git/git.git
16 years ago[PATCH] Redo rename/copy detection logic.
Junio C Hamano [Tue, 24 May 2005 08:10:48 +0000 (01:10 -0700)]
[PATCH] Redo rename/copy detection logic.

Earlier implementation had a major screw-up in the memory
management area.  Rename/copy logic sometimes borrowed a pointer
to a structure without any provision for downstream to determine
which pointer is shared and which is not.  This resulted in the
later clean-up code to sometimes double free such structure,
resulting in a segfault.  This made -M and -C useless.

Another problem the earlier implementation had was that it
reordered the patches, and forced the logic to differentiate
renames and copies to depend on that particular order.  This
problem was fixed by teaching rename/copy detection logic not to
do any reordering, and rename-copy differentiator not to depend
on the order of the patches.  The diffs will leave rename/copy
detector in the same destination path order as the patch that
was fed into it.  Some test vectors have been reordered to
accommodate this change.

It also adds a sanity check logic to the human-readable diff-raw
output to detect paths with embedded TAB and LF characters,
which cannot be expressed with that format.  This idea came up
during a discussion with Chris Wedgwood.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agogit-apply: more consistency checks on gitdiff filenames
Linus Torvalds [Tue, 24 May 2005 02:54:55 +0000 (19:54 -0700)]
git-apply: more consistency checks on gitdiff filenames

There's some duplication of filenames when doing filename operations
(creates, deletes, renames and copies), and this makes us verify that
the pathnames match when they should.

16 years ago[PATCH] adjust git-deltafy-script to the new diff-tree output format
Nicolas Pitre [Tue, 24 May 2005 01:58:00 +0000 (21:58 -0400)]
[PATCH] adjust git-deltafy-script to the new diff-tree output format

Also prevent 'sort' from sorting on the sha1 which was screwing the
history listing.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Update git-diff-cache documentation.
Junio C Hamano [Tue, 24 May 2005 01:20:39 +0000 (18:20 -0700)]
[PATCH] Update git-diff-cache documentation.

The recent diff updates gave diff-cache the same ability to
filter paths, which was not properly documented.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Fix diff-pruning logic which was running prune too early.
Junio C Hamano [Tue, 24 May 2005 01:14:03 +0000 (18:14 -0700)]
[PATCH] Fix diff-pruning logic which was running prune too early.

For later stages to reorder patches, pruning logic and rename detection
logic should not decide which delete to discard (because another entry
said it will take over the file as a rename) until the very end.

Also fix some tests that were assuming the earlier "last one is rename
or keep everything else is copy" semantics of diff-raw format, which no
longer is true.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agogit-apply: start using the index file information.
Linus Torvalds [Tue, 24 May 2005 02:13:55 +0000 (19:13 -0700)]
git-apply: start using the index file information.

Right now we only use it to figure out what the filename might
be when that is ambiguous, but we'll get there..

16 years agogit-apply: if no input files specified, apply stdin
Linus Torvalds [Mon, 23 May 2005 23:42:21 +0000 (16:42 -0700)]
git-apply: if no input files specified, apply stdin

This makes it act more like a traditional UNIX thing (eg "cat").

16 years agodiff-tree: don't write headers if the diff queue is empty
Linus Torvalds [Mon, 23 May 2005 23:37:47 +0000 (16:37 -0700)]
diff-tree: don't write headers if the diff queue is empty

This is not a pickaxe-specific thing, we do this regardless of
what has pruned down the diff queue.

16 years agogit-apply: unknown modes are zero, not -1
Linus Torvalds [Mon, 23 May 2005 23:32:19 +0000 (16:32 -0700)]
git-apply: unknown modes are zero, not -1

16 years ago[PATCH] diff-raw format update take #2.
Junio C Hamano [Mon, 23 May 2005 21:55:33 +0000 (14:55 -0700)]
[PATCH] diff-raw format update take #2.

This changes the diff-raw format again, following the mailing
list discussion.  The new format explicitly expresses which one
is a rename and which one is a copy.

The documentation and tests are updated to match this change.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agogit-apply: parse the diff headers (both traditional and new)
Linus Torvalds [Mon, 23 May 2005 23:09:09 +0000 (16:09 -0700)]
git-apply: parse the diff headers (both traditional and new)

.. and print out the information. This sets up all the pathname
information, and whether it's a new file, deleted file, rename,
copy or whatever.

It's slowly getting to the point where it all comes together,
and we can actually apply all the information that we've gathered.

16 years agogit-apply: improve error detection and messages
Linus Torvalds [Mon, 23 May 2005 21:38:49 +0000 (14:38 -0700)]
git-apply: improve error detection and messages

In particular, give line numbers when detecting corrupt patches.
This makes the tool a lot more friendly (indeed, much more so
than regular "patch", I think).

16 years agogit-apply: bad patch fragments are fatal
Linus Torvalds [Mon, 23 May 2005 19:31:59 +0000 (12:31 -0700)]
git-apply: bad patch fragments are fatal

Don't just stop at them and look for the next header. Die,
die, die!

