Stephen Boyd [Thu, 5 Nov 2009 06:33:53 +0000 (22:33 -0800)]
t1200: Make documentation and test agree
There were some differences between t1200 and the gitcore-tutorial. Add
missing tests for manually merging two branches, and use the same
commands in both files.
Signed-off-by: Stephen Boyd <bebarino@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Stephen Boyd [Thu, 5 Nov 2009 06:33:52 +0000 (22:33 -0800)]
t1200: cleanup and modernize test style
Many parts of the tests in t1200 are run outside the test harness,
circumventing the usefulness of -v and spewing messages to stdout when
-v isn't used. Fix these problems by modernizing the test a bit.
An extra test_done has existed since commit 6a74642 (git-commit --amend:
two fixes., 2006-04-20) leading to the last 6 tests never being run.
Remove it and teach the resolve merge test about fast-forward merges.
Also fix the last test's incorrect find command and prune before
checking for unpacked objects so we remove the unreachable conflict-marked
blob.
Finally, we remove the TODO notes, because fetch, push, and clone have
their own tests since t1200 was introduced and we're not going to add
them here 4 years later.
Signed-off-by: Stephen Boyd <bebarino@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Petr Baudis [Fri, 6 Nov 2009 15:08:41 +0000 (16:08 +0100)]
gitweb: Fix blob linenr links in pathinfo mode
In pathinfo mode, we use <base href> that refers to the base location
of gitweb in order for various external media links to work well.
However, this means that for the page to refer to itself, it must
regenerate full link, and this is exactly what the blob view page
did not do for line numbers.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Nicolas Pitre [Wed, 4 Nov 2009 21:32:46 +0000 (16:32 -0500)]
pack-objects: move thread autodetection closer to relevant code
Let's keep thread stuff close together if possible. And in this case,
this even reduces the #ifdef noise, and allows for skipping the
autodetection altogether if delta search is not needed (like with a pure
clone).
Signed-off-by: Nicolas Pitre <nico@fluxnic.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Daniel Barkalow [Wed, 4 Nov 2009 02:52:35 +0000 (21:52 -0500)]
Allow curl helper to work without a local repository
It's okay to use the curl helper without a local repository, so long
as you don't use "fetch". There aren't any git programs that would try
to use it, and it doesn't make sense to try it (since there's nowhere
to write the results), but we may as well be clear.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Daniel Barkalow [Wed, 4 Nov 2009 02:38:51 +0000 (21:38 -0500)]
Require a struct remote in transport_get()
cmd_ls_remote() was calling transport_get() with a NULL remote and a
non-NULL url in the case where it was run outside a git
repository. This involved a bunch of ill-tested special
cases. Instead, simply get the struct remote for the URL with
remote_get(), which works fine outside a git repository, and can also
take global options into account.
This fixes a tiny and obscure bug where "git ls-remote" without a repo
didn't support global url.*.insteadOf, even though "git clone" and
"git ls-remote" in any repo did.
Also, enforce that all callers provide a struct remote to transport_get().
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Vietor Liu [Sat, 31 Oct 2009 06:36:03 +0000 (14:36 +0800)]
imap-send.c: fix compiler warnings for OpenSSL 1.0
The openssl/CHANGES file says:
Let the TLSv1_method() etc. functions return a 'const' SSL_METHOD
pointer and make the SSL_METHOD parameter in SSL_CTX_new,
SSL_CTX_set_ssl_version and SSL_set_ssl_method 'const'.
In older versions, unqualified pointers were used, so we unfortunately
cannot unconditionally update the type of the variable we use.
Signed-off-by: Vietor Liu <vietor@vxwo.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Fri, 30 Oct 2009 17:09:06 +0000 (10:09 -0700)]
diff --color-words: bit of clean-up
When we introduced the "word diff" mode, we could have done one of three
things:
* change fn_out_consume() to "this is called every time a line worth of
diff becomes ready from the lower-level diff routine. This function
knows two sets of helpers (one for line-oriented diff, another for word
diff), and each set has various functions to be called at certain
places (e.g. hunk header, context, ...). The function's role is to
inspect the incoming line, and dispatch appropriate helpers to produce
either line- or word- oriented diff output."
* introduce fn_out_consume_word_diff() that is "this is called every time
a line worth of diff becomes ready from the lower-level diff routine,
and here is what we do to prepare word oriented diff using that line."
without touching fn_out_consume() at all.
