git.git
18 years agoMerge branch 'jc/lt-tree-n-cache-tree' into next
Junio C Hamano [Tue, 30 May 2006 05:16:56 +0000 (22:16 -0700)]
Merge branch 'jc/lt-tree-n-cache-tree' into next

* jc/lt-tree-n-cache-tree:
  adjust to the rebased series by Linus.
  Remove last vestiges of generic tree_entry_list
  Convert fetch.c: process_tree() to raw tree walker
  Convert "mark_tree_uninteresting()" to raw tree walker
  Remove unused "zeropad" entry from tree_list_entry
  fsck-objects: avoid unnecessary tree_entry_list usage
  Remove "tree->entries" tree-entry list from tree parser
  builtin-read-tree.c: avoid tree_entry_list in prime_cache_tree_rec()
  Switch "read_tree_recursive()" over to tree-walk functionality
  Make "tree_entry" have a SHA1 instead of a union of object pointers
  Make "struct tree" contain the pointer to the tree buffer
  Make git-diff-tree indicate when it flushes
  Remove unnecessary output from t3600-rm.

18 years agoMerge branch 'jc/lt-tree-n-cache-tree' into lt/tree-2
Junio C Hamano [Tue, 30 May 2006 05:13:09 +0000 (22:13 -0700)]
Merge branch 'jc/lt-tree-n-cache-tree' into lt/tree-2

* jc/lt-tree-n-cache-tree:
  adjust to the rebased series by Linus.
  Remove "tree->entries" tree-entry list from tree parser
  Switch "read_tree_recursive()" over to tree-walk functionality
  Make "tree_entry" have a SHA1 instead of a union of object pointers
  Add raw tree buffer info to "struct tree"

This results as if an "ours" merge absorbed the previous "next"
branch change into the 10-patch series, but it really is a result
of an honest merge.

nothing to commit

18 years agoadjust to the rebased series by Linus.
Junio C Hamano [Tue, 30 May 2006 05:00:21 +0000 (22:00 -0700)]
adjust to the rebased series by Linus.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRemove last vestiges of generic tree_entry_list
Linus Torvalds [Mon, 29 May 2006 19:21:28 +0000 (12:21 -0700)]
Remove last vestiges of generic tree_entry_list

The old tree_entry_list is dead, long live the unified single tree
parser.

Yes, we now still have a compatibility function to create a bogus
tree_entry_list in builtin-read-tree.c, but that is now entirely local
to that very messy piece of code.

I'd love to clean read-tree.c up too, but I'm too scared right now, so
the best I can do is to just contain the damage, and try to make sure
that no new users of the tree_entry_list sprout up by not having it as
an exported interface any more.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoConvert fetch.c: process_tree() to raw tree walker
Linus Torvalds [Mon, 29 May 2006 19:20:48 +0000 (12:20 -0700)]
Convert fetch.c: process_tree() to raw tree walker

This leaves only the horrid code in builtin-read-tree.c using the old
interface. Some day I will gather the strength to tackle that one too.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoConvert "mark_tree_uninteresting()" to raw tree walker
Linus Torvalds [Mon, 29 May 2006 19:20:14 +0000 (12:20 -0700)]
Convert "mark_tree_uninteresting()" to raw tree walker

Not very many users to go..

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRemove unused "zeropad" entry from tree_list_entry
Linus Torvalds [Mon, 29 May 2006 19:19:37 +0000 (12:19 -0700)]
Remove unused "zeropad" entry from tree_list_entry

That was a hack, only needed because 'git fsck-objects' didn't look at
the raw tree format.  Now that fsck traverses the tree itself, we can
drop it.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agofsck-objects: avoid unnecessary tree_entry_list usage
Linus Torvalds [Mon, 29 May 2006 19:19:02 +0000 (12:19 -0700)]
fsck-objects: avoid unnecessary tree_entry_list usage

Prime example of where the raw tree parser is easier for everybody.

[jc: "Aieee" one-liner fix from the list applied. ]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRemove "tree->entries" tree-entry list from tree parser
Linus Torvalds [Mon, 29 May 2006 19:18:33 +0000 (12:18 -0700)]
Remove "tree->entries" tree-entry list from tree parser

Instead, just use the tree buffer directly, and use the tree-walk
infrastructure to walk the buffers instead of the tree-entry list.

The tree-entry list is inefficient, and generates tons of small
allocations for no good reason. The tree-walk infrastructure is
generally no harder to use than following a linked list, and allows
us to do most tree parsing in-place.

Some programs still use the old tree-entry lists, and are a bit
painful to convert without major surgery. For them we have a helper
function that creates a temporary tree-entry list on demand.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobuiltin-read-tree.c: avoid tree_entry_list in prime_cache_tree_rec()
Linus Torvalds [Mon, 29 May 2006 19:18:00 +0000 (12:18 -0700)]
builtin-read-tree.c: avoid tree_entry_list in prime_cache_tree_rec()

Use the raw tree walker instead.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSwitch "read_tree_recursive()" over to tree-walk functionality
Linus Torvalds [Mon, 29 May 2006 19:17:28 +0000 (12:17 -0700)]
Switch "read_tree_recursive()" over to tree-walk functionality

Don't use the tree_entry list any more.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake "tree_entry" have a SHA1 instead of a union of object pointers
Linus Torvalds [Mon, 29 May 2006 19:16:46 +0000 (12:16 -0700)]
Make "tree_entry" have a SHA1 instead of a union of object pointers

This is preparatory work for further cleanups, where we try to make
tree_entry look more like the more efficient tree-walk descriptor.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake "struct tree" contain the pointer to the tree buffer
Linus Torvalds [Mon, 29 May 2006 19:16:12 +0000 (12:16 -0700)]
Make "struct tree" contain the pointer to the tree buffer

This allows us to avoid allocating information for names etc, because
we can just use the information from the tree buffer directly.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake git-diff-tree indicate when it flushes
Paul Mackerras [Tue, 30 May 2006 02:01:38 +0000 (19:01 -0700)]
Make git-diff-tree indicate when it flushes

There are times when gitk needs to know that the commits it has sent
to git-diff-tree --stdin did not match, and it needs to know in a
timely fashion even if none of them match.  At the moment,
git-diff-tree outputs nothing for non-matching commits, so it is
impossible for gitk to distinguish between git-diff-tree being slow
and git-diff-tree saying no.