16 years ago[PATCH] NUL terminate diff-tree header lines under -z.
Junio C Hamano [Mon, 23 May 2005 19:07:55 +0000 (12:07 -0700)]
[PATCH] NUL terminate diff-tree header lines under -z.

Thomas Glanzmann noticed that diff-tree -z HEAD piped to
diff-helper -z did not work.  Since diff-helper -z expects NUL
terminated lines, we should generate such.

The output side of the diff-helper should always be using '\n'
termination; earlier it used the same line_termination used for
the input side, which was a mistake.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Performance fix for pickaxe.
Junio C Hamano [Mon, 23 May 2005 07:25:15 +0000 (00:25 -0700)]
[PATCH] Performance fix for pickaxe.

The pickaxe was expanding the blobs and searching in them even
when it should have already known that both sides are the same.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Rename/copy detection fix.
Junio C Hamano [Mon, 23 May 2005 04:26:09 +0000 (21:26 -0700)]
[PATCH] Rename/copy detection fix.

The rename/copy detection logic in earlier round was only good
enough to show patch output and discussion on the mailing list
about the diff-raw format updates revealed many problems with
it.  This patch fixes all the ones known to me, without making
things I want to do later impossible, mostly related to patch
reordering.

 (1) Earlier rename/copy detector determined which one is rename
     and which one is copy too early, which made it impossible
     to later introduce diffcore transformers to reorder
     patches.  This patch fixes it by moving that logic to the
     very end of the processing.

 (2) Earlier output routine diff_flush() was pruning all the
     "no-change" entries indiscriminatingly.  This was done due
     to my false assumption that one of the requirements in the
     diff-raw output was not to show such an entry (which
     resulted in my incorrect comment about "diff-helper never
     being able to be equivalent to built-in diff driver").  My
     special thanks go to Linus for correcting me about this.
     When we produce diff-raw output, for the downstream to be
     able to tell renames from copies, sometimes it _is_
     necessary to output "no-change" entries, and this patch
     adds diffcore_prune() function for doing it.

 (3) Earlier diff_filepair structure was trying to be not too
     specific about rename/copy operations, but the purpose of
     the structure was to record one or two paths, which _was_
     indeed about rename/copy.  This patch discards xfrm_msg
     field which was trying to be generic for this wrong reason,
     and introduces a couple of fields (rename_score and
     rename_rank) that are explicitly specific to rename/copy
     logic.  One thing to note is that the information in a
     single diff_filepair structure _still_ does not distinguish
     renames from copies, and it is deliberately so.  This is to
     allow patches to be reordered in later stages.

 (4) This patch also adds some tests about diff-raw format
     output and makes sure that necessary "no-change" entries
     appear on the output.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Be careful with symlinks when detecting renames and copies.
Junio C Hamano [Mon, 23 May 2005 04:24:49 +0000 (21:24 -0700)]
[PATCH] Be careful with symlinks when detecting renames and copies.

Earlier round was not treating symbolic links carefully enough,
and would have produced diff output that renamed/copied then
edited the contents of a symbolic link, which made no practical
sense.  Change it to detect only pure renames.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agoStart implementing "git-apply"
Linus Torvalds [Mon, 23 May 2005 17:52:17 +0000 (10:52 -0700)]
Start implementing "git-apply"

This applies git patches (and old-style unified diffs)
in the index, rather than doing it in the working directory.

That allows for a lot more flexibility, and means that if a
patch fails, we aren't going to mess up the working directory.

NOTE! This is just the first cut at it, and right now it only
parses the incoming patch, it doesn't actually apply it yet.

16 years agoDon't care about st_dev in the index file
Linus Torvalds [Sun, 22 May 2005 22:08:15 +0000 (15:08 -0700)]
Don't care about st_dev in the index file

Thomas Glanzmann points out that it doesn't work well with different
clients accessing the repository over NFS - they have different views
on what the "device" for the filesystem is.

Of course, other filesystems may not even have stable inode numbers.
But we don't care. At least for now.

16 years agoSome more sparse warning fixes
Linus Torvalds [Sun, 22 May 2005 21:33:43 +0000 (14:33 -0700)]
Some more sparse warning fixes

Proper function declarations and NULL pointer usage.

16 years agoFix up git-fsck-cache documentation
Linus Torvalds [Sun, 22 May 2005 21:32:26 +0000 (14:32 -0700)]
Fix up git-fsck-cache documentation

Notably, the "--unreachable" flag no longer depends on specified heads,
and we should document what happens if no heads are given.

16 years agoInclude file cleanups..
Linus Torvalds [Sun, 22 May 2005 18:54:17 +0000 (11:54 -0700)]
Include file cleanups..

Add <limits.h> to the include files handled by "cache.h", and remove
extraneous #include directives from various .c files. The rule is that
"cache.h" gets all the basic stuff, so that we'll have as few system
dependencies as possible.

