Code

git.git
17 years agobuiltin-prune: memory diet.
Junio C Hamano [Fri, 5 Jan 2007 21:31:43 +0000 (13:31 -0800)]
builtin-prune: memory diet.

Somehow we forgot to turn save_commit_buffer off while walking
the reachable objects.  Releasing the memory for commit object
data that we do not use matters for large projects (for example,
about 90MB is saved while traversing linux-2.6 history).

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoBlame "linenr" link jumps to previous state at "orig_lineno"
Luben Tuikov [Fri, 5 Jan 2007 02:37:45 +0000 (18:37 -0800)]
Blame "linenr" link jumps to previous state at "orig_lineno"

Blame currently displays the commit id which introduced a
block of one or more lines, the line numbers wrt the current
listing of the file and the file's line contents.

The commit id displayed is hyperlinked to the commit.

Currently the linenr links are hyperlinked to the same
commit id displayed to the left, which is _no_ different
than the block of lines displayed, since it is the _same
commit_ that is hyperlinked.  And thus clicking on it leads
to the same state of the file for that chunk of
lines. I.e. data mining is not currently possible with
gitweb given a chunk of lines introduced by a commit.

This patch makes such data mining possible.

The line numbers are now hyperlinked to the parent of the
commit id of the block of lines.  Furthermore they are
linked to the line where that block was introduced.

Thus clicking on a linenr link will show you the file's
line(s) state prior to the commit id you were viewing.

So clicking continually on a linenr link shows you how this
line and its line number changed over time, leading to the
initial commit where it was first introduced.

Signed-off-by: Luben Tuikov <ltuikov@yahoo.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Fix "Use of uninitialized value" warning in git_tags_body
Jakub Narebski [Wed, 3 Jan 2007 21:54:29 +0000 (22:54 +0100)]
gitweb: Fix "Use of uninitialized value" warning in git_tags_body

Fix "Use of uninitialized value" warning in git_tags_body generated
for lightweight tags of tree and blob object; those don't have age
($tag{'age'}) defined.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: make --repack work consistently between fetch and multi-fetch
Eric Wong [Fri, 5 Jan 2007 02:09:56 +0000 (18:09 -0800)]
git-svn: make --repack work consistently between fetch and multi-fetch

Since fetch reforks itself at most every 1000 revisions, we
need to update the counter in the parent process to have a
working count if we set our repack interval to be > ~1000
revisions.  multi-fetch has always done this correctly
because of an extra process; now fetch uses the extra process;
as well.

While we're at it, only compile the $sha1 regex that checks for
repacking once.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: update documentation for multi-{init|fetch}
Eric Wong [Fri, 5 Jan 2007 02:04:06 +0000 (18:04 -0800)]
git-svn: update documentation for multi-{init|fetch}

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: make multi-init less confusing
Eric Wong [Fri, 5 Jan 2007 02:02:00 +0000 (18:02 -0800)]
git-svn: make multi-init less confusing

It now requires at least one of the (trunk|branch|tags) arguments
(either from the command-line or in .git/config).  Also we make
sure that anything that is passed as a URL ('help') in David's
case is actually a URL.

Thanks to David Kågedal for reporting this issue.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRemove shadowing variable from traverse_trees()
René Scharfe [Thu, 4 Jan 2007 18:33:36 +0000 (19:33 +0100)]
Remove shadowing variable from traverse_trees()

The variable named entry is allocated using malloc() and then
forgotten, it being shadowed by an automatic variable of the
same name.  Fixing the array size at 3 worked so far because
the only caller of traverse_trees() needed only as much
entries.  Simply remove the shadowing varaible and we're able
to traverse more than three trees and save stack space at the
same time!

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMake check target depend on common-cmds.h
René Scharfe [Thu, 4 Jan 2007 18:33:48 +0000 (19:33 +0100)]
Make check target depend on common-cmds.h

This fixes sparse complaining about a missing include file
if 'make check' is run on clean sources.

Signed-off-by: Rene Scharfe <rene.scharfe@lsrfire.ath.cx>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agorerere: Fix removal of already resolved path.
Junio C Hamano [Thu, 4 Jan 2007 08:06:50 +0000 (00:06 -0800)]
rerere: Fix removal of already resolved path.

There was an obvious thinko in memmove() to remove an entry that
was resolved from the in-core data structure.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'maint'
Junio C Hamano [Fri, 5 Jan 2007 06:28:21 +0000 (22:28 -0800)]
Merge branch 'maint'

* maint:
  pack-check.c::verify_packfile(): don't run SHA-1 update on huge data
  Fix infinite loop when deleting multiple packed refs.

17 years agopack-check.c::verify_packfile(): don't run SHA-1 update on huge data v1.4.4.4
Junio C Hamano [Thu, 4 Jan 2007 07:09:08 +0000 (23:09 -0800)]
pack-check.c::verify_packfile(): don't run SHA-1 update on huge data

Running the SHA1_Update() on the whole packfile in a single call
revealed an overflow problem we had in the SHA-1 implementation
on POWER architecture some time ago, which was fixed with commit
b47f509b (June 19, 2006).  Other SHA-1 implementations may have
a similar problem.

The sliding mmap() series already makes chunked calls to
SHA1_Update(), so this patch itself will become moot when it
graduates to "master", but in the meantime, run the hash
function in smaller chunks to prevent possible future problems.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Fix shortlog only showing HEAD revision.
Robert Fitzsimons [Wed, 3 Jan 2007 20:25:55 +0000 (20:25 +0000)]
gitweb: Fix shortlog only showing HEAD revision.

