopped reporting 'v1.5.0-rc0-gN'
and instead started reporting 'v1.4.4.4-gN'.  This appeared to be
a backward jump in version numbering.

  maint     o-------------------4
            \                    \
  master     o-o-o-o-o-o-o-5-o-C-o-W

The issue is that commit C in the diagram claims it is version
1.5.0, as the tag v1.5.0 is placed on commit 5.  Yet commit W
claims it is version 1.4.4.4 as the tag v1.5.0 has an older tag
date than the v1.4.4.4 tag.

As it turns out this situation is very common.  A bug fix applied
to maint and later merged into master occurs frequently enough that
it should Just Work Right(tm).

Rather than taking the first tag that gets found git-describe will
now generate a list of all possible tags and select the one which
has the most number of commits in common with HEAD (or whatever
revision the user requested the description of).

This rule is based on the principle shown in the diagram above.
There are a large number of commits on the primary development branch
'master' which do not appear in the 'maint' branch, and many of
these are already tagged as part of v1.5.0-rc0.  Additionally these
commits are not in v1.4.4.4, as they are part of the v1.5.0 release
still being developed.  The v1.5.0-rc0 tag is more descriptive of
W than v1.4.4.4 is, and therefore should be used.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMerge branch 'jc/bare'
Junio C Hamano [Fri, 12 Jan 2007 00:50:36 +0000 (16:50 -0800)]
Merge branch 'jc/bare'

* jc/bare:
  Disallow working directory commands in a bare repository.
  git-fetch: allow updating the current branch in a bare repository.
  Introduce is_bare_repository() and core.bare configuration variable
  Move initialization of log_all_ref_updates

14 years agoMerge branch 'ar/merge-recursive'
Junio C Hamano [Fri, 12 Jan 2007 00:48:28 +0000 (16:48 -0800)]
Merge branch 'ar/merge-recursive'

* ar/merge-recursive:
  merge-recursive: do not use on-file index when not needed.
  Speed-up recursive by flushing index only once for all entries

14 years agoMerge branch 'jc/detached-head'
Junio C Hamano [Fri, 12 Jan 2007 00:47:34 +0000 (16:47 -0800)]
Merge branch 'jc/detached-head'

* jc/detached-head:
  git-checkout: handle local changes sanely when detaching HEAD
  git-checkout: safety check for detached HEAD checks existing refs
  git-checkout: fix branch name output from the command
  git-checkout: safety when coming back from the detached HEAD state.
  git-checkout: rewording comments regarding detached HEAD.
  git-checkout: do not warn detaching HEAD when it is already detached.
  Detached HEAD (experimental)
  git-branch: show detached HEAD
  git-status: show detached HEAD

14 years agogit-status: wording update to deal with deleted files.
Junio C Hamano [Thu, 11 Jan 2007 23:34:41 +0000 (15:34 -0800)]
git-status: wording update to deal with deleted files.

If you do:

$ /bin/rm foo
$ git status

we used to say "git add ... to add content to commit".  But
suggsting "git add" to record the deletion of a file is simply
insane.

So this rewords various things:

 - The section header is the old "Changed but not updated",
   instead of "Changed but not added";

 - Suggestion is "git add ... to update what will be committed",
   instead of "... to add content to commit";

 - If there are removed paths, the above suggestion becomes "git
   add/rm ... to update what will be committed";

 - For untracked files, the suggestion is "git add ... to
   include in what will be committed".

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-rm: do not fail on already removed file.
Junio C Hamano [Thu, 11 Jan 2007 22:58:47 +0000 (14:58 -0800)]
git-rm: do not fail on already removed file.

Often the user would do "/bin/rm foo" before telling git, but
then want to tell git about it.  "git rm foo" however would fail
because it cannot unlink(2) foo.

Treat ENOENT error return from unlink(2) as if a successful
removal happened.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoAvoid errors and warnings when attempting to do I/O on zero bytes
Eric Wong [Thu, 11 Jan 2007 21:43:40 +0000 (13:43 -0800)]
Avoid errors and warnings when attempting to do I/O on zero bytes

Unfortunately, while {read,write}_in_full do take into account
zero-sized reads/writes; their die and whine variants do not.

I have a repository where there are zero-sized files in
the history that was triggering these things.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoBetter error messages for corrupt databases
Linus Torvalds [Thu, 11 Jan 2007 22:09:31 +0000 (14:09 -0800)]
Better error messages for corrupt databases

This fixes another problem that Andy's case showed: git-fsck-objects
reports nonsensical results for corrupt objects.

There were actually two independent and confusing problems:

 - when we had a zero-sized file and used map_sha1_file, mmap() would
   return EINVAL, and git-fsck-objects would report that as an insane and
   confusing error. I don't know when this was introduced, it might have
   been there forever.

 - when "parse_object()" returned NULL, fsck would say "object not found",
   which can be very confusing, since obviously the object might "exist",
   it's just unparseable because it's totally corrupt.