16 years ago[PATCH] Makefile: Solaris fix: call $(MAKE) instead of make for subdirectories
Thomas Glanzmann [Sun, 22 May 2005 18:27:28 +0000 (20:27 +0200)]
[PATCH] Makefile: Solaris fix: call $(MAKE) instead of make for subdirectories

Signed-off-by: Thomas Glanzmann <sithglan@stud.uni-erlangen.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Docs - delta object
David Greaves [Sun, 22 May 2005 17:44:17 +0000 (18:44 +0100)]
[PATCH] Docs - delta object

Added delta documentation

Signed-off-by: David Greaves <david@dgreaves.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Docs - tag object, git- prefix and s/changeset/commit/g
David Greaves [Sun, 22 May 2005 17:44:17 +0000 (18:44 +0100)]
[PATCH] Docs - tag object, git- prefix and s/changeset/commit/g

Add docs for tag type
Rename commands to have git- prefix
Rename changeset to commit throughout

Signed-off-by: David Greaves <david@dgreaves.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Docs - include README in git.txt
David Greaves [Sun, 22 May 2005 17:44:16 +0000 (18:44 +0100)]
[PATCH] Docs - include README in git.txt

Include the README in the git.txt

Signed-off-by: David Greaves <david@dgreaves.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Docs - asciidoc changes
David Greaves [Sun, 22 May 2005 17:44:16 +0000 (18:44 +0100)]
[PATCH] Docs - asciidoc changes

Whitespace and asciidoc formatting changes only in preparation for
content changes.

Signed-off-by: David Greaves <david@dgreaves.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Docs - Makefile update
David Greaves [Sun, 22 May 2005 17:44:15 +0000 (18:44 +0100)]
[PATCH] Docs - Makefile update

A Makefile that works just fine when the 6 character patch is applied
to asciidoc

Signed-off-by: David Greaves <david@dgreaves.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agoSplit up git-pull-script into separate "fetch" and "merge" phases.
Linus Torvalds [Sun, 22 May 2005 18:03:24 +0000 (11:03 -0700)]
Split up git-pull-script into separate "fetch" and "merge" phases.

This allows you to just fetch stuff first, inspect it, and then
resolve the merge separately if everything looks good.

16 years ago[PATCH] Diffcore updates.
Junio C Hamano [Sun, 22 May 2005 17:04:37 +0000 (10:04 -0700)]
[PATCH] Diffcore updates.

This moves the path selection logic from individual programs to a new
diffcore transformer (diff-tree still needs to have its own for
performance reasons).  Also the header printing code in diff-tree was
tweaked not to produce anything when pickaxe is in effect and there is
nothing interesting to report.  An interesting example is the following
in the GIT archive itself:

    $ git-whatchanged -p -C -S'or something in a real script'

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Add the code to set default minimum score back in.
Junio C Hamano [Sun, 22 May 2005 06:33:32 +0000 (23:33 -0700)]
[PATCH] Add the code to set default minimum score back in.

When the minimum score is specified as 0 (meaning "use default
value"), set it to the default as we are told.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Fix tweak in similarity estimator.
Junio C Hamano [Sun, 22 May 2005 08:31:28 +0000 (01:31 -0700)]
[PATCH] Fix tweak in similarity estimator.

There was a screwy math bug in the estimator that confused what
-C1 meant and what -C9 meant, only in one of the early "cheap"
check, which resulted in quite confusing behaviour.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] The diff-raw format updates.
Junio C Hamano [Sun, 22 May 2005 02:42:18 +0000 (19:42 -0700)]
[PATCH] The diff-raw format updates.

Update the diff-raw format as Linus and I discussed, except that
it does not use sequence of underscore '_' letters to express
nonexistence.  All '0' mode is used for that purpose instead.

The new diff-raw format can express rename/copy, and the earlier
restriction that -M and -C _must_ be used with the patch format
output is no longer necessary.  The patch makes -M and -C flags
independent of -p flag, so you need to say git-whatchanged -M -p
to get the diff/patch format.

Updated are both documentations and tests.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Prepare diffcore interface for diff-tree header supression.
Junio C Hamano [Sun, 22 May 2005 02:40:36 +0000 (19:40 -0700)]
[PATCH] Prepare diffcore interface for diff-tree header supression.

This does not actually supress the extra headers when pickaxe is
used, but prepares enough support for diff-tree to implement it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Tweak diffcore-rename heuristics.
Junio C Hamano [Sat, 21 May 2005 22:55:18 +0000 (15:55 -0700)]
[PATCH] Tweak diffcore-rename heuristics.

The heuristics so far was to compare file size change and xdelta
size against the average of file size before and after the
change.  This patch uses the smaller of pre- and post- change
file size instead.

It also makes a very small performance fix.  I didn't measure
it; I do not expect it to make any practical difference, but
while scanning an already sorted list, breaking out in the
middle is the right thing.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agodiff-tree: don't print multiple headers for merges when silent.
Linus Torvalds [Sat, 21 May 2005 22:42:53 +0000 (15:42 -0700)]
diff-tree: don't print multiple headers for merges when silent.

Normally we show every facet of a merge, but when we're silent,
there's little point.

16 years ago[PATCH] Constness fix for pickaxe option.
Junio C Hamano [Sat, 21 May 2005 22:02:51 +0000 (15:02 -0700)]
[PATCH] Constness fix for pickaxe option.

Constness fix for pickaxe option.