My change in 190d7fdcf325bb444fa806f09ebbb403a4ae4ee6 had a small bug
found by Michael Krufky which caused the passed in hash value to be
ignored, so shortlog would only show the HEAD revision.

Signed-off-by: Robert Fitzsimons <robfitz@273k.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-verify-tag: make sure we remove temporary file.
Junio C Hamano [Wed, 3 Jan 2007 20:34:20 +0000 (12:34 -0800)]
git-verify-tag: make sure we remove temporary file.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-tag: add flag to verify a tag
Santi Béjar [Wed, 3 Jan 2007 12:59:00 +0000 (13:59 +0100)]
git-tag: add flag to verify a tag

This way "git tag -v $tag" is the UI for git-verify-tag.

Signed-off-by: Santi Béjar <sbejar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRefactor print-functions in builtin-branch
Lars Hjemli [Wed, 3 Jan 2007 20:10:09 +0000 (21:10 +0100)]
Refactor print-functions in builtin-branch

This moves the guts of print_ref_list() into a revamped print_ref_info(),
which at the same time gets renamed to print_ref_item().

Signed-off-by: Lars Hjemli <hjemli@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: There can be empty patches (in git_patchset_body)
Jakub Narebski [Wed, 3 Jan 2007 19:47:25 +0000 (20:47 +0100)]
gitweb: There can be empty patches (in git_patchset_body)

We now do not skip over empty patches in git_patchset_body
(where empty means that they consist only of git diff header,
and of extended diff header), so uncomment branch of code dealing
with empty patches (patches which do not have even two-line
from/to header)

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Fix bug in git_difftree_body (was '!=' instead of 'ne')
Jakub Narebski [Wed, 3 Jan 2007 19:47:24 +0000 (20:47 +0100)]
gitweb: Fix bug in git_difftree_body (was '!=' instead of 'ne')

Fix bug in git_difftree_body subroutine; it was used '!=' comparison
operator for strings (file type) instead of correct 'ne'.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation/tutorial: misc updates
Santi Béjar [Wed, 3 Jan 2007 12:53:27 +0000 (13:53 +0100)]
Documentation/tutorial: misc updates

 - Teach how to delete a branch with "git branch -d name".
 - Usually a commit has one parent; merge has more.
 - Teach "git show" instead of "git cat-file -p".

Signed-off-by: Santi Béjar <sbejar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agotutorial: misc updates.
Junio C Hamano [Wed, 3 Jan 2007 16:38:01 +0000 (08:38 -0800)]
tutorial: misc updates.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoSkip excessive blank lines before commit body
Lars Hjemli [Wed, 3 Jan 2007 13:34:13 +0000 (14:34 +0100)]
Skip excessive blank lines before commit body

This modifies pretty_print_commit() to make the output of git-rev-list and
friends a bit more predictable.

A commit body starting with blank lines might be unheard-of, but still possible
to create using git-commit-tree (so is bound to appear somewhere, sometime).

Signed-off-by: Lars Hjemli <hjemli@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAdd documentation for git-branch's color configuration.
Brian Gernhardt [Wed, 3 Jan 2007 15:36:29 +0000 (10:36 -0500)]
Add documentation for git-branch's color configuration.

Added color.branch and color.branch.<slot> to configuration list.
Style copied from color.status and meanings derived from the code.

Moved the color meanings from color.diff.<slot> to color.branch.<slot>
since the latter comes first alphabetically.

Added --color and --no-color to git-branch's usage and documentation.

Signed-off-by: Brian Gernhardt <benji@silverinsanity.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogitweb: Fix error in git_project_index subroutine
Jakub Narebski [Wed, 3 Jan 2007 15:03:01 +0000 (16:03 +0100)]
gitweb: Fix error in git_project_index subroutine

Instead of "$projectroot/$pr->{'path'}" to get the path to project
GIT_DIR, it was used "$projectroot/$project" which is valid only
for actions where project parameter is set, and 'project_index' is not
one of them.

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix infinite loop when deleting multiple packed refs.
Junio C Hamano [Tue, 2 Jan 2007 19:19:05 +0000 (11:19 -0800)]
Fix infinite loop when deleting multiple packed refs.

It was stupid to link the same element twice to lock_file_list
and end up in a loop, so we certainly need a fix.

But it is not like we are taking a lock on multiple files in
this case.  It is just that we leave the linked element on the
list even after commit_lock_file() successfully removes the
cruft.

We cannot remove the list element in commit_lock_file(); if we
are interrupted in the middle of list manipulation, the call to
remove_lock_file_on_signal() will happen with a broken list
structure pointed by lock_file_list, which would cause the cruft
to remain, so not removing the list element is the right thing
to do.  Instead we should be reusing the element already on the
list.

There is already a code for that in lock_file() function in
lockfile.c.  The code checks lk->next and the element is linked
only when it is not already on the list -- which is incorrect
for the last element on the list (which has NULL in its next
field), but if you read the check as "is this element already on
the list?" it actually makes sense.  We do not want to link it
on the list again, nor we would want to set up signal/atexit
over and over.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agosend pack check for failure to send revisions list
Andy Whitcroft [Tue, 2 Jan 2007 14:12:09 +0000 (14:12 +0000)]
send pack check for failure to send revisions list

When passing the revisions list to pack-objects we do not check for
errors nor short writes.  Introduce a new write_in_full which will
handle short writes and report errors to the caller.  Use this to
short cut the send on failure, allowing us to wait for and report
the child in case the failure is its fault.

Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoinstaweb: load Apache mime and dir modules if they are needed
Eric Wong [Tue, 2 Jan 2007 08:57:11 +0000 (00:57 -0800)]
instaweb: load Apache mime and dir modules if they are needed

I've noticed that Apache 2.2 on a Debian etch machine has
these compiled as modules.