So this just makes "xmmap()" return NULL for a zero-sized object (which is
a valid thing pointer, exactly the same way "malloc()" can return NULL for
a zero-sized allocation). That fixes the first problem (but we could have
fixed it in the caller too - I don't personally much care whichever way it
goes, but maybe somebody should check that the NO_MMAP case does
something sane in this case too?).

And the second problem is solved by just making the error message slightly
clearer - the failure to parse an object may be because it's missing or
corrupt, not necessarily because it's not "found".

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoconfig-set: check write-in-full returns in set_multivar
Junio C Hamano [Thu, 11 Jan 2007 21:16:26 +0000 (13:16 -0800)]
config-set: check write-in-full returns in set_multivar

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoindex-pack: write-or-die instead of unchecked write-in-full.
Junio C Hamano [Thu, 11 Jan 2007 21:15:51 +0000 (13:15 -0800)]
index-pack: write-or-die instead of unchecked write-in-full.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agowrite_in_full: really write in full or return error on disk full.
Linus Torvalds [Thu, 11 Jan 2007 21:04:11 +0000 (13:04 -0800)]
write_in_full: really write in full or return error on disk full.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDocument git-init
Junio C Hamano [Thu, 11 Jan 2007 20:58:10 +0000 (12:58 -0800)]
Document git-init

These days, the command does a lot more than just initialise the
object database (such as setting default config-variables,
installing template hooks...), and "git init" is actually a more
sensible name nowadays.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agowrite-cache: do not leak the serialized cache-tree data.
Linus Torvalds [Thu, 11 Jan 2007 20:25:16 +0000 (12:25 -0800)]
write-cache: do not leak the serialized cache-tree data.

It is not used after getting written, and just is leaking every time
we write the index out.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDisallow working directory commands in a bare repository.
Shawn O. Pearce [Sun, 31 Dec 2006 04:32:38 +0000 (23:32 -0500)]
Disallow working directory commands in a bare repository.

If the user tries to run a porcelainish command which requires
a working directory in a bare repository they may get unexpected
results which are difficult to predict and may differ from command
to command.

Instead we should detect that the current repository is a bare
repository and refuse to run the command there, as there is no
working directory associated with it.

[jc: updated Shawn's original somewhat -- bugs are mine.]

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agomerge-recursive: do not use on-file index when not needed.
Junio C Hamano [Wed, 10 Jan 2007 19:20:58 +0000 (11:20 -0800)]
merge-recursive: do not use on-file index when not needed.

This revamps the merge-recursive implementation following the
outline in:

Message-ID: <7v8xgileza.fsf@assigned-by-dhcp.cox.net>

There is no need to write out the index until the very end just
once from merge-recursive.  Also there is no need to write out
the resulting tree object for the simple case of merging with a
single merge base.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoSpeed-up recursive by flushing index only once for all entries
Alex Riesen [Thu, 4 Jan 2007 10:22:47 +0000 (11:22 +0100)]
Speed-up recursive by flushing index only once for all entries

The merge-recursive implementation in C inherited the invariant
that the on-file index file is written out and later read back
after any index operations and writing trees from the original
Python implementation.  But it was only because the original
implementation worked at the scripting level.

There is no need to write out the index file after handling
every path.

Signed-off-by: Alex Riesen <raa.lkml@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoProvide better feedback for the untracked only case in status output
Jürgen Rühle [Wed, 10 Jan 2007 22:25:03 +0000 (23:25 +0100)]
Provide better feedback for the untracked only case in status output

Since 98bf8a47c296f51ea9722fef4bb81dbfb70cd4bb status would claim that
git-commit could be useful even if there are no changes except untracked files.

Since wt-status is already computing all the information needed go the whole
way and actually track the (non-)emptiness of all three sections separately,
unify the code, and provide useful messages for each individual case.

Thanks to Junio and Michael Loeffler for suggestions.

Signed-off-by: Jürgen Rühle <j-r@online.de>
14 years agoMerge branch 'js/reflog'
Junio C Hamano [Wed, 10 Jan 2007 22:16:16 +0000 (14:16 -0800)]
Merge branch 'js/reflog'

* js/reflog:
  Sanitize for_each_reflog_ent()

14 years agoMakefile: remove $foo when $foo.exe is built/installed.
Junio C Hamano [Wed, 10 Jan 2007 20:24:54 +0000 (12:24 -0800)]
Makefile: remove $foo when $foo.exe is built/installed.

On Cygwin, newly builtins are not recognized, because there exist both
the executable binaries (with .exe extension) _and_ the now-obsolete
scripts (without extension), but the script is executed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agosend-email: work around double encoding of in-body From field.
Jürgen Rühle [Wed, 10 Jan 2007 21:36:39 +0000 (13:36 -0800)]
send-email: work around double encoding of in-body From field.

git-send-email sends out the message taken from format-patch
output without quoting nor encoding.  When copying the From:
line to form in-body From: field, it should not copy it
verbatim, because the From: for the header is quoted according
to RFC 2047 when not ASCII.