Signed-off-by: Junio C Hamano <junkio@cox.net>
16 years agodiff-tree: prettify output slightly
Linus Torvalds [Sat, 21 May 2005 18:04:19 +0000 (11:04 -0700)]
diff-tree: prettify output slightly

Make the commit explanation buffer larger, and make sure that if
we truncate it, we put a "..." marker there to visually tell people
about the truncation (tested with a much smaller buffer to make
sure it looks sane).

Also make sure that the explanation is properly line-terminated,
and add an extra newline iff we have a diff.

16 years agot/t4003-diff-rename-1: use modern options to "diff"
Linus Torvalds [Sat, 21 May 2005 17:11:37 +0000 (10:11 -0700)]
t/t4003-diff-rename-1: use modern options to "diff"

Don't do "-u0", use "--unified=0" which is accepted by modern GNU
diff versions.

16 years ago"make clean" should also clean up documentation
Linus Torvalds [Sat, 21 May 2005 16:59:37 +0000 (09:59 -0700)]
"make clean" should also clean up documentation

(Or, if somebody disagrees, we should have a "make distclean").

16 years ago[PATCH] Diff overhaul, adding the other half of copy detection.
Junio C Hamano [Sat, 21 May 2005 09:42:35 +0000 (02:42 -0700)]
[PATCH] Diff overhaul, adding the other half of copy detection.

This patch extends diff-cache and diff-files to report the
unmodified files to diff-core as well when -C (copy detection)
is in effect, so that the unmodified files can also be used as
the source candidates.  The existing test t4003 has been
extended to cover this case.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Introducing software archaeologist's tool "pickaxe".
Junio C Hamano [Sat, 21 May 2005 09:40:01 +0000 (02:40 -0700)]
[PATCH] Introducing software archaeologist's tool "pickaxe".

This steals the "pickaxe" feature from JIT and make it available
to the bare Plumbing layer.  From the command line, the user
gives a string he is intersted in.

Using the diff-core infrastructure previously introduced, it
filters the differences to limit the output only to the diffs
between <src> and <dst> where the string appears only in one but
not in the other.  For example:

 $ ./git-rev-list HEAD | ./git-diff-tree -Sdiff-tree-helper --stdin -M

would show the diffs that touch the string "diff-tree-helper".

In real software-archaeologist application, you would typically
look for a few to several lines of code and see where that code
came from.

The "pickaxe" module runs after "rename/copy detection" module,
so it even crosses the file rename boundary, as the above
example demonstrates.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Diff overhaul, adding half of copy detection.
Junio C Hamano [Sat, 21 May 2005 09:39:09 +0000 (02:39 -0700)]
[PATCH] Diff overhaul, adding half of copy detection.

This introduces the diff-core, the layer between the diff-tree
family and the external diff interface engine.  The calls to the
interface diff-tree family uses (diff_change and diff_addremove)
have not changed and will not change.  The purpose of the
diff-core layer is to provide an infrastructure to transform the
set of differences sent from the applications, before sending
them to the external diff interface.

The recently introduced rename detection code has been rewritten
to use the diff-core facility.  When applications send in
separate creates and deletes, matching ones are transformed into
a single rename-and-edit diff, and sent out to the external diff
interface as such.

This patch also enhances the rename detection code further to be
able to detect copies.  Currently this happens only as long as
copy sources appear as part of the modified files, but there
already is enough provision for callers to report unmodified
files to diff-core, so that they can be also used as copy source
candidates.  Extending the callers this way will be done in a
separate patch.

Please see and marvel at how well this works by trying out the
newly added t/t4003-diff-rename-1.sh test script.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agogit-whatchanged: allow other pagers
Linus Torvalds [Sat, 21 May 2005 16:44:16 +0000 (09:44 -0700)]
git-whatchanged: allow other pagers

(but still try to use '-S' if using less)

16 years ago[PATCH] Fix use of wc in t0000-basic
Daniel Barkalow [Sat, 21 May 2005 00:49:13 +0000 (20:49 -0400)]
[PATCH] Fix use of wc in t0000-basic

The version of wc I have (GNU textutils-2.1) puts spaces at the beginning
of lines. This patch should work for any version of wc.

Signed-off-by: Daniel Barkalow <barkalow@iabervon.org>
Acked-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] delta creation
Nicolas Pitre [Fri, 20 May 2005 21:00:23 +0000 (17:00 -0400)]
[PATCH] delta creation

This adds the ability to actually create delta objects using a new tool:
git-mkdelta.  It uses an ordered list of potential objects to deltafy
against earlier objects in the list.  A cap on the depth of delta
references can be provided as well, otherwise the default is to not have
any limit.  A limit of 0 will also undeltafy any given object.

Also provided is the beginning of a script to deltafy an entire
repository.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] delta check
Nicolas Pitre [Fri, 20 May 2005 20:59:17 +0000 (16:59 -0400)]
[PATCH] delta check

This adds knowledge of delta objects to fsck-cache and various object
parsing code.  A new switch to git-fsck-cache is provided to display the
maximum delta depth found in a repository.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] delta read
Nicolas Pitre [Fri, 20 May 2005 20:57:28 +0000 (16:57 -0400)]
[PATCH] delta read

This makes the core code aware of delta objects and undeltafy them as
needed.  The convention is to use read_sha1_file() to have
undeltafication done automatically (most users do that already so this
is transparent).