Also set ServerName to avoid a warning at startup.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agofetch-pack: do not use lockfile structure on stack.
Junio C Hamano [Tue, 2 Jan 2007 19:22:08 +0000 (11:22 -0800)]
fetch-pack: do not use lockfile structure on stack.

They are used in atexit() for clean-up, and you will be
accessing unallocated memory at that point.

See 31f584c2 for the fix for a similar problem.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRemove unused variable (git-commit.sh)
Junio C Hamano [Tue, 2 Jan 2007 07:42:53 +0000 (23:42 -0800)]
Remove unused variable (git-commit.sh)

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUpdate clone/fetch documentation with --depth (shallow clone) option
Junio C Hamano [Mon, 1 Jan 2007 23:08:06 +0000 (15:08 -0800)]
Update clone/fetch documentation with --depth (shallow clone) option

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoStrongly discourage --update-head-ok in fetch-options documentation.
Junio C Hamano [Mon, 1 Jan 2007 23:07:35 +0000 (15:07 -0800)]
Strongly discourage --update-head-ok in fetch-options documentation.

"Use it with care" is a wrong wording to say "this is purely internal
and you are supposed to know what you are doing if you use this".

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'sp/merge' (early part)
Junio C Hamano [Mon, 1 Jan 2007 22:40:37 +0000 (14:40 -0800)]
Merge branch 'sp/merge' (early part)

* 'sp/merge' (early part):
  Use merge-recursive in git-am -3.
  Allow merging bare trees in merge-recursive.
  Move better_branch_name above get_ref in merge-recursive.

17 years agoDocumentation: remove master:origin example from pull-fetch-param.txt
J. Bruce Fields [Mon, 1 Jan 2007 21:40:23 +0000 (16:40 -0500)]
Documentation: remove master:origin example from pull-fetch-param.txt

This is no longer a useful example.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: update git-pull.txt for new clone behavior
J. Bruce Fields [Mon, 1 Jan 2007 21:39:06 +0000 (16:39 -0500)]
Documentation: update git-pull.txt for new clone behavior

Update examples, stop using branch named "origin" as an example.
Remove large example of use of remotes; that particular case is
nicely automated by default, so it's not so pressing to explain, and
we can refer to git-repo-config for the details.

Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-fetch: remove .keep file at the end.
Junio C Hamano [Mon, 1 Jan 2007 20:24:15 +0000 (12:24 -0800)]
git-fetch: remove .keep file at the end.

Removal of them is needed regardless of errors.  The original
code had the removal outside of the process which sets the flag
to tell the later step what to remove, but it runs as a
downstream of a pipeline and its effect was lost.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agofail pull/merge early in the middle of conflicted merge
Junio C Hamano [Mon, 1 Jan 2007 07:21:50 +0000 (23:21 -0800)]
fail pull/merge early in the middle of conflicted merge

After a pull that results in a conflicted merge, a new user
often tries another "git pull" in desperation.  When the index
is unmerged, merge backends correctly bail out without touching
either index nor the working tree, so this does not make the
wound any worse.

The user will however see several lines of messsages during this
process, such as "filename: needs merge", "you need to resolve
your current index first", "Merging...", and "Entry ... would be
overwritten by merge. Cannot merge.".  They are unnecessarily
alarming, and cause useful conflict messages from the first pull
scroll off the top of the terminal.

This changes pull and merge to run "git-ls-files -u" upfront and
stop them much earlier than we currently do.  Old timers may
know better and would not to try pulling again before cleaning
things up; this change adds extra overhead that is unnecessary
for them.  But this would be worth paying for to save new people
from needless confusion.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUpdate send-pack pipeline documentation.
Junio C Hamano [Mon, 1 Jan 2007 20:07:26 +0000 (12:07 -0800)]
Update send-pack pipeline documentation.

The pipeline was much more complex and needed documentation, but
now it is trivial and straightforward.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: t/t91??-*: optimize the tests a bit
Eric Wong [Mon, 1 Jan 2007 05:49:47 +0000 (21:49 -0800)]
git-svn: t/t91??-*: optimize the tests a bit

This removes some unnecessary 'svn up' calls throughout

t9103-git-svn-graft-branches.sh:
  * removed an 'svn log' call that was leftover from debugging
  * removed multiple git-svn calls with a multi-init / multi-fetch
    combination (which weren't tested before, either)
  * replaced `rev-list ... | head -n1` with `rev-parse ...`
    (not sure what I was thinking when I wrote that)

All this saves about 9 seconds from a test run
(53s -> 44s for 'make t91*') on my 1.3GHz Athlon

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: t/t9100-git-svn-basic: remove old check for NO_SYMLINK
Eric Wong [Mon, 1 Jan 2007 05:49:46 +0000 (21:49 -0800)]
git-svn: t/t9100-git-svn-basic: remove old check for NO_SYMLINK

We don't support the svn command-line client anymore; nor
do we support anything before SVN 1.1.0, so we can be certain
symlinks will be supported in the SVN repository.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: remove svnadmin dependency from the tests
Eric Wong [Mon, 1 Jan 2007 05:49:45 +0000 (21:49 -0800)]
git-svn: remove svnadmin dependency from the tests

We require the libraries now, so we can create repositories
using them (and save some executable load time while we're at
it).

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoi18n: do not leak 'encoding' header even when we cheat the conversion.
Junio C Hamano [Mon, 1 Jan 2007 02:18:23 +0000 (18:18 -0800)]
i18n: do not leak 'encoding' header even when we cheat the conversion.