The original came from Jürgen Rühle, but I moved the
string munging into a separate function so that later other
people can tweak it more easily.  Bugs introduced during the
translation are mine.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoAdd git-init documentation.
Nicolas Pitre [Wed, 10 Jan 2007 17:56:41 +0000 (12:56 -0500)]
Add git-init documentation.

Oops. Commit 515377ea9ec6192f82a2fa5c5b5b7651d9d6cf6c missed one
file, git-init documentation.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoFix t1410 for core.filemode==false
Johannes Schindelin [Wed, 10 Jan 2007 12:22:50 +0000 (13:22 +0100)]
Fix t1410 for core.filemode==false

Since c869753e, core.filemode is hardwired to false on Cygwin.
So this test had no chance to succeed, since an early commit
(changing just the filemode) failed, and therefore all subsequent
tests.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMake git-describe a builtin.
Shawn O. Pearce [Wed, 10 Jan 2007 11:36:36 +0000 (06:36 -0500)]
Make git-describe a builtin.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDon't save the commit buffer in git-describe.
Shawn O. Pearce [Wed, 10 Jan 2007 11:36:29 +0000 (06:36 -0500)]
Don't save the commit buffer in git-describe.

The commit buffer (message of the commit) is not actually
used by the git-describe process.  We can save some memory
by not keeping it around.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoFix warnings in sha1_file.c - use C99 printf format if available
Pavel Roskin [Wed, 10 Jan 2007 04:07:11 +0000 (23:07 -0500)]
Fix warnings in sha1_file.c - use C99 printf format if available

Signed-off-by: Pavel Roskin <proski@gnu.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years ago-u is now default for 'git-mailinfo'.
Junio C Hamano [Wed, 10 Jan 2007 05:31:36 +0000 (21:31 -0800)]
-u is now default for 'git-mailinfo'.

Originally from David Woodhouse, but also adjusts the callers of
mailinfo to the new default.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years ago-u is now default for 'git-applymbox'
Junio C Hamano [Wed, 10 Jan 2007 05:20:01 +0000 (21:20 -0800)]
-u is now default for 'git-applymbox'

It has '-n' to disable it just in case, but do not even bother
documenting it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-am: should work when "--no-utf8 --utf8" is given
Junio C Hamano [Wed, 10 Jan 2007 05:16:45 +0000 (21:16 -0800)]
git-am: should work when "--no-utf8 --utf8" is given

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-checkout: handle local changes sanely when detaching HEAD
Junio C Hamano [Wed, 10 Jan 2007 04:39:09 +0000 (20:39 -0800)]
git-checkout: handle local changes sanely when detaching HEAD

When switching branches, we usually first try read-tree to make
sure that we do not lose the local changes and then updated the
HEAD using update-ref.  However, we detached and updated HEAD
before these checks, which was quite bad in a repository with
local changes.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDon't die in git-http-fetch when fetching packs.
Shawn O. Pearce [Wed, 10 Jan 2007 01:04:52 +0000 (20:04 -0500)]
Don't die in git-http-fetch when fetching packs.

My sp/mmap changes to pack-check.c modified the function such that
it expects packed_git.pack_size to be populated with the total
bytecount of the packfile by the caller.

But that isn't the case for packs obtained by git-http-fetch as
pack_size was not initialized before being accessed.  This caused
verify_pack to think it had 2^32-21 bytes available when the
downloaded pack perhaps was only 305 bytes in length.  The use_pack
function then later dies with "offset beyond end of packfile"
when computing the overall file checksum.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-checkout: safety check for detached HEAD checks existing refs
Junio C Hamano [Wed, 10 Jan 2007 01:37:50 +0000 (17:37 -0800)]
git-checkout: safety check for detached HEAD checks existing refs

Checking for reachability from refs does not help much if the
state we are currently on is somewhere in the middle.  We will
lose where we were.

So this makes sureh that HEAD is something directly pointed at
by one of the existing refs (most likely a tag for a user who
has been "sightseeing").

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoUpdate git-svn manpage to remove the implication that SVN::* is optional.
Steven Grimm [Wed, 10 Jan 2007 00:20:17 +0000 (16:20 -0800)]
Update git-svn manpage to remove the implication that SVN::* is optional.

Now that git-svn requires the SVN::* Perl library, the manpage doesn't need
to describe what happens when you don't have it.

Signed-off-by: Steven Grimm <koreth@midwinter.com>
Acked-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoReplacing the system call pread() with lseek()/xread()/lseek() sequence.
Stefan-W. Hahn [Tue, 9 Jan 2007 21:04:12 +0000 (22:04 +0100)]
Replacing the system call pread() with lseek()/xread()/lseek() sequence.

Using cygwin with cygwin.dll before 1.5.22 the system call pread() is buggy.
This patch introduces NO_PREAD. If NO_PREAD is set git uses a sequence of
lseek()/xread()/lseek() to emulate pread.

Signed-off-by: Stefan-W. Hahn <stefan.hahn@s-hahn.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogitweb: Fix git_patchset_body not closing <div class="patch">
Jakub Narebski [Tue, 9 Jan 2007 23:07:43 +0000 (00:07 +0100)]
gitweb: Fix git_patchset_body not closing <div class="patch">

Fix case when git_patchset_body didn't close <div class="patch">,
for patchsets with last patch empty.