This makes git-diff-tree flush its output and echo back the
input line when it is not a valid-looking object name.  Gitk, or
other users of git-diff-tree --stdin, can use a blank line or
any other "marker line" to indicate that git-diff-tree has
processed all the commits on its input up to the echoed back
marker line, and any commits that have not been output do not
match.

[jc: re-done after a couple of back-and-forth discussion on the list.]

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRemove unnecessary output from t3600-rm.
Shawn Pearce [Mon, 29 May 2006 08:06:48 +0000 (04:06 -0400)]
Remove unnecessary output from t3600-rm.

Moved the setup commands into test_expect_success blocks so their
output is hidden unless -v is used.  This makes the test suite look
a little cleaner when the rm test-file setup step fails (and was
expected to fail for most cases).

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/lt-tree-n-cache-tree' into next
Junio C Hamano [Mon, 29 May 2006 06:05:02 +0000 (23:05 -0700)]
Merge branch 'jc/lt-tree-n-cache-tree' into next

* jc/lt-tree-n-cache-tree:
  Remove "tree->entries" tree-entry list from tree parser
  Switch "read_tree_recursive()" over to tree-walk functionality
  Make "tree_entry" have a SHA1 instead of a union of object pointers
  Add raw tree buffer info to "struct tree"
  Don't use "sscanf()" for tree mode scanning
  git-fetch: avoid using "case ... in (arm)"

18 years agoMerge branch 'lt/apply'
Junio C Hamano [Mon, 29 May 2006 06:00:50 +0000 (23:00 -0700)]
Merge branch 'lt/apply'

* lt/apply:
  apply: force matching at the beginning.
  Add a test-case for git-apply trying to add an ending line
  apply: treat EOF as proper context.

18 years agoMerge branch 'jc/cache-tree'
Junio C Hamano [Mon, 29 May 2006 05:57:47 +0000 (22:57 -0700)]
Merge branch 'jc/cache-tree'

* jc/cache-tree: (26 commits)
  builtin-rm: squelch compiler warnings.
  git-write-tree writes garbage on sparc64
  Fix crash when reading the empty tree
  fsck-objects: do not segfault on missing tree in cache-tree
  cache-tree: a bit more debugging support.
  read-tree: invalidate cache-tree entry when a new index entry is added.
  Fix test-dump-cache-tree in one-tree disappeared case.
  fsck-objects: mark objects reachable from cache-tree
  cache-tree: replace a sscanf() by two strtol() calls
  cache-tree.c: typefix
  test-dump-cache-tree: validate the cached data as well.
  cache_tree_update: give an option to update cache-tree only.
  read-tree: teach 1-way merege and plain read to prime cache-tree.
  read-tree: teach 1 and 2 way merges about cache-tree.
  update-index: when --unresolve, smudge the relevant cache-tree entries.
  test-dump-cache-tree: report number of subtrees.
  cache-tree: sort the subtree entries.
  Teach fsck-objects about cache-tree.
  index: make the index file format extensible.
  cache-tree: protect against "git prune".
  ...

Conflicts:

Makefile, builtin.h, git.c: resolved the same way as in next.

18 years agoMerge branch 'lt/tree' into jc/lt-tree-n-cache-tree
Junio C Hamano [Mon, 29 May 2006 05:47:53 +0000 (22:47 -0700)]
Merge branch 'lt/tree' into jc/lt-tree-n-cache-tree

* lt/tree: (98 commits)
  Remove "tree->entries" tree-entry list from tree parser
  Switch "read_tree_recursive()" over to tree-walk functionality
  Make "tree_entry" have a SHA1 instead of a union of object pointers
  Add raw tree buffer info to "struct tree"
  Don't use "sscanf()" for tree mode scanning
  git-fetch: avoid using "case ... in (arm)"
  mailinfo: skip bogus UNIX From line inside body
  mailinfo: More carefully parse header lines in read_one_header_line()
  Allow in body headers beyond the in body header prefix.
  More accurately detect header lines in read_one_header_line
  In handle_body only read a line if we don't already have one.
  Refactor commit messge handling.
  Move B and Q decoding into check header.
  Make read_one_header_line return a flag not a length.
  Fix memory leak in "git rev-list --objects"
  gitview: Move the console error messages to message dialog
  gitview: Add key binding for F5.
  Let git-clone to pass --template=dir option to git-init-db.
  Make cvsexportcommit create parent directories as needed.
  Document current cvsexportcommit limitations.
  ...

Conflicts:

Makefile, builtin.h, git.c are trivially resolved.
builtin-read-tree.c needed adjustment for the tree
parser change.

18 years agoMerge branch 'jc/dirwalk-n-cache-tree' into jc/cache-tree
Junio C Hamano [Mon, 29 May 2006 05:34:34 +0000 (22:34 -0700)]
Merge branch 'jc/dirwalk-n-cache-tree' into jc/cache-tree

* jc/dirwalk-n-cache-tree: (212 commits)
  builtin-rm: squelch compiler warnings.
  Add builtin "git rm" command
  Move pathspec matching from builtin-add.c into dir.c
  Prevent bogus paths from being added to the index.
  builtin-add: fix unmatched pathspec warnings.
  Remove old "git-add.sh" remnants
  builtin-add: warn on unmatched pathspecs
  Do "git add" as a builtin
  Clean up git-ls-file directory walking library interface
  libify git-ls-files directory traversal
  Add a conversion tool to migrate remote information into the config
  fetch, pull: ask config for remote information
  Fix build procedure for builtin-init-db
  read-tree -m -u: do not overwrite or remove untracked working tree files.
  apply --cached: do not check newly added file in the working tree
  Implement a --dry-run option to git-quiltimport
  Implement git-quiltimport
  Revert "builtin-grep: workaround for non GNU grep."
  builtin-grep: workaround for non GNU grep.
  builtin-grep: workaround for non GNU grep.
  ...