We special case the case where encoding recorded in the commit
and the output encoding are the same and do not call iconv().
But we should drop 'encoding' header for this case as well for
consistency.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDo not merge random set of refs out of wildcarded refs
Junio C Hamano [Mon, 1 Jan 2007 01:44:37 +0000 (17:44 -0800)]
Do not merge random set of refs out of wildcarded refs

When your fetch configuration has only the wildcards, we would
pick the lexicographically first ref from the remote side for
merging, which was complete nonsense.  Make sure nothing except
the one that is specified with branch.*.merge is merged in this
case.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix formatting for urls section of fetch, pull, and push manpages
Theodore Tso [Sun, 31 Dec 2006 04:11:52 +0000 (23:11 -0500)]
Fix formatting for urls section of fetch, pull, and push manpages

Updated to make the nroff'ed man pages look nicer.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: update tutorial's discussion of origin
J. Bruce Fields [Sun, 31 Dec 2006 23:47:38 +0000 (18:47 -0500)]
Documentation: update tutorial's discussion of origin

Update tutorial's discussion of origin branch to reflect new defaults,
and include a brief mention of git-repo-config.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: update glossary entry for "origin"
J. Bruce Fields [Sun, 31 Dec 2006 23:47:36 +0000 (18:47 -0500)]
Documentation: update glossary entry for "origin"

Update glossary entry for "origin" to reflect fact that it normally now refers
to a remote repository, not a branch.

Also, warning not to work on remote-tracking branches is no longer necessary
since git doesn't allow that.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: update git-clone.txt for clone's new default behavior
J. Bruce Fields [Sun, 31 Dec 2006 23:47:34 +0000 (18:47 -0500)]
Documentation: update git-clone.txt for clone's new default behavior

Fix a couple remaining references to the origin branch.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocs: update cvs-migration.txt to reflect clone's new default behavior
J. Bruce Fields [Sun, 31 Dec 2006 23:47:33 +0000 (18:47 -0500)]
Docs: update cvs-migration.txt to reflect clone's new default behavior

I couldn't think of a really quick way to give all the details, so just refer
readers to the git-repo-config man page instead.

I haven't tested recent cvs import behavior--some time presumably it should be
updated to do something more similar to clone.

Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agosend-pack: tell pack-objects to use its internal rev-list.
Junio C Hamano [Sun, 31 Dec 2006 09:26:53 +0000 (01:26 -0800)]
send-pack: tell pack-objects to use its internal rev-list.

This means one less process in the pipeline to worry about, and
removes about 1/8 of the code.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agosend-pack.c: use is_null_sha1()
Junio C Hamano [Sun, 31 Dec 2006 08:59:53 +0000 (00:59 -0800)]
send-pack.c: use is_null_sha1()

Everybody else uses is_null_sha1() -- there is no point to have its
own is_zero_sha1() anymore.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUpdate documentation for update hook.
Junio C Hamano [Sun, 31 Dec 2006 08:52:25 +0000 (00:52 -0800)]
Update documentation for update hook.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/send-pack-pipeline'
Junio C Hamano [Sun, 31 Dec 2006 08:31:26 +0000 (00:31 -0800)]
Merge branch 'jc/send-pack-pipeline'

* jc/send-pack-pipeline:
  Documentation: illustrate send-pack pipeline.
  send-pack: fix pipeline.

17 years agoAdd test case for update hooks in receive-pack.
Shawn O. Pearce [Sun, 31 Dec 2006 07:20:23 +0000 (02:20 -0500)]
Add test case for update hooks in receive-pack.

Verify that the update hooks work as documented/advertised.  This is
a simple set of tests to check that the update hooks run with the
parameters expected, have their STDOUT and STDERR redirected to
the client side of the connection, and that their STDIN does not
contain any data (as its actually /dev/null).

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation/config.txt (and repo-config manpage): mark-up fix.
Junio C Hamano [Sun, 31 Dec 2006 06:39:24 +0000 (22:39 -0800)]
Documentation/config.txt (and repo-config manpage): mark-up fix.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoTeach Git how to parse standard power of 2 suffixes.
Shawn O. Pearce [Sun, 31 Dec 2006 03:13:05 +0000 (22:13 -0500)]
Teach Git how to parse standard power of 2 suffixes.

Sometimes its necessary to supply a value as a power of two in a
configuration parameter.  In this case the user may want to use the
standard suffixes such as K, M, or G to indicate that the numerical
value should be multiplied by a constant base before being used.

Shell scripts/etc. can also benefit from this automatic option
parsing with `git repo-config --int`.

[jc: with a couple of test and a slight input tightening]

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUse /dev/null for update hook stdin.
Shawn O. Pearce [Sun, 31 Dec 2006 02:55:22 +0000 (21:55 -0500)]
Use /dev/null for update hook stdin.

Currently the update hook invoked by receive-pack has its stdin
connected to the pushing client.  The hook shouldn't attempt to
read from this stream, and doing so may consume data that was
meant for receive-pack.  Instead we should give the update hook
/dev/null as its stdin, ensuring that it always receives EOF and
doesn't disrupt the protocol if it attempts to read any data.

The post-update hook is similar, as it gets invoked with /dev/null
on stdin to prevent the hook from reading data from the client.
Previously we had invoked it with stdout also connected to /dev/null,
throwing away anything on stdout, to prevent client protocol errors.
Instead we should redirect stdout to stderr, like we do with the
update hook.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRedirect update hook stdout to stderr.
Shawn O. Pearce [Sun, 31 Dec 2006 02:55:19 +0000 (21:55 -0500)]
Redirect update hook stdout to stderr.