This patch also removes some commented out code in git_patchset_body.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Acked-by: Luben Tuikov <ltuikov@yahoo.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit.el: Define the propertize function if needed, for XEmacs compatibility.
Alexandre Julliard [Tue, 9 Jan 2007 20:27:40 +0000 (21:27 +0100)]
git.el: Define the propertize function if needed, for XEmacs compatibility.

Also use `concat' instead of `format' in the pretty-printer since
format doesn't preserve properties under XEmacs.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-clone: Make sure the master branch exists before running cat on it.
Alexandre Julliard [Tue, 9 Jan 2007 20:26:52 +0000 (21:26 +0100)]
git-clone: Make sure the master branch exists before running cat on it.

Otherwise we get an error like this on stderr:

  cat: [...]/.git/refs/remotes/origin/master: No such file or directory

which makes it look like git-clone failed.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-apply: Remove directories that have become empty after deleting a file.
Alexandre Julliard [Tue, 9 Jan 2007 20:25:46 +0000 (21:25 +0100)]
git-apply: Remove directories that have become empty after deleting a file.

Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoget_tree_entry: map blank requested entry to tree root
Jeff King [Tue, 9 Jan 2007 16:11:47 +0000 (11:11 -0500)]
get_tree_entry: map blank requested entry to tree root

This means that
  git show HEAD:
will now return HEAD^{tree}, which is logically consistent with
  git show HEAD:Documentation

Signed-off-by: Jeff King <peff@peff.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agobuiltin-archive: do not free a tree held by the object layer.
Junio C Hamano [Tue, 9 Jan 2007 22:07:59 +0000 (14:07 -0800)]
builtin-archive: do not free a tree held by the object layer.

Found by running "git archive --format=tar HEAD" in Documentation/
directory.

It's surprising that nobody has noticed this from the beginning...

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMerge branch 'maint'
Junio C Hamano [Tue, 9 Jan 2007 20:04:30 +0000 (12:04 -0800)]
Merge branch 'maint'

* maint:
  Fix "Do not ignore a detected patchfile brokenness."
  Do not ignore a detected patchfile brokenness.

14 years agoFix "Do not ignore a detected patchfile brokenness."
Junio C Hamano [Tue, 9 Jan 2007 19:50:53 +0000 (11:50 -0800)]
Fix "Do not ignore a detected patchfile brokenness."

Returning negative value from there does not stop the caller from using
the earlier part.

Noticed by Linus.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoSanitize for_each_reflog_ent()
Johannes Schindelin [Mon, 8 Jan 2007 00:59:54 +0000 (01:59 +0100)]
Sanitize for_each_reflog_ent()

It used to ignore the return value of the helper function; now, it
expects it to return 0, and stops iteration upon non-zero return
values; this value is then passed on as the return value of
for_each_reflog_ent().

Further, it makes no sense to force the parsing upon the helper
functions; for_each_reflog_ent() now calls the helper function with
old and new sha1, the email, the timestamp & timezone, and the message.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDo not ignore a detected patchfile brokenness.
Junio C Hamano [Tue, 9 Jan 2007 10:52:31 +0000 (02:52 -0800)]
Do not ignore a detected patchfile brokenness.

find_header() function is used to read and parse the patchfile
and it detects errors in the patch, but one place ignored the
error and went ahead, which was quite bad.

Noticed by Jeff Garzik.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agomerge-base: do not leak commit list
Junio C Hamano [Tue, 9 Jan 2007 07:10:49 +0000 (23:10 -0800)]
merge-base: do not leak commit list

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoAuto-quote config values in config.c:store_write_pair()
Brian Gernhardt [Tue, 9 Jan 2007 05:27:41 +0000 (00:27 -0500)]
Auto-quote config values in config.c:store_write_pair()

Suggested by Jakub Narebski <jnareb@gmail.com> on the list.

When we send a value to store_write_pair(), make sure that the value
that gets read out matches the one passed in.  This means that for any
value that contains leading or trailing whitespace or any comment
character (# and ;), we need to surround it in quotes.

Signed-off-by: Brian Gernhardt <benji@silverinsanity.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoIgnore git-init and git-remote
Brian Gernhardt [Tue, 9 Jan 2007 05:27:33 +0000 (00:27 -0500)]
Ignore git-init and git-remote

These new commands weren't added to .gitignore.  Add them so we don't
end up with copies of them in the repo.

Signed-off-by: Brian Gernhardt <benji@silverinsanity.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agorm git-rerere.perl -- it is now a built-in.
Junio C Hamano [Tue, 9 Jan 2007 05:52:38 +0000 (21:52 -0800)]
rm git-rerere.perl -- it is now a built-in.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agocvsserver: fix revision number during file adds
Martin Langhoff [Tue, 9 Jan 2007 02:10:41 +0000 (15:10 +1300)]
cvsserver: fix revision number during file adds

With this patch, cvs add / cvs commit echoes back to the client
the correct file version (1.1) so that the file in the checkout
is recognised as up-to-date.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agocvsserver: detect early of we are up to date and avoid costly rev-list
Martin Langhoff [Tue, 9 Jan 2007 02:10:32 +0000 (15:10 +1300)]
cvsserver: detect early of we are up to date and avoid costly rev-list

if the SHA1 of our head matches the last SHA1 seen in the DB, avoid further
processing.