18 years agoRemove "tree->entries" tree-entry list from tree parser
Linus Torvalds [Sun, 28 May 2006 22:13:53 +0000 (15:13 -0700)]
Remove "tree->entries" tree-entry list from tree parser

This finally removes the tree-entry list from "struct tree", since most of
the users can just use the tree-walk infrastructure to walk the raw tree
buffers instead of the tree-entry list.

The tree-entry list is inefficient, and generates tons of small
allocations for no good reason. The tree-walk infrastructure is generally
no harder to use than following a linked list, and allows us to do most
tree parsing in-place.

Some programs still use the old tree-entry lists, and are a bit painful to
convert without major surgery. For them we have a helper function that
creates a temporary tree-entry list on demand. We can convert those too
eventually, but with this they no longer affect any users who don't need
the explicit lists.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoSwitch "read_tree_recursive()" over to tree-walk functionality
Linus Torvalds [Sun, 28 May 2006 22:11:28 +0000 (15:11 -0700)]
Switch "read_tree_recursive()" over to tree-walk functionality

Don't use the tree_entry list, it really had no major reason not to just
walk the raw tree instead.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake "tree_entry" have a SHA1 instead of a union of object pointers
Linus Torvalds [Sun, 28 May 2006 22:10:04 +0000 (15:10 -0700)]
Make "tree_entry" have a SHA1 instead of a union of object pointers

This is preparatory work for further cleanups, where we try to make
tree_entry look more like the more efficient tree-walk descriptor.

Instead of having a union of pointers to blob/tree/objects, this just
makes "struct tree_entry" have the raw SHA1, and makes all the users use
that instead (often that implies adding a "lookup_tree(..)" on the sha1,
but sometimes the user just wanted the SHA1 in the first place, and it
just avoids an unnecessary indirection).

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd raw tree buffer info to "struct tree"
Linus Torvalds [Sun, 28 May 2006 22:07:07 +0000 (15:07 -0700)]
Add raw tree buffer info to "struct tree"

This allows us to avoid allocating information for names etc, because
we can just use the information from the tree buffer directly.

We still keep the old "tree_entry_list" in struct tree as well, so old
users aren't affected, apart from the fact that the allocations are
different (if you free a tree entry, you should no longer free the name
allocation for it, since it's allocated as part of "tree->buffer")

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDon't use "sscanf()" for tree mode scanning
Linus Torvalds [Sun, 28 May 2006 23:16:15 +0000 (16:16 -0700)]
Don't use "sscanf()" for tree mode scanning

Doing an oprofile run on the result of my git rev-list memory leak fixes
and tree parsing cleanups, I was surprised by the third-highest entry
being

samples  %        image name               app name                 symbol name
179751    2.7163  libc-2.4.so              libc-2.4.so              _IO_vfscanf@@GLIBC_2.4

where that 2.7% is actually more than 5% of one CPU, because this was run
on a dual CPU setup with the other CPU just being idle.

That seems to all be from the use of 'sscanf(tree, "%o", &mode)' for the
tree buffer parsing.

So do the trivial octal parsing by hand, which also gives us where the
first space in the string is (and thus where the pathname starts) so we
can get rid of the "strchr(tree, ' ')" call too.

This brings the "git rev-list --all --objects" time down from 63 seconds
to 55 seconds on the historical kernel archive for me, so it's quite
noticeable - tree parsing is a lot of what we end up doing when following
all the objects.

[ I also see a 5% speedup on a full "git fsck-objects" on the current
  kernel archive, so that sscanf() really does seem to have hurt our
  performance by a surprising amount ]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-fetch: avoid using "case ... in (arm)"
Junio C Hamano [Mon, 29 May 2006 02:24:28 +0000 (19:24 -0700)]
git-fetch: avoid using "case ... in (arm)"

NetBSD ash chokes on the optional open parenthesis for case arms.  Inside
$(command substitution), however, bash barfs without.  So adjust things
accordingly.

Originally pointed out by Dennis Stosberg.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/cache-tree' into next
Junio C Hamano [Sun, 28 May 2006 20:39:51 +0000 (13:39 -0700)]
Merge branch 'jc/cache-tree' into next

* jc/cache-tree:
  git-write-tree writes garbage on sparc64

18 years agoMerge branch 'master' into next
Junio C Hamano [Sun, 28 May 2006 20:39:45 +0000 (13:39 -0700)]
Merge branch 'master' into next

* master:
  Fix memory leak in "git rev-list --objects"
  gitview: Move the console error messages to message dialog
  gitview: Add key binding for F5.
  Let git-clone to pass --template=dir option to git-init-db.
  Make cvsexportcommit create parent directories as needed.
  Document current cvsexportcommit limitations.
  Do not call 'cmp' with non-existant -q flag.
  Fix "--abbrev=xyz" for revision listing
  t1002: use -U0 instead of --unified=0
  format-patch: -n and -k are mutually exclusive.

18 years agoMerge branch 'jc/mailinfo'
Junio C Hamano [Sun, 28 May 2006 20:39:05 +0000 (13:39 -0700)]
Merge branch 'jc/mailinfo'

* jc/mailinfo:
  mailinfo: skip bogus UNIX From line inside body

18 years agoMerge branch 'eb/mailinfo'
Junio C Hamano [Sun, 28 May 2006 20:36:37 +0000 (13:36 -0700)]
Merge branch 'eb/mailinfo'

* eb/mailinfo:
  mailinfo: More carefully parse header lines in read_one_header_line()
  Allow in body headers beyond the in body header prefix.
  More accurately detect header lines in read_one_header_line
  In handle_body only read a line if we don't already have one.
  Refactor commit messge handling.
  Move B and Q decoding into check header.
  Make read_one_header_line return a flag not a length.

18 years agogit-write-tree writes garbage on sparc64
Dennis Stosberg [Sun, 28 May 2006 19:08:08 +0000 (21:08 +0200)]
git-write-tree writes garbage on sparc64

In the "next" branch, write_index_ext_header() writes garbage on a
64-bit big-endian machine; the written index file will be unreadable.
I noticed this on NetBSD/sparc64. Reproducible with:

$ git init-db
$ :>file
$ git-update-index --add file
$ git-write-tree
$ git-update-index
error: index uses  extension, which we do not understand
fatal: index file corrupt

Signed-off-by: Dennis Stosberg <dennis@stosberg.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix memory leak in "git rev-list --objects"
Linus Torvalds [Sun, 28 May 2006 18:37:23 +0000 (11:37 -0700)]
Fix memory leak in "git rev-list --objects"

Martin Langhoff points out that "git repack -a" ends up using up a lot of
memory for big archives, and that git cvsimport probably should do only
incremental repacks in order to avoid having repacking flush all the
caches.