If an update hook outputs to stdout then that output will be sent
back over the wire to the push client as though it were part of
the git protocol.  This tends to cause protocol errors on the
client end of the connection, as the hook output is not expected
in that context.  Most hook developers work around this by making
sure their hook outputs everything to stderr.

But hooks shouldn't need to perform such special behavior.  Instead
we can just dup stderr to stdout prior to invoking the update hook.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRemove unnecessary argc parameter from run_command_v.
Shawn O. Pearce [Sun, 31 Dec 2006 02:55:15 +0000 (21:55 -0500)]
Remove unnecessary argc parameter from run_command_v.

The argc parameter is never used by the run_command_v family of
functions.  Instead they require that the passed argv[] be NULL
terminated so they can rely on the operating system's execvp
function to correctly pass the arguments to the new process.

Making the caller pass the argc is just confusing, as the caller
could be mislead into believing that the argc might take precendece
over the argv, or that the argv does not need to be NULL terminated.
So goodbye argc.  Don't come back.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAutomatically detect a bare git repository.
Shawn O. Pearce [Sun, 31 Dec 2006 04:30:19 +0000 (23:30 -0500)]
Automatically detect a bare git repository.

Many users find it unfriendly that they can create a bare git
repository easily with `git clone --bare` but are then unable to
run simple commands like `git log` once they cd into that newly
created bare repository.  This occurs because we do not check to
see if the current working directory is a git repository.

Instead of failing out with "fatal: Not a git repository" we should
try to automatically detect if the current working directory is
a bare repository and use that for GIT_DIR, and fail out only if
that doesn't appear to be true.

We test the current working directory only after we have tried
searching up the directory tree.  This is to retain backwards
compatibility with our previous behavior on the off chance that
a user has a 'refs' and 'objects' subdirectories and a 'HEAD'
file that looks like a symref, all stored within a repository's
associated working directory.

This change also consolidates the validation logic between the case
of GIT_DIR being supplied and GIT_DIR not being supplied, cleaning
up the code.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoReplace "GIT_DIR" with GIT_DIR_ENVIRONMENT.
Shawn O. Pearce [Sun, 31 Dec 2006 04:29:11 +0000 (23:29 -0500)]
Replace "GIT_DIR" with GIT_DIR_ENVIRONMENT.

We tend to use the nice constant GIT_DIR_ENVIRONMENT when we
are referring to the "GIT_DIR" constant, but git.c didn't do
so.  Now it does.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUse PATH_MAX constant for --bare.
Shawn O. Pearce [Sun, 31 Dec 2006 04:28:53 +0000 (23:28 -0500)]
Use PATH_MAX constant for --bare.

For easier portability we prefer PATH_MAX over seemingly random
constants like 1024.  Make it so.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoForce core.filemode to false on Cygwin.
Shawn O. Pearce [Sun, 31 Dec 2006 04:53:55 +0000 (23:53 -0500)]
Force core.filemode to false on Cygwin.

Many users have noticed that core.filemode doesn't appear to be
automatically set right on Cygwin when using a repository stored
on NTFS.  The issue is that Cygwin and NTFS correctly supports
the executable mode bit, and Git properly detected that, but most
native Windows applications tend to create files such that Cygwin
sees the executable bit set when it probably shouldn't be.

This is especially bad if the user's favorite editor deletes the
file then recreates it whenever they save (vs. just overwriting)
as now a file that was created with mode 0644 by checkout-index
appears to have mode 0755.

So we introduce NO_TRUSTABLE_FILEMODE, settable at compile time.
Setting this option forces core.filemode to false, even if the
detection code would have returned true.  This option should be
enabled by default on Cygwin.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix formatting for urls section of fetch, pull, and push manpages
Theodore Ts'o [Sun, 31 Dec 2006 01:03:51 +0000 (20:03 -0500)]
Fix formatting for urls section of fetch, pull, and push manpages

The line:

[remote "<remote>"]

was getting swallowed up by asciidoc, causing a critical line in the
explanation for how to store the .git/remotes information in .git/config
to go missing from the git-fetch, git-pull, and git-push manpages.

Put all of the examples into delimited blocks to fix this problem and to
make them look nicer.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoFix yet another subtle xdl_merge() bug
Johannes Schindelin [Sun, 31 Dec 2006 01:07:41 +0000 (02:07 +0100)]
Fix yet another subtle xdl_merge() bug

In very obscure cases, a merge can hit an unexpected code path (where the
original code went as far as saying that this was a bug). This failing
merge was noticed by Alexandre Juillard.

The problem is that the original file contains something like this:

-- snip --
two non-empty lines
before two empty lines

after two empty lines
-- snap --

and this snippet is reduced to _one_ empty line in _both_ new files.
However, it is ambiguous as to which hunk takes the empty line: the first
or the second one?

Indeed in Alexandre's example files, the xdiff algorithm attributes the
empty line to the first hunk in one case, and to the second hunk in the
other case.

(Trimming down the example files _changes_ that behaviour!)

Thus, the call to xdl_merge_cmp_lines() has no chance to realize that the
change is actually identical in both new files. Therefore,
xdl_refine_conflicts() finds an empty diff script, which was not expected
there, because (the original author of xdl_merge() thought)
xdl_merge_cmp_lines() would catch that case earlier.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoi18n: drop "encoding" header in the output after re-coding.
Junio C Hamano [Sat, 30 Dec 2006 23:49:32 +0000 (15:49 -0800)]
i18n: drop "encoding" header in the output after re-coding.