If the delta object itself has to be accessed then it must be done
through map_sha1_file() and unpack_sha1_file().

In that context mktag.c has been switched to read_sha1_file() as there
is no reason to do the full map+unpack manually.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agosparse cleanup
Linus Torvalds [Fri, 20 May 2005 18:46:10 +0000 (11:46 -0700)]
sparse cleanup

Fix various things that sparse complains about:
 - use NULL instead of 0
 - make sure we declare everything properly, or mark it static
 - use proper function declarations ("fn(void)" instead of "fn()")

Sparse is always right.

16 years ago[PATCH] Simplify "reverse-diff" logic in the diff core.
Junio C Hamano [Fri, 20 May 2005 16:54:07 +0000 (09:54 -0700)]
[PATCH] Simplify "reverse-diff" logic in the diff core.

Instead of swapping the arguments just before output, this patch
makes the swapping happen on the input side of the diff core,
when "reverse-diff" is in effect.  This greatly simplifies the
logic, but more importantly it is necessary for upcoming "copy
detection" work.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Diff-files fix with more tests.
Junio C Hamano [Fri, 20 May 2005 16:48:38 +0000 (09:48 -0700)]
[PATCH] Diff-files fix with more tests.

The same check we added earlier to update-cache to catch ENOTDIR
turns out to be missing from diff-files.  This causes a
difference not being reported when you have DF/DF (a file in a
subdirectory) in the cache and DF is a file on the filesystem.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Add tests for diff-tree
Junio C Hamano [Fri, 20 May 2005 15:32:07 +0000 (08:32 -0700)]
[PATCH] Add tests for diff-tree

This adds and reorganizes some tests for diff-tree

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agodiff-tree: use new base_name_compare() helper function
Linus Torvalds [Fri, 20 May 2005 16:11:46 +0000 (09:11 -0700)]
diff-tree: use new base_name_compare() helper function

This fixes diff-tree sorting of directories vs files (we used to
use just the regular cache_name_compare() which only works on
full file pathnames).

16 years agoIntroduce "base_name_compare()" helper function
Linus Torvalds [Fri, 20 May 2005 16:09:18 +0000 (09:09 -0700)]
Introduce "base_name_compare()" helper function

This one compares two pathnames that may be partial basenames, not
full paths. We need to get the path sorting right, since a directory
name will sort as if it had the final '/' at the end.

16 years ago[PATCH] Document -R option for git-diff-tree.
Junio C Hamano [Fri, 20 May 2005 07:26:27 +0000 (00:26 -0700)]
[PATCH] Document -R option for git-diff-tree.

Obviously we would want to document this as well.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agofsck-cache: fix segfault on nonexistent referenced object
Linus Torvalds [Fri, 20 May 2005 14:49:17 +0000 (07:49 -0700)]
fsck-cache: fix segfault on nonexistent referenced object

Noted by Frank Sorenson and Petr Baudis, patch rewritten by me.

16 years agoFix up previous commit
Linus Torvalds [Fri, 20 May 2005 05:39:27 +0000 (22:39 -0700)]
Fix up previous commit

Add '-R' flag to diff-tree, and change the test subdirectory
shell files to be executable (something that Junio couldn't
get me to do through the pure patch with my current patch
handling infrastructure).

16 years ago[PATCH] diff overhaul
Junio C Hamano [Fri, 20 May 2005 02:00:36 +0000 (19:00 -0700)]
[PATCH] diff overhaul

This cleans up the way calls are made into the diff core from diff-tree
family and diff-helper.  Earlier, these programs had "if
(generating_patch)" sprinkled all over the place, but those ugliness are
gone and handled uniformly from the diff core, even when not generating
patch format.

This also allowed diff-cache and diff-files to acquire -R
(reverse) option to generate diff in reverse.  Users of
diff-tree can swap two trees easily so I did not add -R there.

[ Linus' note: I'll add -R to "diff-tree" too, since a "commit
  diff" doesn't have another tree to switch around: the other
  tree is always the parent(s) of the commit ]

Also -M<digits-as-mantissa> suggestion made by Linus has been
implemented.

Documentation updates are also included.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agodiff-tree: add "--root" flag to show a root commit as a big creation event.
Linus Torvalds [Thu, 19 May 2005 20:44:29 +0000 (13:44 -0700)]
diff-tree: add "--root" flag to show a root commit as a big creation event.

"Let there be light"

16 years ago[PATCH] cleanup of in-code names
Alexey Nezhdanov [Thu, 19 May 2005 11:17:16 +0000 (15:17 +0400)]
[PATCH] cleanup of in-code names

Fixes all in-code names that leaved during "big name change".

Signed-off-by: Alexey Nezhdanov <snake@penza-gsm.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Declare stacked variables before the first statement.
Thomas Glanzmann [Thu, 19 May 2005 13:23:18 +0000 (15:23 +0200)]
[PATCH] Declare stacked variables before the first statement.

Signed-off-by: Thomas Glanzmann <sithglan@stud.uni-erlangen.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Detect renames in diff family.
Junio C Hamano [Thu, 19 May 2005 17:13:01 +0000 (10:13 -0700)]
[PATCH] Detect renames in diff family.