The big majority of the memory usage of repacking is from git rev-list
tracking all objects, and this patch should go a long way in avoiding the
excessive memory usage: the bulk of it was due to the object names being
leaked from the tree parser.

For the historic Linux kernel archive, this simple patch does:

Before:
/usr/bin/time git-rev-list --all --objects > /dev/null

72.45user 0.82system 1:13.55elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+125376minor)pagefaults 0swaps

After:
/usr/bin/time git-rev-list --all --objects > /dev/null

75.22user 0.48system 1:16.34elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (0major+43921minor)pagefaults 0swaps

where we do end up wasting a bit of time on some extra strdup()s (which
could be avoided, but that would require tracking where the pathnames came
from), but we avoid a lot of memory usage.

Minor page faults track maximum RSS very closely (each page fault maps in
one page into memory), so the reduction from 125376 page faults to 43921
means a rough reduction of VM footprint from almost half a gigabyte to
about a third of that. Those numbers were also double-checked by looking
at "top" while the process was running.

(Side note: at least part of the remaining VM footprint is the mapping of
the 177MB pack-file, so the remaining memory use is at least partly "well
behaved" from a project caching perspective).

For the current git archive itself, the memory usage for a "--all
--objects" rev-list invocation dropped from 7128 pages to 2318 (27MB to
9MB), so the reduction seems to hold for much smaller projects too.

For regular "git-rev-list" usage (ie without the "--objects" flag) this
patch has no impact.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogitview: Move the console error messages to message dialog
Aneesh Kumar K.V [Sat, 27 May 2006 10:25:32 +0000 (15:55 +0530)]
gitview: Move the console error messages to message dialog

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogitview: Add key binding for F5.
Aneesh Kumar K.V [Sat, 27 May 2006 10:24:14 +0000 (15:54 +0530)]
gitview: Add key binding for F5.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoLet git-clone to pass --template=dir option to git-init-db.
Junio C Hamano [Sun, 28 May 2006 17:14:38 +0000 (10:14 -0700)]
Let git-clone to pass --template=dir option to git-init-db.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake cvsexportcommit create parent directories as needed.
Yann Dirson [Sat, 27 May 2006 16:39:35 +0000 (18:39 +0200)]
Make cvsexportcommit create parent directories as needed.

This is a quick port of my initial patch for 1.0.7, that I had forgotten to
post.  Possibly needs some testing before applying to master.

Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocument current cvsexportcommit limitations.
Yann Dirson [Sat, 27 May 2006 16:39:33 +0000 (18:39 +0200)]
Document current cvsexportcommit limitations.

Since there is no bugtracker that I know of, let's just use the scripts
themselves to document their limitations.

Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDo not call 'cmp' with non-existant -q flag.
Yann Dirson [Sat, 27 May 2006 16:39:31 +0000 (18:39 +0200)]
Do not call 'cmp' with non-existant -q flag.

I cannot find when that flag was removed if it ever existed, I can find
nothing about it in the ChangeLog and NEWS file of GNU diff.  The current
flag is -s aka --quiet aka --silent, so let's use -s, assuming it is a
portable flag.  Feel free to lart me with a POSIX bible if needed.

Signed-off-by: Yann Dirson <ydirson@altern.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoFix "--abbrev=xyz" for revision listing
Linus Torvalds [Sat, 27 May 2006 19:24:30 +0000 (12:24 -0700)]
Fix "--abbrev=xyz" for revision listing

The revision argument parsing was happily parsing "--abbrev", but it
didn't parse "--abbrev=<n>".

Which was hidden by the fact that the diff options _would_ parse
--abbrev=<n>, so it would actually silently parse it, it just
wouldn't use it for the same things that a plain "--abbrev" was
used for.

Which seems a bit insane.

With this patch, if you do "git log --abbrev=10" it will abbreviate the
merge parent commit ID's to ten hex characters, which was probably what
you expected.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot1002: use -U0 instead of --unified=0
Linus Torvalds [Fri, 26 May 2006 05:41:02 +0000 (22:41 -0700)]
t1002: use -U0 instead of --unified=0

Using "-U0" is definitely more portable than using "--unified=0",
so we should do that regardless.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoformat-patch: -n and -k are mutually exclusive.
Junio C Hamano [Sun, 28 May 2006 16:23:29 +0000 (09:23 -0700)]
format-patch: -n and -k are mutually exclusive.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Fri, 26 May 2006 23:36:20 +0000 (16:36 -0700)]
Merge branch 'master' into next

* master:
  Call builtin ls-tree in git-cat-file -p
  built-in format-patch: various fixups.
  Add instructions to commit template.

18 years agoCall builtin ls-tree in git-cat-file -p
Petr Baudis [Fri, 26 May 2006 16:59:17 +0000 (18:59 +0200)]
Call builtin ls-tree in git-cat-file -p

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobuilt-in format-patch: various fixups.
Junio C Hamano [Fri, 26 May 2006 18:30:49 +0000 (11:30 -0700)]
built-in format-patch: various fixups.

 - The --start-number handling introduced breakage in the normal
   code path.  It started numbering at 0 when not --numbered,
   for example.

 - When generating one file per patch, we needlessly added an
   extra blank line in front for second and subsequent files.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoAdd instructions to commit template.
Martin Waitz [Thu, 25 May 2006 23:42:18 +0000 (01:42 +0200)]
Add instructions to commit template.

New users can be irritated by the git status text in their editor.
Let's give them a short help.