After re-coding the commit message into the encoding the user
specified (either with core.logoutputencidng or --encoding
option), this drops the "encoding" header altogether.  The
output is after re-coding as the user asked (either with the
config or --encoding=<encoding> option), and the extra header
becomes redundant information.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocommit-tree: cope with different ways "utf-8" can be spelled.
Junio C Hamano [Sat, 30 Dec 2006 20:20:43 +0000 (12:20 -0800)]
commit-tree: cope with different ways "utf-8" can be spelled.

People can spell config.commitencoding differently from what we
internally have ("utf-8") to mean UTF-8.  Try to accept them and
treat them equally.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMove commit reencoding parameter parsing to revision.c
Junio C Hamano [Sat, 30 Dec 2006 19:59:08 +0000 (11:59 -0800)]
Move commit reencoding parameter parsing to revision.c

This way, git-rev-list and git-diff-tree with --pretty can use
it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: minor rewording for git-log and git-show pages.
Junio C Hamano [Sat, 30 Dec 2006 10:21:48 +0000 (02:21 -0800)]
Documentation: minor rewording for git-log and git-show pages.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: i18n commit log message notes.
Junio C Hamano [Sat, 30 Dec 2006 10:22:38 +0000 (02:22 -0800)]
Documentation: i18n commit log message notes.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot3900: test log --encoding=none
Junio C Hamano [Sat, 30 Dec 2006 10:35:14 +0000 (02:35 -0800)]
t3900: test log --encoding=none

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocommit re-encoding: fix confusion between no and default conversion.
Junio C Hamano [Sat, 30 Dec 2006 10:18:24 +0000 (02:18 -0800)]
commit re-encoding: fix confusion between no and default conversion.

Telling the git-log family not to do any character conversion is
done with --encoding=none, which sets log_output_encoding to an
empty string.  However, logmsg_reencode() confused this with
log_output_encoding and commit_encoding set to NULL.  The latter
means we should use the default encoding (i.e. utf-8).

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoDocumentation: illustrate send-pack pipeline.
Junio C Hamano [Fri, 29 Dec 2006 20:14:30 +0000 (12:14 -0800)]
Documentation: illustrate send-pack pipeline.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agosend-pack: fix pipeline.
Junio C Hamano [Fri, 29 Dec 2006 10:35:40 +0000 (02:35 -0800)]
send-pack: fix pipeline.

send-pack builds a pipeline that runs "rev-list | pack-objects"
and sends the output from pack-objects to the other side, while
feeding the input side of that pipe from itself.  However, the
file descriptor that is given to this pipeline (so that it can
be dup2(2)'ed into file descriptor 1 of pack-objects) is closed
by the caller before the complex fork+exec dance!  Worse yet,
the caller already dup2's it to 1, so the child process did not
even have to.

I do not understand how this code could possibly have been
working, but it somehow was working by accident.

Merging the sliding mmap() code reveals this problem, presumably
because it keeps one extra file descriptor open for a packfile
and changes the way file descriptors are allocated.  I am too
tired to diagnose the problem now, but this seems to be a
sensible fix.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/curl'
Junio C Hamano [Fri, 29 Dec 2006 19:36:21 +0000 (11:36 -0800)]
Merge branch 'jc/curl'

* jc/curl:
  Work around http-fetch built with cURL 7.16.0

17 years agoFix 'git add' with .gitignore
Junio C Hamano [Fri, 29 Dec 2006 19:01:31 +0000 (11:01 -0800)]
Fix 'git add' with .gitignore

When '*.ig' is ignored, and you have two files f.ig and d.ig/foo
in the working tree,

$ git add .

correctly ignored f.ig but failed to ignore d.ig/foo.  This was
caused by a thinko in an earlier commit 4888c534, when we tried
to allow adding otherwise ignored files.

After reverting that commit, this takes a much simpler approach.
When we have an unmatched pathspec that talks about an existing
pathname, we know it is an ignored path the user tried to add,
so we include it in the set of paths directory walker returned.

This does not let you say "git add -f D" on an ignored directory
D and add everything under D.  People can submit a patch to
further allow it if they want to, but I think it is a saner
behaviour to require explicit paths to be spelled out in such a
case.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoRevert "read_directory: show_both option."
Junio C Hamano [Fri, 29 Dec 2006 18:08:19 +0000 (10:08 -0800)]
Revert "read_directory: show_both option."

This reverts commit 4888c534099012d71d24051deb5b14319747bd1a.

17 years agoAdd info about new test families (8 and 9) to t/README
Jakub Narebski [Fri, 29 Dec 2006 13:39:09 +0000 (14:39 +0100)]
Add info about new test families (8 and 9) to t/README

Signed-off-by: Jakub Narebski <jnareb@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot5400 send-pack test: try a bit more nontrivial transfer.
Junio C Hamano [Fri, 29 Dec 2006 10:25:04 +0000 (02:25 -0800)]
t5400 send-pack test: try a bit more nontrivial transfer.

Not that this reveals anything new, but I did test_tick shell
function in test-lib and found it rather cute and nice.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUse merge-recursive in git-am -3.
Shawn O. Pearce [Thu, 28 Dec 2006 07:35:27 +0000 (02:35 -0500)]
Use merge-recursive in git-am -3.

By switching from merge-resolve to merge-recursive in the 3-way
fallback behavior of git-am we gain a few benefits:

 * renames are automatically handled, like in rebase -m;
 * conflict hunks can reference the patch name;
 * its faster on Cygwin (less forks).

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoAllow merging bare trees in merge-recursive.
Shawn O. Pearce [Thu, 28 Dec 2006 07:35:24 +0000 (02:35 -0500)]
Allow merging bare trees in merge-recursive.