A bit of clean-up of diff.c which fixes up some comments and removes a
memory leak.

This also re-introduces the rename score debugging fprintf(), but leaves
it #idef'ed it out for normal use.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] manpage name conflict
Sebastian Kuzminsky [Thu, 19 May 2005 16:24:54 +0000 (10:24 -0600)]
[PATCH] manpage name conflict

This moves the git manpage to man7, since "git" isn't a direct command
per se.  It also does two other things:

    * Sort of works around the asciidoc 6.0.3 bug where the manpages all
      get called "git.1".  It just renames them to what they should have
      been called.

    * Fixes a cut-n-paste bug in git-diff-helper.txt that was making
      asciidoc choke.

16 years ago[PATCH] Implement git-checkout-cache -u to update stat information in the cache.
Junio C Hamano [Sun, 15 May 2005 21:23:12 +0000 (14:23 -0700)]
[PATCH] Implement git-checkout-cache -u to update stat information in the cache.

With -u flag, git-checkout-cache picks up the stat information
from newly created file and updates the cache.  This removes the
need to run git-update-cache --refresh immediately after running
git-checkout-cache.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agodiff.c: remove left-over scoring debug message
Linus Torvalds [Thu, 19 May 2005 16:20:00 +0000 (09:20 -0700)]
diff.c: remove left-over scoring debug message

It may be wonderful for rating the scoring, but it's
not appropriate for actual use ;)

16 years agogit-whatchanged: use 'less -S' on the output to make it more user friendly
Linus Torvalds [Thu, 19 May 2005 16:07:19 +0000 (09:07 -0700)]
git-whatchanged: use 'less -S' on the output to make it more user friendly

16 years ago[PATCH] Detect renames in diff family.
Junio C Hamano [Thu, 19 May 2005 10:32:35 +0000 (03:32 -0700)]
[PATCH] Detect renames in diff family.

This rips out the rename detection engine from diff-helper and moves it
to the diff core, and updates the internal calling convention used by
diff-tree family into the diff core.  In order to give the same option
name to diff-tree family as well as to diff-helper, I've changed the
earlier diff-helper '-r' option to '-M' (stands for Move; sorry but the
natural abbreviation 'r' for 'rename' is already taken for 'recursive').

Although I did a fair amount of test with the git-diff-tree with
existing rename commits in the core GIT repository, this should still be
considered beta (preview) release.  This patch depends on the diff-delta
infrastructure just committed.

This implements almost everything I wanted to see in this series of
patch, except a few minor cleanups in the calling convention into diff
core, but that will be a separate cleanup patch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Deltification library work by Nicolas Pitre.
Nicolas Pitre [Thu, 19 May 2005 14:27:14 +0000 (10:27 -0400)]
[PATCH] Deltification library work by Nicolas Pitre.

This patch adds the basic library functions to create and replay delta
information.  Also included is a test-delta utility to validate the
code.

diff-delta was based on LibXDiff written by Davide Libenzi

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] A test case addition for strbuf regression
Junio C Hamano [Thu, 19 May 2005 06:55:07 +0000 (23:55 -0700)]
[PATCH] A test case addition for strbuf regression

This test would have caught the strbuf eof condition gotcha,
hopefully fixed with my previous patch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] fix strbuf take #2
Junio C Hamano [Thu, 19 May 2005 06:34:03 +0000 (23:34 -0700)]
[PATCH] fix strbuf take #2

I just remembered why I placed that bogus "sb->len ==0 implies
sb->eof" condition there.  We need at least something like this
to catch the normal EOF (that is, line termination immediately
followed by EOF) case.  "if (feof(fp))" fires when we have
already read the eof, not when we are about read it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agomerge-base: use the new lookup_commit_reference() helper function
Linus Torvalds [Wed, 18 May 2005 23:16:51 +0000 (16:16 -0700)]
merge-base: use the new lookup_commit_reference() helper function

This allows you to use tags for merge bases.

16 years agocommit: add "lookup_commit_reference()" helper function
Linus Torvalds [Wed, 18 May 2005 23:14:22 +0000 (16:14 -0700)]
commit: add "lookup_commit_reference()" helper function

It's pretty much the same as "lookup_commit()", but it will take
tags too, and look up the commit (if any) associated with them.

16 years ago[PATCH] fix show_date() for positive timezones
Nicolas Pitre [Wed, 18 May 2005 21:11:07 +0000 (17:11 -0400)]
[PATCH] fix show_date() for positive timezones

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agodiff-tree: don't match non-directories as partial pathnames
Linus Torvalds [Wed, 18 May 2005 21:17:22 +0000 (14:17 -0700)]
diff-tree: don't match non-directories as partial pathnames

This normally doesn't matter, but if you have a filename that is
sometimes a directory and sometimes a regular file (or symlink),
we don't want the regular file case to trigger a "partial match".

16 years agodiff-tree: fix "whole sub-tree disappeared or appeared" case
Linus Torvalds [Wed, 18 May 2005 21:07:42 +0000 (14:07 -0700)]
diff-tree: fix "whole sub-tree disappeared or appeared" case

We still need to check which part of the sub-tree is interesting.