[jc: an "Oops, please amend" patch rolled in]

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDocumentation: add git-remote man page
J. Bruce Fields [Tue, 9 Jan 2007 04:28:51 +0000 (23:28 -0500)]
Documentation: add git-remote man page

Add a preliminary man page for git-remote.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMerge branch 'jc/reflog'
Junio C Hamano [Mon, 8 Jan 2007 23:56:51 +0000 (15:56 -0800)]
Merge branch 'jc/reflog'

* jc/reflog:
  reflog --fix-stale: do not check the same trees and commits repeatedly.
  reflog expire --fix-stale
  Move traversal of reachable objects into a separate library.
  builtin-prune: separate ref walking from reflog walking.
  builtin-prune: make file-scope static struct to an argument.

14 years agoshort i/o: fix config updates to use write_in_full
Andy Whitcroft [Mon, 8 Jan 2007 15:58:38 +0000 (15:58 +0000)]
short i/o: fix config updates to use write_in_full

We need to check that the writes we perform during the update of
the users configuration work.  Convert to using write_in_full().

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoshort i/o: fix calls to write to use xwrite or write_in_full
Andy Whitcroft [Mon, 8 Jan 2007 15:58:23 +0000 (15:58 +0000)]
short i/o: fix calls to write to use xwrite or write_in_full

We have a number of badly checked write() calls.  Often we are
expecting write() to write exactly the size we requested or fail,
this fails to handle interrupts or short writes.  Switch to using
the new write_in_full().  Otherwise we at a minimum need to check
for EINTR and EAGAIN, where this is appropriate use xwrite().

Note, the changes to config handling are much larger and handled
in the next patch in the sequence.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoshort i/o: fix calls to read to use xread or read_in_full
Andy Whitcroft [Mon, 8 Jan 2007 15:58:08 +0000 (15:58 +0000)]
short i/o: fix calls to read to use xread or read_in_full

We have a number of badly checked read() calls.  Often we are
expecting read() to read exactly the size we requested or fail, this
fails to handle interrupts or short reads.  Add a read_in_full()
providing those semantics.  Otherwise we at a minimum need to check
for EINTR and EAGAIN, where this is appropriate use xread().

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoshort i/o: clean up the naming for the write_{in,or}_xxx family
Andy Whitcroft [Mon, 8 Jan 2007 15:57:52 +0000 (15:57 +0000)]
short i/o: clean up the naming for the write_{in,or}_xxx family

We recently introduced a write_in_full() which would either write
the specified object or emit an error message and fail.  In order
to fix the read side we now want to introduce a read_in_full()
but without an error emit.  This patch cleans up the naming
of this family of calls:

1) convert the existing write_or_whine() to write_or_whine_pipe()
   to better indicate its pipe specific nature,
2) convert the existing write_in_full() calls to write_or_whine()
   to better indicate its nature,
3) introduce a write_in_full() providing a write or fail semantic,
   and
4) convert write_or_whine() and write_or_whine_pipe() to use
   write_in_full().

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years ago--prune is now default for 'pack-refs'
Junio C Hamano [Mon, 8 Jan 2007 22:40:33 +0000 (14:40 -0800)]
--prune is now default for 'pack-refs'

There is no reason not to, really.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years ago--utf8 is now default for 'git-am'
Junio C Hamano [Mon, 8 Jan 2007 21:59:15 +0000 (13:59 -0800)]
--utf8 is now default for 'git-am'

Since we are talking about allowing potentially incompatible UI
changes in v1.5.0 iff the change improves the general situation,
I would say why not.

There is --no-utf8 flag to avoid re-coding from botching the log
message just in case, but we may not even need it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-commit: do not fail to print the diffstat even if there is a file named HEAD
Michael Loeffler [Mon, 8 Jan 2007 19:23:13 +0000 (20:23 +0100)]
git-commit: do not fail to print the diffstat even if there is a file named HEAD

Signed-off-by: Michael Loeffler <zvpunry@zvpunry.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agossh-upload: prevent buffer overrun
Andy Whitcroft [Mon, 8 Jan 2007 11:45:44 +0000 (11:45 +0000)]
ssh-upload: prevent buffer overrun

Prevent a client from overrunning the on stack ref buffer.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-checkout: fix branch name output from the command
Junio C Hamano [Mon, 8 Jan 2007 10:44:55 +0000 (02:44 -0800)]
git-checkout: fix branch name output from the command

When switching branches with "git checkout", we internally did $arg^0
(aka $arg^{commit}) suffix but there was no need to.