To support wider use cases, such as from within `git am -3`, the
merge-recursive utility needs to accept not just commit-ish but
also tree-ish as arguments on its command line.

If given a tree-ish then merge-recursive will create a virtual commit
wrapping it, with the subject of the commit set to the best name we
can derive for that tree, which is either the command line string
(probably the SHA1), or whatever string appears in GITHEAD_*.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMove better_branch_name above get_ref in merge-recursive.
Shawn O. Pearce [Thu, 28 Dec 2006 07:35:20 +0000 (02:35 -0500)]
Move better_branch_name above get_ref in merge-recursive.

To permit the get_ref function to use the static better_branch_name
function to generate a string on demand I'm moving it up earlier.
The actual logic was not affected in this change.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/utf8'
Junio C Hamano [Fri, 29 Dec 2006 03:03:02 +0000 (19:03 -0800)]
Merge branch 'jc/utf8'

* jc/utf8:
  t3900: test conversion to non UTF-8 as well
  Rename t3900 test vector file
  UTF-8: introduce i18n.logoutputencoding.
  Teach log family --encoding
  i18n.logToUTF8: convert commit log message to UTF-8
  Move encoding conversion routine out of mailinfo to utf8.c

Conflicts:

commit.c

17 years agoAllow non-fast-forward of remote tracking branches in default clone
Junio C Hamano [Fri, 29 Dec 2006 00:32:17 +0000 (16:32 -0800)]
Allow non-fast-forward of remote tracking branches in default clone

This changes the default remote.origin.fetch configuration
created by git-clone so that it allows non-fast-forward updates.

When using the separate-remote layout with reflog enabled, it
does not make much sense to refuse to update the remote tracking
branch just because some of them do not fast-forward.  git-fetch
issues warnings on non-fast-forwardness, and the user can peek
at what the previous state was using the reflog.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agocore.logallrefupdates: log remotes/ tracking branches.
Junio C Hamano [Fri, 29 Dec 2006 00:32:05 +0000 (16:32 -0800)]
core.logallrefupdates: log remotes/ tracking branches.

Not using reflog for tags/ was very sensible; not giving reflog
for the remotes/ was not.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoGIT_SKIP_TESTS: allow users to omit tests that are known to break
Junio C Hamano [Fri, 29 Dec 2006 01:58:00 +0000 (17:58 -0800)]
GIT_SKIP_TESTS: allow users to omit tests that are known to break

In some environments, certain tests have no way of succeeding
due to platform limitation, such as lack of 'unzip' program, or
filesystem that do not allow arbitrary sequence of non-NUL bytes
as pathnames.

You should be able to say something like

$ cd t
$ GIT_SKIP_TESTS=t9200.8 t9200-git-cvsexport-commit.sh

and even:

$ GIT_SKIP_TESTS='t[0-4]??? t91?? t9200.8' make test

to omit such tests.  The value of the environment variable is a
SP separated list of patterns that tells which tests to skip,
and either can match the "t[0-9]{4}" part to skip the whole
test, or t[0-9]{4} followed by ".$number" to say which
particular test to skip.

Note that some tests in the existing test suite rely on previous
test item, so you cannot arbitrarily disable one and expect the
remainder of test to check what the test originally was intended
to check.

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot3900: test conversion to non UTF-8 as well
Junio C Hamano [Fri, 29 Dec 2006 01:36:35 +0000 (17:36 -0800)]
t3900: test conversion to non UTF-8 as well

Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'jc/make'
Junio C Hamano [Fri, 29 Dec 2006 00:43:27 +0000 (16:43 -0800)]
Merge branch 'jc/make'

* jc/make:
  gcc does not necessarily pass runtime libpath with -R

17 years agoupdate hook: redirect _both_ diagnostic lines to stderr upon tag failure
Jim Meyering [Thu, 28 Dec 2006 15:05:02 +0000 (16:05 +0100)]
update hook: redirect _both_ diagnostic lines to stderr upon tag failure

Otherwise, sending the diagnostic to stdout would provoke a
protocol failure.

Signed-off-by: Jim Meyering <jim@meyering.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoxdl_merge(): fix a segmentation fault when refining conflicts
Johannes Schindelin [Thu, 28 Dec 2006 16:13:33 +0000 (17:13 +0100)]
xdl_merge(): fix a segmentation fault when refining conflicts

The function xdl_refine_conflicts() tries to break down huge
conflicts by doing a diff on the conflicting regions. However,
this does not make sense when one side is empty.

Worse, when one side is not only empty, but after EOF, the code
accessed unmapped memory.

Noticed by Luben Tuikov, Shawn Pearce and Alexandre Julliard, the
latter providing a test case.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: sort multi-init output
Eric Wong [Thu, 28 Dec 2006 09:16:22 +0000 (01:16 -0800)]
git-svn: sort multi-init output

This looks a bit more pleasant for users.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: verify_ref() should actually --verify
Eric Wong [Thu, 28 Dec 2006 09:16:21 +0000 (01:16 -0800)]
git-svn: verify_ref() should actually --verify

Not sure how I missed this the first time around...

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: print out the SVN library version in --version, too
Eric Wong [Thu, 28 Dec 2006 09:16:20 +0000 (01:16 -0800)]
git-svn: print out the SVN library version in --version, too

This could be useful in finding new problems and helping users
debug.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agogit-svn: remove non-delta fetch code paths
Eric Wong [Thu, 28 Dec 2006 09:16:19 +0000 (01:16 -0800)]
git-svn: remove non-delta fetch code paths

We have less code to worry about now.  As a bonus, --revision
can be used to reliably skip parts of history whenever fetch is
run, not just the first time.  I'm not sure why anybody would
want to skip history in the middle, however...