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'eb/mailinfo' into next
Junio C Hamano [Fri, 26 May 2006 07:51:05 +0000 (00:51 -0700)]
Merge branch 'eb/mailinfo' into next

* eb/mailinfo:
  mailinfo: More carefully parse header lines in read_one_header_line()

18 years agoMerge branch 'ew/tests' into next
Junio C Hamano [Fri, 26 May 2006 07:51:01 +0000 (00:51 -0700)]
Merge branch 'ew/tests' into next

* ew/tests:
  t6000lib: workaround a possible dash bug
  t5500-fetch-pack: remove local (bashism) usage.
  tests: Remove heredoc usage inside quotes
  t3300-funny-names: shell portability fixes
  git-format-patch --start-number <n>
  Don't write directly to a make target ($@).
  bogus "fatal: Not a git repository"
  Documentation/Makefile: remove extra /
  cvsimport: avoid "use" with :tag

18 years agomailinfo: More carefully parse header lines in read_one_header_line()
Junio C Hamano [Fri, 26 May 2006 07:46:58 +0000 (00:46 -0700)]
mailinfo: More carefully parse header lines in read_one_header_line()

We exited prematurely from header parsing loop when the header
field did not have a space after the colon but we insisted on
it, and we got the check wrong because we forgot that we strip
the trailing whitespace before we do the check.

The space after the colon is not even required by RFC2822, so
stop requiring it.  While we are at it, the header line is
specified to be more strict than "anything with a colon in it"
(there must be one or more characters before the colon, and they
must not be controls, SP or non US-ASCII), so implement that
check as well, lest we mistakenly think something like:

Bogus not a header line: this is not.

as a header line.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot6000lib: workaround a possible dash bug
Eric Wong [Fri, 26 May 2006 02:06:18 +0000 (19:06 -0700)]
t6000lib: workaround a possible dash bug

pdksh doesn't need this patch, of course bash works fine since
that what most users use.

Normally, 'var=val command' seems to work fine with dash, but
perhaps there's something weird going on with "$@".  dash is
pretty widespread, so it'll be good to support this even though
it does seem like a bug in dash.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot5500-fetch-pack: remove local (bashism) usage.
Eric Wong [Fri, 26 May 2006 02:06:17 +0000 (19:06 -0700)]
t5500-fetch-pack: remove local (bashism) usage.

None of the variables seem to conflict, so local was unnecessary.

Also replaced ${var:pos:len} with the sed equivalent.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agotests: Remove heredoc usage inside quotes
Eric Wong [Fri, 26 May 2006 02:06:16 +0000 (19:06 -0700)]
tests: Remove heredoc usage inside quotes

The use of heredoc inside quoted strings doesn't seem to be
supported by dash.  pdksh seems to handle it fine, however.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agot3300-funny-names: shell portability fixes
Eric Wong [Fri, 26 May 2006 02:06:15 +0000 (19:06 -0700)]
t3300-funny-names: shell portability fixes

echo isn't remotely standardized for handling backslashes,
so cat + heredoc seems better

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-format-patch --start-number <n>
Johannes Schindelin [Thu, 25 May 2006 21:55:11 +0000 (23:55 +0200)]
git-format-patch --start-number <n>

Since the "a..b c..d" syntax is interpreted as "b ^a d ^c" as other
range-ish commands, if you want to format a..b and then c..d and end
up with files consecutively numbered, the second run needs to be able
to tell the command what number to start from.

This does not imply --numbered (which gives [PATCH n/m] to the subject).

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDon't write directly to a make target ($@).
Jim Meyering [Thu, 25 May 2006 16:52:01 +0000 (18:52 +0200)]
Don't write directly to a make target ($@).

Otherwise, if make is suspended, or killed with prejudice, or if the
system crashes, you could be left with an up-to-date, yet corrupt,
generated file.

I left off the `clean' addition, because I believe "make clean" should
not remove wildcard patterns like "*+", on the off-chance that someone
uses names like that for files they care about.  Besides, in practice,
those temporary files are left behind so rarely that they're not a bother,
and they're removed again as part of the next build.

[jc: sign-off?]

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agobogus "fatal: Not a git repository"
Linus Torvalds [Thu, 25 May 2006 15:22:42 +0000 (08:22 -0700)]
bogus "fatal: Not a git repository"

I was just testing that "git ls-remote" change by Junio, and when you're
not in a git repository, it gives this totally bogus warning. The _target_
obviously has to be a git repository, but there's no reason why you'd have
to be in a local git repo when doing an ls-remote.

The reason is commit 73136b2e8a8ee024320c5ac6a0f14f912432bf03 by Dscho: it
adds calls to git-repo-config in git-parse-remote.sh to get the remote
shorthands etc.

Now, either we should just hide and ignore the error from git-repo-config
(probably bad, because some errors _are_ valid - like git-repo-config
failing due to bad syntax in the config file), or we should just make
git-repo-config quietly handle the case of not being in a git repository.

This does the latter: just quietly accepting (and doing nothing - trying
to set a value will result in the lock-file failing) our lot in life
sounds better than dying with a bogus error message.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Acked-By: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoDocumentation/Makefile: remove extra /
Martin Waitz [Thu, 25 May 2006 12:37:46 +0000 (14:37 +0200)]
Documentation/Makefile: remove extra /

As both DESTDIR and the prefix are supposed to be absolute pathnames
they can simply be concatenated without an extra / (like in the main Makefile).
The extra slash may even break installation on Windows.

[jc: adjusted an earlier workaround for this problem in the dist-doc
 target in the main Makefile as well. ]

Signed-off-by: Martin Waitz <tali@admingilde.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsimport: avoid "use" with :tag
Jeff King [Wed, 24 May 2006 13:58:28 +0000 (09:58 -0400)]
cvsimport: avoid "use" with :tag

Avoid "use POSIX qw(strftime dup2 :errno_h)"; it was reported
that a Perl installations on Mandrake 9.1 did not like it, even
though it understood "use POSIX qw(:errno_h)".  Funny.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'sp/reflog' into next
Junio C Hamano [Thu, 25 May 2006 05:32:44 +0000 (22:32 -0700)]
Merge branch 'sp/reflog' into next

* sp/reflog:
  Test that git-branch -l works.
  Verify git-commit provides a reflog message.

