This reverts 6b7d25d97b commit.
It turns out that the new algorithm has a really bad corner
case, that literally spends minutes for inputs that takes less
than a quater seconds to delta with the old algorithm. The
resulting delta is 50% smaller which is admirable, but the
performance degradation is simply unacceptable for unconditional
use.
Some example cases are these blobs in Linux 2.6 repository:
4917ec509720a42846d513addc11cbd25e0e3c4f
9af06ba723df75fed49f7ccae5b6c9c34bc5115f
dfc9cd58dc065d17030d875d3fea6e7862ede143
Signed-off-by: Junio C Hamano <junkio@cox.net>
This lowers the default merge threshold score to 75% from
earlier 80%. The break threshold stays the same at 50% for now,
but we might want to revisit it (and the rename detection limit
as well).
* break score: this much edit (both insertion of new material
and deletion of old material) needs to be there in the file
before we consider this _might_ be a rewrite and break the
filepair.
* merge score: after a filepair is broken by the above criteria
and goes through rename detection, if their pieces did not
match with other files as rename/copy, we merge them back
into one as if nothing happened. If the filepair had at
least this much deletion of old material, however, we say
this is completely rewritten with dissimilarity index X% when
we do so.
The updated delta code by Nico is so good that what we earlier
thought to be complete rewrite now reuses a lot more from the
source material (reducing the counted "delete"), so this
adjustment is needed to keep the perceived behaviour similar to
what we had earlier.
Signed-off-by: Junio C Hamano <junkio@cox.net>
The previous one wrongly coalesced a span with the next one
even though the span being added does not reach it.
Signed-off-by: Junio C Hamano <junkio@cox.net>
With the finer grained delta algorithm, count-delta algorithm
started overcounting copied source material, since the new delta
output tends to reuse the same source range more than once and
more aggressively. This broke an earlier assumption that the
number of bytes copied out from the source buffer is a good
approximation how much source material is actually remaining in
the result.
This uses fairly inefficient algorithm to keep track of ranges
of source material that are actually copied out to the
destination buffer. With this tweak, the obvious rename/break
detection tests in the testsuite start to work again.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Indexing based on adler32 has a match precision based on the block size
(currently 16). Lowering the block size would produce smaller deltas
but the indexing memory and computing cost increases significantly.
For optimal delta result the indexing block size should be 3 with an
increment of 1 (instead of 16 and 16). With such low params the adler32
becomes a clear overhead increasing the time for git-repack by a factor
of 3. And with such small blocks the adler 32 is not very useful as the
whole of the block bits can be used directly.
This patch replaces the adler32 with an open coded index value based on
3 characters directly. This gives sufficient bits for hashing and
allows for optimal delta with reasonable CPU cycles.
The resulting packs are 6% smaller on average. The increase in CPU time
is about 25%. But this cost is now hidden by the delta reuse patch
while the saving on data transfers is always there.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Testing for realloc and size limit can be done with only one test per
loop. Make it so and fix a theoretical off-by-one comparison error in
the process.
The output buffer memory allocation is also bounded by max_size when
specified.
Finally make some variable unsigned to allow the handling of files up to
4GB in size instead of 2GB.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This change provides a 8% saving on the pack size with a 4% CPU time
increase for git-repack -a on the current git archive.
Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* fix:
git-push: Update documentation to describe the no-refspec behavior.
format-patch: pretty-print timestamp correctly.
git-add: Add support for --, documentation, and test.
* jc/perl:
cvsimport: avoid open "-|" list form for Perl 5.6
svnimport: avoid open "-|" list form for Perl 5.6
send-email: avoid open "-|" list form for Perl 5.6
rerere: avoid open "-|" list form for Perl 5.6
fmt-merge-msg: avoid open "-|" list form for Perl 5.6
* jc/pack-reuse:
pack-objects: avoid delta chains that are too long.
git-repack: allow passing a couple of flags to pack-objects.
pack-objects: finishing touches.
pack-objects: reuse data from existing packs.
* jc/nostat:
cache_name_compare() compares name and stage, nothing else.
"assume unchanged" git: documentation.
ls-files: split "show-valid-bit" into a different option.
"Assume unchanged" git: --really-refresh fix.
ls-files: debugging aid for CE_VALID changes.
"Assume unchanged" git: do not set CE_VALID with --refresh
"Assume unchanged" git
It turns out that the git-push documentation didn't describe what it
would do when not given a refspec, (not on the command line, nor in a
remotes file). This is fairly important for the user who is trying to
understand operations such as:
git clone git://something/some/where
# hack, hack, hack
git push origin
I tracked the mystery behavior down to git-send-pack and lifted the
relevant portion of its documentation up to git-push, (namely that all
refs existing both locally and remotely are updated).
Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This patch address the below:
Use monospace font to draw branch and tag name
set the font size to 13.
Make the graph column resizable. This helps to accommodate large tag names
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This fix the below bug
Junio C Hamano <junkio@cox.net> writes:
>
> It does not work in my repository, since you do not seem to
> handle branch and tag names with slashes in them. All of my
> topic branches live in directories with two-letter names
> (e.g. ak/gitview).
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
I found a paper thin man page for git-rebase, but was quite happy to
see something much more useful in the usage statement of the script
when I went there to find out how this thing worked. Here it is
cleaned up slightly and expanded a bit into the actual documentation.
Signed-off-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This adds support to git-add to allow the common -- to separate
command-line options and file names. It adds documentation and a new
git-add test case as well.
[jc: this should apply to 1.2.X maintenance series, so I reworked
git-ls-files --error-unmatch test. ]
Signed-off-by: Junio C Hamano <junkio@cox.net>
In my attempt to port git to IRIX, I broke it. Sorry.
Signed-off-by: Johannes E. Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
- Solaris 9 and up do not need -liconv, so NEEDS_LIBICONV should be set
only for S8.
- Move the declaration of the uname variables to early in the Makefile
so they can be referenced by prefix and gitexecdir variables.
- gitexecdir defaults to being same as bindir, it might as well reference
that variable.
[jc: corrupt patch, sneakily tried to remove inclusion of GIT-VERSION-FILE
I do not know why I am applying this...]
Signed-off-by: Paul Jakma <paul@quagga.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This reads data in the format a (non recursive) ls-tree outputs
and writes a tree object to the object database. The created
tree object name is output to the standard output.
For convenience, the input data does not need to be sorted; the
command sorts the input lines internally.
By request from Tommi Virtanen.
Signed-off-by: Junio C Hamano <junkio@cox.net>
* jc/ident:
Keep Porcelainish from failing by broken ident after making changes.
Delay "empty ident" errors until they really matter.
Make "empty ident" error message a bit more helpful.
It said "after fixing up, commit the result using -F .msg", but
it was not clear for new people how "fix up" should be done.
Hint "git-update-index <path>".
We could recommend "git commit -a -F .msg" instead, but I am
hesitant to give that suggestion in the blind -- you could do a
cherry-pick, revert or a merge in general in a dirty working
tree as long as local modifications do not overlap with the
merge, but using "commit -a" would include them in the result.
Signed-off-by: Junio C Hamano <junkio@cox.net>
None of these were really show-stoppers (or even triggered)
on most of the trees I've tracked.
* Node change prevention for identically named nodes. This is
a limitation of SVN, but we find the error and exit before
it's passed to SVN so we don't dirty our working tree when our
commit fails. git-svn will exit with an error code 1 if any
of the following conditions are found:
1. a directory is removed and a file of the same name of the
removed directory is created
1a. a file has its parent directory removed and the file is
takes the name of the removed parent directory::
baz/zzz => baz
2. a file is removed and a directory of the same name of the
removed file is created.
2a. a file is moved into a deeper directory that shares the
previous name of the file::
dir/$file => dir/file/$file
Since SVN cannot handle these cases, the user will have to
manually split the commit into several parts.
* --rmdir now handles nested/deep removals. If dir/a/b/c/d/e/file
is removed, and everything else is in the dir/ hierarchy is
otherwise empty, then dir/ will be deleted when file is deleted
from svn and --rmdir specified.
* Always assert that we have written the tree we want to write
on commits. This helped me find several bugs in the symlink
handling code (which as been fixed).
* Several symlink handling fixes. We now refuse to set
permissions on symlinks. We also always unlink a file
if we're going to overwrite it.
* Apply changes in a pre-determined order, so we always have
rename from locations handy before we delete them.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
XML::Simple was originally required back when I made svn-arch-mirror
because I needed to explictly track renames with Arch. Then I carried
it over to git-svn because I was afraid somebody could commit an svn
log message that could throw off a non-XML log parser. Then I noticed
the <n> lines column in the header. So, no more XML :)
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Allow 'from..to' notation from the command line.
More liberal sha1 parsing when reading from stdin no longer requires the
sha1 to be the first character, so a leading 'commit ' string is OK.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Both of these options are passed directly to git-diff-tree when
committing to a SVN repository.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Just a typo, I doubt anybody would use (and I highly recommend not
using) this option anyways. But you never know...
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