The improvement is easily visible in the change to an existing
test t/3200-branch.sh in this commit; it was expecting rather
ugly message.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-checkout: safety when coming back from the detached HEAD state.
Junio C Hamano [Mon, 8 Jan 2007 10:42:30 +0000 (02:42 -0800)]
git-checkout: safety when coming back from the detached HEAD state.

After making commits in the detached HEAD state, if you run "git
checkout" to switch to an existing branch, you will lose your
work.  Make sure the switched-to branch is a fast-forward of the
current HEAD, or require -f when switching.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-checkout: rewording comments regarding detached HEAD.
Junio C Hamano [Mon, 8 Jan 2007 10:08:47 +0000 (02:08 -0800)]
git-checkout: rewording comments regarding detached HEAD.

We used to say "you are not on a branch" before the initial
commit.  This is incorrect -- the user is on a branch yet to be
born, but its name has been already determined.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-checkout: do not warn detaching HEAD when it is already detached.
Junio C Hamano [Mon, 8 Jan 2007 10:19:38 +0000 (02:19 -0800)]
git-checkout: do not warn detaching HEAD when it is already detached.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDetached HEAD (experimental)
Junio C Hamano [Tue, 2 Jan 2007 07:31:08 +0000 (23:31 -0800)]
Detached HEAD (experimental)

This allows "git checkout v1.4.3" to dissociate the HEAD of
repository from any branch.  After this point, "git branch"
starts reporting that you are not on any branch.  You can go
back to an existing branch by saying "git checkout master", for
example.

This is still experimental.  While I think it makes sense to
allow commits on top of detached HEAD, it is rather dangerous
unless you are careful in the current form.  Next "git checkout
master" will obviously lose what you have done, so we might want
to require "git checkout -f" out of a detached HEAD if we find
that the HEAD commit is not an ancestor of any other branches.
There is no such safety valve implemented right now.

On the other hand, the reason the user did not start the ad-hoc
work on a new branch with "git checkout -b" was probably because
the work was of a throw-away nature, so the convenience of not
having that safety valve might be even better.  The user, after
accumulating some commits on top of a detached HEAD, can always
create a new branch with "git checkout -b" not to lose useful
work done while the HEAD was detached.

We'll see.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-branch: show detached HEAD
Lars Hjemli [Wed, 3 Jan 2007 20:10:10 +0000 (21:10 +0100)]
git-branch: show detached HEAD

This makes git-branch show a detached HEAD as '* (no branch)'.

Signed-off-by: Lars Hjemli <hjemli@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-status: show detached HEAD
Junio C Hamano [Wed, 3 Jan 2007 09:09:34 +0000 (01:09 -0800)]
git-status: show detached HEAD

This makes git-status to state when you are not on any branch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agocvsimport: cleanup temporary cvsps file
Martin Langhoff [Mon, 8 Jan 2007 08:08:46 +0000 (21:08 +1300)]
cvsimport: cleanup temporary cvsps file

It is bad manners to leave these sizable files
around when we are done with them.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agocvsimport: document -S and -L options
Martin Langhoff [Mon, 8 Jan 2007 06:53:03 +0000 (19:53 +1300)]
cvsimport: document -S and -L options

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agocvsimport: skip commits that are too recent (option and documentation)
Martin Langhoff [Mon, 8 Jan 2007 06:43:39 +0000 (19:43 +1300)]
cvsimport: skip commits that are too recent (option and documentation)

This makes the earlier "wait for 10 minutes before importing" safety
overridable with "-a(ll)" flag, and adds necessary documentation.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-fetch: allow updating the current branch in a bare repository.
Junio C Hamano [Sun, 7 Jan 2007 10:17:52 +0000 (02:17 -0800)]
git-fetch: allow updating the current branch in a bare repository.

Sometimes, people have only fetch access into a bare repository
that is used as a back-up location (or a distribution point) but
does not have a push access for networking reasons, e.g. one end
being behind a firewall, and updating the "current branch" in
such a case is perfectly fine.

This allows such a fetch without --update-head-ok, which is a
flag that should never be used by end users otherwise.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoIntroduce is_bare_repository() and core.bare configuration variable
Junio C Hamano [Sun, 7 Jan 2007 10:00:28 +0000 (02:00 -0800)]
Introduce is_bare_repository() and core.bare configuration variable

This removes the old is_bare_git_dir(const char *) to ask if a
directory, if it is a GIT_DIR, is a bare repository, and
replaces it with is_bare_repository(void *).  The function looks
at core.bare configuration variable if exists but uses the old
heuristics: if it is ".git" or ends with "/.git", then it does
not look like a bare repository, otherwise it does.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMove initialization of log_all_ref_updates
Junio C Hamano [Sun, 7 Jan 2007 09:35:34 +0000 (01:35 -0800)]
Move initialization of log_all_ref_updates

The patches to prevent Porcelainish that require working tree
from doing any damage in a bare repository make a lot of sense,
and I want to make the is_bare_git_dir() function more reliable.