18 years agoMerge branch 'jc/fetchupload' into next
Junio C Hamano [Thu, 25 May 2006 05:32:38 +0000 (22:32 -0700)]
Merge branch 'jc/fetchupload' into next

* jc/fetchupload:
  fetch-pack: give up after getting too many "ack continue"
  cat-file: document -p option
  Built git-upload-tar should be ignored.
  ls-remote: fix rsync:// to report HEAD

18 years agofetch-pack: give up after getting too many "ack continue"
Junio C Hamano [Thu, 25 May 2006 04:48:34 +0000 (21:48 -0700)]
fetch-pack: give up after getting too many "ack continue"

If your repository have more roots than the remote repository
you ask an object for, the remote upload-pack keeps responding
"ack continue" until it fills up its received-have buffer
(currently 256 entries).  Usually this is not a problem because
the requester stops traversing the ancestry chain from the commit
it gets "ack continue" for, but this mechanism does not work as
a roadblock when it traverses down the path to the root the
other side does not have.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocat-file: document -p option
Jeff King [Thu, 25 May 2006 01:22:32 +0000 (21:22 -0400)]
cat-file: document -p option

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoTest that git-branch -l works.
Shawn Pearce [Thu, 25 May 2006 03:34:04 +0000 (23:34 -0400)]
Test that git-branch -l works.

If the user supplies -l to git-branch when creating a new branch
then the new branch's log should be created automatically and the
branch creation should be logged in that log.

Further if a branch is being deleted and it had a log then also
verify that the log was deleted.

Test git-checkout -b foo -l for creating a new branch foo with a
log and checking out that branch.

Fixed git-checkout -b foo -l as the branch variable name was
incorrect in the script.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoVerify git-commit provides a reflog message.
Shawn Pearce [Thu, 25 May 2006 03:33:18 +0000 (23:33 -0400)]
Verify git-commit provides a reflog message.

The reflog message from git-commit should include the first line
of the commit message as supplied by the user.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoBuilt git-upload-tar should be ignored.
Shawn Pearce [Thu, 25 May 2006 02:48:39 +0000 (22:48 -0400)]
Built git-upload-tar should be ignored.

Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agols-remote: fix rsync:// to report HEAD
Junio C Hamano [Thu, 25 May 2006 04:36:14 +0000 (21:36 -0700)]
ls-remote: fix rsync:// to report HEAD

This prevented recent git-clone from checking out the working
tree files in the cloned repository.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'sp/reflog' into next
Junio C Hamano [Wed, 24 May 2006 23:51:09 +0000 (16:51 -0700)]
Merge branch 'sp/reflog' into next

* sp/reflog:
  fetch.c: remove an unused variable and dead code.

18 years agoMerge branch 'master' into sp/reflog
Junio C Hamano [Wed, 24 May 2006 23:49:24 +0000 (16:49 -0700)]
Merge branch 'master' into sp/reflog

* master: (90 commits)
  fetch.c: remove an unused variable and dead code.
  Clean up sha1 file writing
  Builtin git-cat-file
  builtin format-patch: squelch content-type for 7-bit ASCII
  CMIT_FMT_EMAIL: Q-encode Subject: and display-name part of From: fields.
  add more informative error messages to git-mktag
  remove the artificial restriction tagsize < 8kb
  git-rebase: use canonical A..B syntax to format-patch
  git-format-patch: now built-in.
  fmt-patch: Support --attach
  fmt-patch: understand old <his> notation
  Teach fmt-patch about --keep-subject
  Teach fmt-patch about --numbered
  fmt-patch: implement -o <dir>
  fmt-patch: output file names to stdout
  Teach fmt-patch to write individual files.
  built-in tar-tree and remote tar-tree
  Builtin git-diff-files, git-diff-index, git-diff-stages, and git-diff-tree.
  Builtin git-show-branch.
  Builtin git-apply.
  ...

18 years agofetch.c: remove an unused variable and dead code.
Junio C Hamano [Wed, 24 May 2006 23:42:38 +0000 (16:42 -0700)]
fetch.c: remove an unused variable and dead code.

Funnily enough, this variable was never assigned ever since it
was introduced, and has been protecting some code that has never
been executed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'lt/apply' into next
Junio C Hamano [Wed, 24 May 2006 21:11:08 +0000 (14:11 -0700)]
Merge branch 'lt/apply' into next

* lt/apply:
  Clean up sha1 file writing
  Builtin git-cat-file
  apply: force matching at the beginning.
  Add a test-case for git-apply trying to add an ending line
  apply: treat EOF as proper context.

18 years agoMerge branch 'master' into lt/apply
Junio C Hamano [Wed, 24 May 2006 21:08:30 +0000 (14:08 -0700)]
Merge branch 'master' into lt/apply

* master: (40 commits)
  Clean up sha1 file writing
  Builtin git-cat-file
  builtin format-patch: squelch content-type for 7-bit ASCII
  CMIT_FMT_EMAIL: Q-encode Subject: and display-name part of From: fields.
  add more informative error messages to git-mktag
  remove the artificial restriction tagsize < 8kb
  git-rebase: use canonical A..B syntax to format-patch
  git-format-patch: now built-in.
  fmt-patch: Support --attach
  fmt-patch: understand old <his> notation
  Teach fmt-patch about --keep-subject
  Teach fmt-patch about --numbered
  fmt-patch: implement -o <dir>
  fmt-patch: output file names to stdout
  Teach fmt-patch to write individual files.
  Use RFC2822 dates from "git fmt-patch".
  git-fmt-patch: thinkofix to show [PATCH] properly.
  rename internal format-patch wip
  Minor tweak on subject line in --pretty=email
  Tentative built-in format-patch.
  ...

18 years agoClean up sha1 file writing
Linus Torvalds [Wed, 24 May 2006 15:30:54 +0000 (08:30 -0700)]
Clean up sha1 file writing

This cleans up and future-proofs the sha1 file writing in sha1_file.c.

In particular, instead of doing a simple "write()" call and just verifying
that it succeeds (or - as in one place - just assuming it does), it uses
"write_buffer()" to write data to the file descriptor while correctly
checking for partial writes, EINTR etc.