For people (nearly everyone at the moment) without the
do_switch() function in their Perl SVN library, the entire tree
must be refetched if --follow-parent is used and a parent is
found.  Future versions of SVN will have a working do_switch()
function accessible via Perl.

Accessing repositories on the local machine (especially file://
ones) is also slightly slower as a result; but I suspect most
git-svn users will be using it to access remote repositories.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot9200-git-cvsexportcommit.sh: quiet down commit
Eric Wong [Thu, 28 Dec 2006 09:10:52 +0000 (01:10 -0800)]
t9200-git-cvsexportcommit.sh: quiet down commit

Also, fixed an unportable use of 'export'.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agotest-lib: quiet down init-db output for tests
Eric Wong [Thu, 28 Dec 2006 09:10:51 +0000 (01:10 -0800)]
test-lib: quiet down init-db output for tests

I don't think anybody running tests needs to know they're
running init-db and creating a repository for testing.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agot6024-recursive-merge: quiet down this test
Eric Wong [Thu, 28 Dec 2006 09:10:50 +0000 (01:10 -0800)]
t6024-recursive-merge: quiet down this test

We get an extra measure of error checking here as well.
While we're at it, also removed a less portable use of export.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoMerge branch 'js/shallow'
Junio C Hamano [Thu, 28 Dec 2006 09:25:43 +0000 (01:25 -0800)]
Merge branch 'js/shallow'

* js/shallow:
  fetch-pack: Do not fetch tags for shallow clones.
  get_shallow_commits: Avoid memory leak if a commit has been reached already.
  git-fetch: Reset shallow_depth before auto-following tags.
  upload-pack: Check for NOT_SHALLOW flag before sending a shallow to the client.
  fetch-pack: Properly remove the shallow file when it becomes empty.
  shallow clone: unparse and reparse an unshallowed commit
  Why didn't we mark want_obj as ~UNINTERESTING in the old code?
  Why does it mean we do not have to register shallow if we have one?
  We should make sure that the protocol is still extensible.
  add tests for shallow stuff
  Shallow clone: do not ignore shallowness when following tags
  allow deepening of a shallow repository
  allow cloning a repository "shallowly"
  support fetching into a shallow repository
  upload-pack: no longer call rev-list

17 years agoAllow git-merge to select the default strategy.
Shawn O. Pearce [Thu, 28 Dec 2006 07:35:11 +0000 (02:35 -0500)]
Allow git-merge to select the default strategy.

Now that git-merge knows how to use the pull.{twohead,octopus}
configuration options to select the default merge strategy there
is no reason for git-pull to do the same immediately prior to
invoking git-merge.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoHonor pull.{twohead,octopus} in git-merge.
Shawn O. Pearce [Thu, 28 Dec 2006 07:35:05 +0000 (02:35 -0500)]
Honor pull.{twohead,octopus} in git-merge.

If git-merge is invoked without a strategy argument it is probably
being run as a porcelain-ish command directly and is not being run
from within git-pull.  However we still should honor whatever merge
strategy the user may have selected in their configuration, just as
`git-pull .` would have.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoEnsure `git-pull` fails if `git-merge` fails.
Shawn O. Pearce [Thu, 28 Dec 2006 07:35:01 +0000 (02:35 -0500)]
Ensure `git-pull` fails if `git-merge` fails.

If git-merge exits with a non-zero exit status so should git-pull.
This way the caller of git-pull knows the task did not complete
successfully simply by checking the process exit status.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUse branch names in 'git-rebase -m' conflict hunks.
Shawn O. Pearce [Thu, 28 Dec 2006 07:34:56 +0000 (02:34 -0500)]
Use branch names in 'git-rebase -m' conflict hunks.

If a three-way merge in git-rebase generates a conflict then we
should take advantage of git-merge-recursive's ability to include
the branch name of each side of the conflict hunk by setting the
GITHEAD_* environment variables.

In the case of rebase there aren't really two clear branches; we
have the branch we are rebasing onto, and we have the branch we are
currently rebasing.  Since most conflicts will be arising between
the user's current branch and the branch they are rebasing onto
we assume the stuff that isn't in the current commit is the "onto"
branch and the stuff in the current commit is the "current" branch.

This assumption may however come up wrong if the user resolves one
conflict in such a way that it conflicts again on a future commit
also being rebased.  In this case the user's prior resolution will
appear to be in the "onto" part of the hunk.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoHonor GIT_REFLOG_ACTION in git-rebase.
Shawn O. Pearce [Thu, 28 Dec 2006 07:34:52 +0000 (02:34 -0500)]
Honor GIT_REFLOG_ACTION in git-rebase.

To help correctly log actions caused by porcelain which invoke
git-reset directly we should honor the setting of GIT_REFLOG_ACTION
which we inherited from our caller.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
17 years agoUse GIT_REFLOG_ACTION environment variable instead.
Shawn O. Pearce [Thu, 28 Dec 2006 07:34:48 +0000 (02:34 -0500)]
Use GIT_REFLOG_ACTION environment variable instead.

Junio rightly pointed out that the --reflog-action parameter
was starting to get out of control, as most porcelain code
needed to hand it to other porcelain and plumbing alike to
ensure the reflog contained the top-level user action and
not the lower-level actions it invoked.

At Junio's suggestion we are introducing the new set_reflog_action
function to all shell scripts, allowing them to declare early on
what their default reflog name should be, but this setting only
takes effect if the caller has not already set the GIT_REFLOG_ACTION
environment variable.

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