16 years agodiff-tree: fix up comparison of "interesting" sub-trees
Linus Torvalds [Wed, 18 May 2005 20:50:24 +0000 (13:50 -0700)]
diff-tree: fix up comparison of "interesting" sub-trees

We used to trigger the "interesting subdirectory" check for any
matching name that started with the same character series, regardless
of whether it had the matching slash or not.

16 years agodiff-tree: show hex sha1 of the single-commit argument case correctly.
Linus Torvalds [Wed, 18 May 2005 20:43:58 +0000 (13:43 -0700)]
diff-tree: show hex sha1 of the single-commit argument case correctly.

We can't just do the "sha1_to_hex()" thing directly, since the
buffer in question will be overwritten by the name of the parent.

So teach diff_tree_commit() to generate the proper hex name itself.

16 years agodiff-tree: fix and extend argument parsing
Linus Torvalds [Wed, 18 May 2005 20:10:17 +0000 (13:10 -0700)]
diff-tree: fix and extend argument parsing

We use "--" to mark end of command line switches, not "-". Also,
allow more flexibility in the passed-in sha1 names, in that a
single sha1 uses the "commit-diff" logic that compares against
its parent(s).

16 years agodiff-tree: clean up diff_tree_stdin() function
Linus Torvalds [Wed, 18 May 2005 20:06:47 +0000 (13:06 -0700)]
diff-tree: clean up diff_tree_stdin() function

Split it into the "one commit" vs "two trees" case, since we'll
want to use the "one commit" case for other things too.

16 years agodiff-helper: pass unrecognized lines through unmodified
Linus Torvalds [Wed, 18 May 2005 18:33:46 +0000 (11:33 -0700)]
diff-helper: pass unrecognized lines through unmodified

(and flush any pending renames)

16 years agostrbuf: allow zero-length lines
Linus Torvalds [Wed, 18 May 2005 18:33:06 +0000 (11:33 -0700)]
strbuf: allow zero-length lines

They aren't EOF.

16 years ago[PATCH] Diff-helper update
Junio C Hamano [Wed, 18 May 2005 06:29:49 +0000 (23:29 -0700)]
[PATCH] Diff-helper update

This patch adds a framework and a stub implementation of rename
detection to diff-helper program.

The current stub code is just enough to detect pure renames in
diff-tree output and not fancier.  The plan is perhaps to use
the same delta code when Nico's delta storage patch is merged
for similarity evaluation purposes.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agofsck-cache: read the default reference information even when
Linus Torvalds [Wed, 18 May 2005 17:19:59 +0000 (10:19 -0700)]
fsck-cache: read the default reference information even when
not doing reachability analysis.

This avoids the dangling head problem, and means that just a
plain "git-fsck-cache" with no parameters will DTRT.

16 years agofsck-cache: walk the 'refs' directory if the user doesn't give any
Linus Torvalds [Wed, 18 May 2005 17:16:14 +0000 (10:16 -0700)]
fsck-cache: walk the 'refs' directory if the user doesn't give any
explicit references for reachability analysis.

We already had that as separate logic in git-prune-script, so this
is not a new special case - it's an old special case moved into
fsck, making normal usage be much simpler.

16 years ago[PATCH] Fix diff output take #4.
Junio C Hamano [Wed, 18 May 2005 16:10:47 +0000 (09:10 -0700)]
[PATCH] Fix diff output take #4.

This implements the output format suggested by Linus in
<Pine.LNX.4.58.0505161556260.18337@ppc970.osdl.org>, except the
imaginary diff option is spelled "diff --git" with double dashes as
suggested by Matthias Urlichs.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years ago[PATCH] Kill a bunch of pointer sign warnings for gcc4
Brian Gerst [Wed, 18 May 2005 12:14:09 +0000 (08:14 -0400)]
[PATCH] Kill a bunch of pointer sign warnings for gcc4

 - Raw hashes should be unsigned char.
 - String functions want signed char.
 - Hash and compress functions want unsigned char.

Signed-off By: Brian Gerst <bgerst@didntduck.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
16 years agoAdd silly "git-whatchanged" script.
Linus Torvalds [Tue, 17 May 2005 18:47:13 +0000 (11:47 -0700)]
Add silly "git-whatchanged" script.

It's a one-liner, but it's useful as documentation if nothing else.

16 years ago[PATCH 4/4] Trivial test harness fixes.
Junio C Hamano [Sun, 15 May 2005 21:21:13 +0000 (14:21 -0700)]
[PATCH 4/4] Trivial test harness fixes.

The documentation of the test harness still refer to old
numbering and also contains an obvious typo.

Also "make test" should be run after making sure we have built
all binaries, since test is designed to test the newly built
ones.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years agoRemove obsolete note about native CPU byte format
Petr Baudis [Sun, 15 May 2005 10:38:55 +0000 (12:38 +0200)]
Remove obsolete note about native CPU byte format

Comment in entry.h said that the cache contents is in the native CPU
byte format, which is really not true anymore for quite some time.