It also splits up write_sha1_to_fd() to be a lot more readable: if we need
to re-create the compressed object, we do so in a separate helper
function, making the logic a whole lot more modular and obvious.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoBuiltin git-cat-file
Timo Hirvonen [Wed, 24 May 2006 11:08:46 +0000 (14:08 +0300)]
Builtin git-cat-file

Signed-off-by: Timo Hirvonen <tihirvon@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoapply: force matching at the beginning.
Junio C Hamano [Wed, 24 May 2006 20:19:50 +0000 (13:19 -0700)]
apply: force matching at the beginning.

When there is no leading context, the patch must match at the
beginning of preimage; otherwise there is a "patch adds these
lines while the other lines were added to the original file"
conflict.

This is the opposite of match_end fix earlier in this series.
Unlike matching at the end case, we can additionally check the
preimage line number recorded in the patch, so the change is not
symmetrical with the earlier one.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'jc/fmt-patch'
Junio C Hamano [Wed, 24 May 2006 19:21:17 +0000 (12:21 -0700)]
Merge branch 'jc/fmt-patch'

* jc/fmt-patch:
  builtin format-patch: squelch content-type for 7-bit ASCII
  CMIT_FMT_EMAIL: Q-encode Subject: and display-name part of From: fields.

18 years agoMerge branch 'be/tag'
Junio C Hamano [Wed, 24 May 2006 19:20:48 +0000 (12:20 -0700)]
Merge branch 'be/tag'

* be/tag:
  add more informative error messages to git-mktag
  remove the artificial restriction tagsize < 8kb

18 years agoMerge branch 'js/fmt-patch'
Junio C Hamano [Wed, 24 May 2006 19:19:47 +0000 (12:19 -0700)]
Merge branch 'js/fmt-patch'

This makes "git format-patch" a built-in.

* js/fmt-patch:
  git-rebase: use canonical A..B syntax to format-patch
  git-format-patch: now built-in.
  fmt-patch: Support --attach
  fmt-patch: understand old <his> notation
  Teach fmt-patch about --keep-subject
  Teach fmt-patch about --numbered
  fmt-patch: implement -o <dir>
  fmt-patch: output file names to stdout
  Teach fmt-patch to write individual files.
  Use RFC2822 dates from "git fmt-patch".
  git-fmt-patch: thinkofix to show [PATCH] properly.
  rename internal format-patch wip
  Minor tweak on subject line in --pretty=email
  Tentative built-in format-patch.

18 years agoMerge branch 'jc/builtin-n-tar-tree'
Junio C Hamano [Wed, 24 May 2006 18:24:50 +0000 (11:24 -0700)]
Merge branch 'jc/builtin-n-tar-tree'

This pulls in "make many commands built-in" branches.

* jc/builtin-n-tar-tree:
  built-in tar-tree and remote tar-tree
  Builtin git-diff-files, git-diff-index, git-diff-stages, and git-diff-tree.
  Builtin git-show-branch.
  Builtin git-apply.
  Builtin git-commit-tree.
  Builtin git-read-tree.
  Builtin git-tar-tree.
  Builtin git-ls-tree.
  Builtin git-ls-files.

18 years agoMerge branch 'jc/fetch-sorted'
Junio C Hamano [Wed, 24 May 2006 18:04:55 +0000 (11:04 -0700)]
Merge branch 'jc/fetch-sorted'

* jc/fetch-sorted:
  fetch-pack: output refs in the order they were given on the command line.

18 years agoMerge branch 'lt/dirwalk'
Junio C Hamano [Wed, 24 May 2006 18:04:16 +0000 (11:04 -0700)]
Merge branch 'lt/dirwalk'

This makes 'git add' and 'git rm' built-ins.

* lt/dirwalk:
  Add builtin "git rm" command
  Move pathspec matching from builtin-add.c into dir.c
  Prevent bogus paths from being added to the index.
  builtin-add: fix unmatched pathspec warnings.
  Remove old "git-add.sh" remnants
  builtin-add: warn on unmatched pathspecs
  Do "git add" as a builtin
  Clean up git-ls-file directory walking library interface
  libify git-ls-files directory traversal

18 years agoMerge branch 'master' into next
Junio C Hamano [Wed, 24 May 2006 10:32:31 +0000 (03:32 -0700)]
Merge branch 'master' into next

* master:
  --summary output should print immediately after stats.
  git-svn: ignore expansion of svn:keywords
  git-svn: starting a 1.1.0-pre development version
  cvsimport: set up commit environment in perl instead of using env

18 years agoAdd a test-case for git-apply trying to add an ending line
Catalin Marinas [Tue, 23 May 2006 21:48:36 +0000 (22:48 +0100)]
Add a test-case for git-apply trying to add an ending line

git-apply adding an ending line doesn't seem to fail if the ending line is
already present in the patched file.

Signed-off-by: Catalin Marinas <catalin.marinas@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years ago--summary output should print immediately after stats.
Sean [Tue, 23 May 2006 09:34:23 +0000 (05:34 -0400)]
--summary output should print immediately after stats.

Currently the summary is displayed after the patch.  Fix this so
that the output order is stat-summary-patch.  As a consequence of
the way this is coded, the --summary option will only actually
display summary data if combined with either the --stat or
--patch-with-stat option.

Signed-off-by: Sean Estabrooks <seanlkml@sympatico.ca>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoapply: treat EOF as proper context.
Linus Torvalds [Wed, 24 May 2006 02:08:01 +0000 (19:08 -0700)]
apply: treat EOF as proper context.

Catalin noticed that we do not treat end-of-file condition shown
in the patch text as the patch context.  This causes a patch
that appends at the end of the file to cleanly apply even if
something else has been appended to the file.  If this happened
in the middle, we would refuse by saying that the file has
conflicting modifications.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: ignore expansion of svn:keywords
Eric Wong [Wed, 24 May 2006 02:23:41 +0000 (19:23 -0700)]
git-svn: ignore expansion of svn:keywords

Unlike my earlier test patch, this also checks svn:eol-style and
makes sure it's applied to working copy updates.  This is
definitely more correct than my original attempt at killing
keyword expansions, but I still haven't tested it enough to
know.  Feedback would be much appreciated.

Also changed assert_svn_wc_clean() to only work on the svn
working copy.  This requires a separate call to assert_tree() to
check wc integrity against git in preparation for another change
I'm planning.