* Do neither of the above, and keep fn_out_consume() to "this is called
every time a line worth of diff becomes ready from the lower-level diff
routine, and here is what we do to output line oriented diff using that
line." but sprinkle a handful of 'are we in word-diff mode? if so do
this totally different thing' at random places.
This patch is to at least abstract the details of "this totally different
thing" out from the main codepath, in order to improve readability.
We can later refactor it by introducing fn_out_consume_word_diff(), taking
the second route above, but that is a separate topic.
There is not enough justification for doing this. We do not update
things in .git/branches and .git/remotes anymore, but still do read
information from there and will keep doing so.
Besides, this breaks quite a lot of tests in t55?? series.
Junio C Hamano [Sat, 31 Oct 2009 03:05:47 +0000 (20:05 -0700)]
Merge branch 'ja/fetch-doc'
* ja/fetch-doc:
Documentation/merge-options.txt: order options in alphabetical groups
Documentation/git-pull.txt: Add subtitles above included option files
Documentation/fetch-options.txt: order options alphabetically
Björn Gustavsson [Thu, 29 Oct 2009 22:08:31 +0000 (23:08 +0100)]
Teach 'git merge' and 'git pull' the option --ff-only
For convenience in scripts and aliases, add the option
--ff-only to only allow fast-forwards (and up-to-date,
despite the name).
Disallow combining --ff-only and --no-ff, since they
flatly contradict each other.
Allow all other options to be combined with --ff-only
(i.e. do not add any code to handle them specially),
including the following options:
* --strategy (one or more): As long as the chosen merge
strategy results in up-to-date or fast-forward, the
command will succeed.
* --squash: I cannot imagine why anyone would want to
squash commits only if fast-forward is possible, but I
also see no reason why it should not be allowed.
* --message: The message will always be ignored, but I see
no need to explicitly disallow providing a redundant message.
Acknowledgements: I did look at Yuval Kogman's earlier
patch (107768 in gmane), mainly as shortcut to find my
way in the code, but I did not copy anything directly.
Signed-off-by: Björn Gustavsson <bgustavsson@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King [Fri, 30 Oct 2009 19:05:52 +0000 (15:05 -0400)]
ls-files: unbreak "ls-files -i"
Commit b5227d8 changed the behavior of "ls-files" with
respect to includes, but accidentally broke the "-i" option
The original behavior was:
1. if no "-i" is given, cull all results according to --exclude*
2. if "-i" is given, show the inverse of (1)
The broken behavior was:
1. if no "-i" is given:
a. for "-o", cull results according to --exclude*
b. for index files, always show all
2. if "-i" is given:
a. for "-o", shows the inverse of (1a)
b. for index files, always show all
The fixed behavior keeps the new (1b) behavior introduced
by b5227d8, but fixes the (2b) behavior to show only ignored
files, not all files.
This patch also tweaks the documentation. The original text
was somewhat obscure in the first place, but it is also now
inaccurate (the relationship between (1b) and (2b) is not
quite a "reverse").
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
diff --color-words -U0: fix the location of hunk headers
Colored word diff without context lines firstly printed all the hunk
headers among each other and then printed the diff.
This was due to the code relying on getting at least one context line at
the end of each hunk, where the colored words would be flushed (it is
done that way to be able to ignore rewrapped lines).
Noticed by Markus Heidelberg.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Make the MSVC projects use PDB/IDB files named after the project
Instead of having all PDB files for all projects named "vc90.pdb", name them
after the respective project to make the relation more clear (and to avoid name
clashes when copying files around).
Signed-off-by: Sebastian Schuberth <sschuberth@gmail.com> Acked-by: Marius Storm-Olsen <mstormo@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When compiling with MSVC on x86-compatible, use an intrinsic for byte swapping.
In contrast to the GCC path, we do not prefer inline assembly here as it is not
supported for the x64 platform.
Signed-off-by: Sebastian Schuberth <sschuberth@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Thomas Rast [Wed, 28 Oct 2009 22:10:06 +0000 (23:10 +0100)]
describe: when failing, tell the user about options that work
Users seem to call git-describe without reading the manpage, and then
wonder why it doesn't work with unannotated tags by default.