In order to allow the repository owner override the heuristic
implemented in is_bare_git_dir() if/when it misidentifies a
particular repository, it would make sense to introduce a new
configuration variable "[core] bare = true/false", and make
is_bare_git_dir() notice it.

The scripts would do a 'repo-config --bool --get core.bare' and
iff the command fails (i.e. there is no such variable in the
configuration file), it would use the heuristic implemented at
the script level [*1*].

However, setup_git_env() which is called a lot earlier than we
even read from the repository configuration currently makes a
call to is_bare_git_dir(), in order to change the default
setting for log_all_ref_updates.  It somehow feels that this is
a hack.

By the way, [*1*] is another thing I hate about the current
config mechanism.  "git-repo-config --get" does not know what
the possible configuration variables are, let alone what the
default values for them are.  It allows us not to maintain a
centralized configuration table, which makes it easy to
introduce ad-hoc variables and gives a warm fuzzy feeling of
being modular, but my feeling is that it is turning out to be a
rather high price to pay for scripts.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-svn: pass an unambiguous ref to rev-list when grafting-branches
Eric Wong [Mon, 8 Jan 2007 03:35:41 +0000 (19:35 -0800)]
git-svn: pass an unambiguous ref to rev-list when grafting-branches

Some users apparently create local heads with the same basename
as the remote branch they're tracking.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-svn: add --prefix= option to multi-init
Eric Wong [Mon, 8 Jan 2007 03:35:40 +0000 (19:35 -0800)]
git-svn: add --prefix= option to multi-init

Also, document --{trunk,branches,tags} options while we're
documenting multi-init options.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoDocumentation: clarify definition of "reachable"
J. Bruce Fields [Mon, 8 Jan 2007 00:25:35 +0000 (19:25 -0500)]
Documentation: clarify definition of "reachable"

Clarify definition of "reachable" (what chain?)

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogit-svnimport: fix edge revisions double importing
Sasha Khapyorsky [Mon, 8 Jan 2007 02:22:42 +0000 (04:22 +0200)]
git-svnimport: fix edge revisions double importing

This fixes newly introduced bug when the incremental cycle edge revisions
are imported twice.

Signed-off-by: Sasha Khapyorsky <sashak@voltaire.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agocvsimport: skip commits that are too recent
Martin Langhoff [Mon, 8 Jan 2007 01:11:23 +0000 (14:11 +1300)]
cvsimport: skip commits that are too recent

With this patch, cvsimport will skip commits made
in the last 10 minutes. The recent-ness test is of
5 minutes + cvsps fuzz window (5 minutes default).

When working with a CVS repository that is in use,
importing commits that are too recent can lead to
partially incorrect trees. This is mainly due to

 - Commits that are within the cvsps fuzz window may later
   be found to have affected more files.

 - When performing incremental imports, clock drift between
   the systems may lead to skipped commits.

This commit helps keep incremental imports of in-use
CVS repositories sane.

Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agogitweb: Remove superfluous "|" in "commit" view
Jakub Narebski [Mon, 8 Jan 2007 01:10:42 +0000 (02:10 +0100)]
gitweb: Remove superfluous "|" in "commit" view

Remove superfluous trailing "|" separator from difftree part of "commit"
view for new files (created in given commit).

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoRemove unnecessary git-rm --cached reference from status output
Jürgen Rühle [Sun, 7 Jan 2007 19:18:26 +0000 (20:18 +0100)]
Remove unnecessary git-rm --cached reference from status output

Since git-reset has learned restoring the absence of paths git-rm --cached is
no longer necessary. Therefore remove it from the cached content header hint.

Also remove the unfortunate wording 'Cached' from the header itself.

Signed-off-by: Jürgen Rühle <j-r@online.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years ago"init-db" can really be just "init"
Nicolas Pitre [Sun, 7 Jan 2007 17:31:29 +0000 (12:31 -0500)]
"init-db" can really be just "init"

Make "init" the equivalent of "init-db". This should make first GIT
impression a little more friendly.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMerge branch 'sp/mmap'
Junio C Hamano [Sun, 7 Jan 2007 08:12:47 +0000 (00:12 -0800)]
Merge branch 'sp/mmap'

* sp/mmap: (27 commits)
  Spell default packedgitlimit slightly differently
  Increase packedGit{Limit,WindowSize} on 64 bit systems.
  Update packedGit config option documentation.
  mmap: set FD_CLOEXEC for file descriptors we keep open for mmap()
  pack-objects: fix use of use_pack().
  Fix random segfaults in pack-objects.
  Cleanup read_cache_from error handling.
  Replace mmap with xmmap, better handling MAP_FAILED.
  Release pack windows before reporting out of memory.
  Default core.packdGitWindowSize to 1 MiB if NO_MMAP.
  Test suite for sliding window mmap implementation.
  Create pack_report() as a debugging aid.
  Support unmapping windows on 'temporary' packfiles.
  Improve error message when packfile mmap fails.
  Ensure core.packedGitWindowSize cannot be less than 2 pages.
  Load core configuration in git-verify-pack.
  Fully activate the sliding window pack access.
  Unmap individual windows rather than entire files.
  Document why header parsing won't exceed a window.
  Loop over pack_windows when inflating/accessing data.
  ...