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agogit-svn: starting a 1.1.0-pre development version
Eric Wong [Wed, 24 May 2006 02:23:40 +0000 (19:23 -0700)]
git-svn: starting a 1.1.0-pre development version

Some not-very-well-tested changes coming...

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agocvsimport: set up commit environment in perl instead of using env
Jeff King [Tue, 23 May 2006 20:59:44 +0000 (16:59 -0400)]
cvsimport: set up commit environment in perl instead of using env

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'master' into next
Junio C Hamano [Tue, 23 May 2006 23:31:05 +0000 (16:31 -0700)]
Merge branch 'master' into next

* master:
  cvsimport: do not barf on creation of an empty file.

18 years agocvsimport: do not barf on creation of an empty file.
Junio C Hamano [Tue, 23 May 2006 23:30:39 +0000 (16:30 -0700)]
cvsimport: do not barf on creation of an empty file.

When the server says "created this file whose length is empty",
we mistakenly said "oops, the server did not say a sensible
thing".  Fix it.

Spotted and fixed by Linus, acked by Martin.

Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMerge branch 'sp/reflog' into next
Junio C Hamano [Tue, 23 May 2006 22:07:47 +0000 (15:07 -0700)]
Merge branch 'sp/reflog' into next

* sp/reflog:
  Enable ref log creation in git checkout -b.
  Create/delete branch ref logs.
  Include ref log detail in commit, reset, etc.
  Change order of -m option to update-ref.
  Correct force_write bug in refs.c
  Change 'master@noon' syntax to 'master@{noon}'.
  Log ref updates made by fetch.
  Force writing ref if it doesn't exist.
  Added logs/ directory to repository layout.
  General ref log reading improvements.
  Fix ref log parsing so it works properly.
  Support 'master@2 hours ago' syntax
  Log ref updates to logs/refs/<ref>
  Convert update-ref to use ref_lock API.
  Improve abstraction of ref lock/write.

18 years agoMerge branch 'eb/mailinfo' into next
Junio C Hamano [Tue, 23 May 2006 21:57:00 +0000 (14:57 -0700)]
Merge branch 'eb/mailinfo' into next

* eb/mailinfo:
  Allow in body headers beyond the in body header prefix.
  More accurately detect header lines in read_one_header_line
  In handle_body only read a line if we don't already have one.
  Refactor commit messge handling.
  Move B and Q decoding into check header.
  Make read_one_header_line return a flag not a length.

18 years agoMerge branch 'be/tag' into next
Junio C Hamano [Tue, 23 May 2006 21:53:17 +0000 (14:53 -0700)]
Merge branch 'be/tag' into next

* be/tag:
  add more informative error messages to git-mktag
  remove the artificial restriction tagsize < 8kb

18 years agoMerge branch 'jc/builtin-n-tar-tree' into next
Junio C Hamano [Tue, 23 May 2006 21:52:45 +0000 (14:52 -0700)]
Merge branch 'jc/builtin-n-tar-tree' into next

* jc/builtin-n-tar-tree:
  Builtin git-diff-files, git-diff-index, git-diff-stages, and git-diff-tree.
  Builtin git-show-branch.
  Builtin git-apply.
  Builtin git-commit-tree.
  Builtin git-read-tree.
  Builtin git-tar-tree.
  Builtin git-ls-tree.
  Builtin git-ls-files.

18 years agoMerge branch 'jc/tartree' into jc/builtin-n-tar-tree
Junio C Hamano [Tue, 23 May 2006 21:44:31 +0000 (14:44 -0700)]
Merge branch 'jc/tartree' into jc/builtin-n-tar-tree

* jc/tartree:
  built-in tar-tree and remote tar-tree

18 years agoAllow in body headers beyond the in body header prefix.
Eric W. Biederman [Tue, 23 May 2006 19:58:36 +0000 (13:58 -0600)]
Allow in body headers beyond the in body header prefix.

- handle_from is fixed to not mangle it's input line.

- Then handle_inbody_header is allowed to look in
  the body of a commit message for additional headers
  that we haven't already seen.

This allows patches with all of the right information in
unfortunate places to be imported.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMore accurately detect header lines in read_one_header_line
Eric W. Biederman [Tue, 23 May 2006 19:53:20 +0000 (13:53 -0600)]
More accurately detect header lines in read_one_header_line

Only count lines of the form '^.*: ' and '^From ' as email
header lines.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoIn handle_body only read a line if we don't already have one.
Eric W. Biederman [Tue, 23 May 2006 19:49:00 +0000 (13:49 -0600)]
In handle_body only read a line if we don't already have one.

This prepares for detecting non-email patches that don't have
mail headers.  In which case we have already read the first
line so handle_body should not ignore it.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoRefactor commit messge handling.
Eric W. Biederman [Tue, 23 May 2006 19:47:28 +0000 (13:47 -0600)]
Refactor commit messge handling.

- Move handle_info into main so it is called once
  after everything has been parsed.  This allows the removal
  of a static variable and removes two duplicate calls.

- Move parsing of inbody headers into handle_commit.
  This means we parse the in-body headers after we have decoded
  the character set, and it removes code duplication between
  handle_multipart_one_part and handle_body.

- Change the flag indicating that we have seen an in body
  prefix header into another bit in seen.
  This is a little more general and allows the possibility of parsing
  in body headers after the body message has begun.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMove B and Q decoding into check header.
Eric W. Biederman [Tue, 23 May 2006 19:45:37 +0000 (13:45 -0600)]
Move B and Q decoding into check header.

B and Q decoding is not appropriate for in body headers, so move
it up to where we explicitly know we have a real email header.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoMake read_one_header_line return a flag not a length.
Eric W. Biederman [Tue, 23 May 2006 19:44:11 +0000 (13:44 -0600)]
Make read_one_header_line return a flag not a length.

Currently we only use the return value from read_one_header line
to tell if the line we have read is a header or not.  So make
it a flag.  This paves the way for better email detection.

Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
18 years agoadd more informative error messages to git-mktag
Björn Engelmann [Tue, 23 May 2006 18:20:09 +0000 (20:20 +0200)]
add more informative error messages to git-mktag

Signed-off-by: Björn Engelmann <BjEngelmann@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>