16 years agoRename some more cache-related functions
Brad Roberts [Sun, 15 May 2005 02:04:25 +0000 (19:04 -0700)]
Rename some more cache-related functions

same_name -> ce_same_name()
remove_entry_at() -> remove_cache_entry_at()

Signed-off-by: Brad Roberts <braddr@puremagic.com>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years agoRename cache_match_stat() to ce_match_stat()
Brad Roberts [Sun, 15 May 2005 02:04:25 +0000 (19:04 -0700)]
Rename cache_match_stat() to ce_match_stat()

Signed-off-by: Brad Roberts <braddr@puremagic.com>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years agoCleanup the x-allocation functions
Brad Roberts [Sun, 15 May 2005 02:04:25 +0000 (19:04 -0700)]
Cleanup the x-allocation functions

xmalloc() and xrealloc() now take their sizes as size_t-type arguments.
Introduced complementary xcalloc().

Signed-off-by: Brad Roberts <braddr@puremagic.com>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years ago[PATCH 3/3] Rename git-diff-tree-helper to git-diff-helper (part 2).
Junio C Hamano [Sat, 14 May 2005 01:41:36 +0000 (18:41 -0700)]
[PATCH 3/3] Rename git-diff-tree-helper to git-diff-helper (part 2).

It used to be that diff-tree needed helper support to parse its
raw output to generate diffs, but these days git-diff-* family
produces the same output and the helper is not tied to diff-tree
anymore.  Drop "tree" from its name.

This follows the "rename only" commit to adjust the contents of
the files involved.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years ago[PATCH 2/3] Rename git-diff-tree-helper to git-diff-helper.
Junio C Hamano [Sat, 14 May 2005 01:40:54 +0000 (18:40 -0700)]
[PATCH 2/3] Rename git-diff-tree-helper to git-diff-helper.

It used to be that diff-tree needed helper support to parse its
raw output to generate diffs, but these days git-diff-* family
produces the same output and the helper is not tied to diff-tree
anymore.  Drop "tree" from its name.

This commit is done separately to record just the rename and no
file content changes. The changes in the renamed files are recorded
in the next commit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Bundled with the changes in the unrenamed files.

Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years agoThe test of the basic diff functionality
Junio C Hamano [Sat, 14 May 2005 05:52:42 +0000 (22:52 -0700)]
The test of the basic diff functionality

This test comes from "[PATCH 2/2] The core GIT tests: recent additions and
fixes" but couldn't be included before since it depended on the modechange
diff output changes.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years ago[PATCH 1/3] Update mode-change strings in diff output.
Junio C Hamano [Sat, 14 May 2005 01:40:14 +0000 (18:40 -0700)]
[PATCH 1/3] Update mode-change strings in diff output.

This updates the mode change strings to be a bit more machine
friendly.  Although this might go against the spirit of
readability for human consumption, these mode bits strings are
shown only when unusual things (mode change, file creation and
deletion) happens, output normalized for machine consumption
would be permissible.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years ago[PATCH] Add the merge test Linus called "test script from hell".
Junio C Hamano [Sat, 14 May 2005 08:46:46 +0000 (01:46 -0700)]
[PATCH] Add the merge test Linus called "test script from hell".

This is an adaptation to the test framework of a historic test
that was used before three way merge form of read-tree was
introduced, and subsequently used to validate the read-tree -m
merge works correctly.  It covers all the tricky cases known
back then and also have been updated to cover conflicting
files/directories cases since then.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years agoFixed misnamed t/t2010-update-cache-badpath.sh
Petr Baudis [Sat, 14 May 2005 23:42:31 +0000 (01:42 +0200)]
Fixed misnamed t/t2010-update-cache-badpath.sh

mv t/t2010-update-cache-badpath.sh t/t2100-update-cache-badpath.sh

16 years ago[PATCH] Test GIT environment use.
Junio C Hamano [Sat, 14 May 2005 08:45:42 +0000 (01:45 -0700)]
[PATCH] Test GIT environment use.

This test makes sure that use of deprecated environment variables still
works, using both new and old names makes new one take
precedence, and GIT_DIR and GIT_ALTERNATE_OBJECT_DIRECTORIES mechanisms
work.

Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Petr Baudis <pasky@ucw.cz>
16 years agoRename some test scripts and describe the naming convention
Petr Baudis [Sat, 14 May 2005 23:34:22 +0000 (01:34 +0200)]
Rename some test scripts and describe the naming convention

First digit: "family", e.g. the absolute basics and global stuff (0),
the basic db-side commands (read-tree, write-tree, commit-tree), the
basic working-tree-side commands (checkout-cache, update-cache), the
other basic commands (ls-files), the diff commands, the pull commands,
exporting commands, revision tree commands...

Second digit: the particular command we are testing

Third digit: (optionally) the particular switch or group of switches
we are testing

Freeform part: commandname-details

Described in the README.

mv t1000-checkout-cache.sh t2000-checkout-cache-clash.sh
mv t1001-checkout-cache.sh t2001-checkout-cache-clash.sh
mv t0200-update-cache.sh t2010-update-cache-badpath.sh
mv t0400-ls-files.sh t3000-ls-files-others.sh
mv t0500-ls-files.sh t3010-ls-files-killed.sh