Conflicts:

cache.h
pack-check.c

14 years agoSpell default packedgitlimit slightly differently
Junio C Hamano [Sun, 7 Jan 2007 08:11:11 +0000 (00:11 -0800)]
Spell default packedgitlimit slightly differently

This is shorter and easier to read, and also makes sure the
constant expression does not overflow integer range.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMerge branch 'jr/status'
Junio C Hamano [Sun, 7 Jan 2007 07:49:24 +0000 (23:49 -0800)]
Merge branch 'jr/status'

* jr/status:
  Improve cached content header of status output
  Support --amend on initial commit in status output
  Improve "nothing to commit" part of status output
  Clarify syntax and role of git-add in status output

14 years agoMerge branch 'jc/remote'
Junio C Hamano [Sun, 7 Jan 2007 07:49:16 +0000 (23:49 -0800)]
Merge branch 'jc/remote'

* jc/remote:
  git-remote

14 years agogit-reset <tree> -- <path> restores absense of <path> in <tree>
Junio C Hamano [Fri, 5 Jan 2007 09:38:56 +0000 (01:38 -0800)]
git-reset <tree> -- <path> restores absense of <path> in <tree>

When <path> exists in the index (either merged or unmerged), and
<tree> does not have it, git-reset should be usable to restore
the absense of it from the tree.  This implements it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agodiff-index --cached --raw: show tree entry on the LHS for unmerged entries.
Junio C Hamano [Fri, 5 Jan 2007 09:25:18 +0000 (01:25 -0800)]
diff-index --cached --raw: show tree entry on the LHS for unmerged entries.

This updates the way diffcore represents an unmerged pair
somewhat.  It used to be that entries with mode=0 on both sides
were used to represent an unmerged pair, but now it has an
explicit flag.  This is to allow diff-index --cached to report
the entry from the tree when the path is unmerged in the index.

This is used in updating "git reset <tree> -- <path>" to restore
absense of the path in the index from the tree.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoreflog --fix-stale: do not check the same trees and commits repeatedly.
Junio C Hamano [Sun, 7 Jan 2007 06:32:41 +0000 (22:32 -0800)]
reflog --fix-stale: do not check the same trees and commits repeatedly.

Since we use the reachability tracking machinery now, we should
keep the already checked trees and commits whose completeness is
known, to avoid checking the same thing over and over again.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoreflog expire --fix-stale
Junio C Hamano [Sat, 6 Jan 2007 10:16:19 +0000 (02:16 -0800)]
reflog expire --fix-stale

The logic in an earlier round to detect reflog entries that
point at a broken commit was not sufficient.  Just like we do
not trust presense of a commit during pack transfer (we trust
only our refs), we should not trust a commit's presense, even if
the tree of that commit is complete.

A repository that had reflog enabled on some of the refs that
was rewound and then run git-repack or git-prune from older
versions of git can have reflog entries that point at a commit
that still exist but lack commits (or trees and blobs needed for
that commit) between it and some commit that is reachable from
one of the refs.

This revamps the logic -- the definition of "broken commit"
becomes: a commit that is not reachable from any of the refs and
there is a missing object among the commit, tree, or blob
objects reachable from it that is not reachable from any of the
refs.  Entries in the reflog that refer to such a commit are
expired.

Since this computation involves traversing all the reachable
objects, i.e. it has the same cost as 'git prune', it is enabled
only when a new option --fix-stale.  Fortunately, once this is
run, we should not have to ever worry about missing objects,
because the current prune and pack-objects know about reflogs
and protect objects referred by them.

Unfortunately, this will be absolutely necessary to help people
migrate to the newer prune and repack.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agoMove traversal of reachable objects into a separate library.
Junio C Hamano [Sat, 6 Jan 2007 10:16:17 +0000 (02:16 -0800)]
Move traversal of reachable objects into a separate library.

This moves major part of builtin-prune into a separate file,
reachable.c.  It is used to mark the objects that are reachable
from refs, and optionally from reflogs.

The patch looks very large, but if you look at it with diff -C,
which this message is formatted in, most of them are copied
lines and there are very little additions.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agobuiltin-prune: separate ref walking from reflog walking.
Junio C Hamano [Sat, 6 Jan 2007 10:16:14 +0000 (02:16 -0800)]
builtin-prune: separate ref walking from reflog walking.

This is necessary for the next step, because the reason I am
making the connectivity walker into a library is because I want
to use it for cleaning up stale reflog entries.

Signed-off-by: Junio C Hamano <junkio@cox.net>
14 years agobuiltin-prune: make file-scope static struct to an argument.
Junio C Hamano [Sat, 6 Jan 2007 10:16:10 +0000 (02:16 -0800)]
builtin-prune: make file-scope static struct to an argument.

I want to make the first part of 'git prune' that marks the
reachable objects callable as a library, so this starts the
first step toward the goal by making the callchain to pass
rev_info structure as an argument.

No functionality change should be in this step.

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