Make a minimal effort towards seeing if there would have been
unannotated tags, and tell the user. Specifically, we say that --tags
could work if we found any unannotated tags. If not, we say that
--always would have given results.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Linus Torvalds [Tue, 27 Oct 2009 18:28:07 +0000 (11:28 -0700)]
Add '--bisect' revision machinery argument
I personally use "git bisect visualize" all the time when I bisect, but it
turns out that that is not a very flexible model. Sometimes I want to do
bisection based on all commits (no pathname limiting), but then visualize
the current bisection tree with just a few pathnames because I _suspect_
those pathnames are involved in the problem but am not totally sure about
them.
And at other times, I want to use other revision parsing logic, none of
which is available with "git bisect visualize".
So this adds "--bisect" as a revision parsing argument, and as a result it
just works with all the normal logging tools. So now I can just do
Junio C Hamano [Wed, 28 Oct 2009 18:21:46 +0000 (11:21 -0700)]
Merge branch 'maint'
* maint:
help -a: do not unnecessarily look for a repository
Do not try to remove directories when removing old links
rebase -i: more graceful handling of invalid commands
help -i: properly error out if no info viewer can be found
David Brown [Wed, 28 Oct 2009 17:13:44 +0000 (10:13 -0700)]
commit: More generous accepting of RFC-2822 footer lines.
'git commit -s' will insert a blank line before the Signed-off-by
line at the end of the message, unless this last line is a
Signed-off-by line itself. Common use has other trailing lines
at the ends of commit text, in the style of RFC2822 headers.
Be more generous in considering lines to be part of this footer.
If the last paragraph of the commit message reasonably resembles
RFC-2822 formatted lines, don't insert that blank line.
The new Signed-off-by line is still only suppressed when the
author's existing Signed-off-by is the last line of the message.
Signed-off-by: David Brown <davidb@quicinc.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
bash completion: difftool accepts the same options as diff
So complete refs, files after the double-dash and some diff options that
make sense for difftool.
Signed-off-by: Markus Heidelberg <markus.heidelberg@web.de> Acked-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
help -a: do not unnecessarily look for a repository
Although 'git help -a' actually doesn't need to be run inside a git
repository and uses no repository-specific information, it looks for a git
directory. On 'git <TAB><TAB>' the bash completion runs 'git help -a' and
unnecessary searching for a git directory can be annoying in auto-mount
environments. With this commit, 'git help' no longer searches for a
repository when run with the -a option.
Reported by Vincent Danjean through http://bugs.debian.org/539273
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Björn Gustavsson [Sat, 17 Oct 2009 09:33:38 +0000 (11:33 +0200)]
bash: complete more options for 'git rebase'
Complete all long options for 'git rebase' except --no-verify
(probably used very seldom) and the long options corresponding
to -v, -q, and -f.
Signed-off-by: Björn Gustavsson <bgustavsson@gmail.com> Acked-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Do not try to remove directories when removing old links
When building Git with MSVC on Windows, directories named after the Git alias
are created for the output files, e.g. there is a "git-merge-index" directory
next to the "git-merge-index.exe" executable in the build root. Previously,
"make all" just checked if "git-merge-index" and "git-merge-index.exe" are the
same file, and if not, tried to remove "git-merge-index". This fails in the
case of "git-merge-index" being a directory, which is why this is checked now.
Signed-off-by: Sebastian Schuberth <sschuberth@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Julian Phillips [Sun, 25 Oct 2009 21:28:12 +0000 (21:28 +0000)]
fetch: Speed up fetch of large numbers of refs
When there are large numbers of refs, calling read_ref for each ref is
inefficent (and infact downright slow) - so instead use for_each_ref
to build up a string list of all the refs that we currently have,
which significantly improves the volume.
Signed-off-by: Julian Phillips <julian@quantumfyre.co.uk> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Julian Phillips [Sun, 25 Oct 2009 21:28:11 +0000 (21:28 +0000)]
remote: Make ref_remove_duplicates faster for large numbers of refs
The ref_remove_duplicates function was very slow at dealing with very
large numbers of refs. This is because it was using a linear search
through all remaining refs to find any duplicates of the current ref.
Rewriting it to use a string list to keep track of which refs have
already been seen and removing duplicates when they are found is much
more efficient.
Signed-off-by: Julian Phillips <julian@quantumfyre.co.uk> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King [Wed, 28 Oct 2009 00:52:57 +0000 (20:52 -0400)]
add-interactive: handle deletion of empty files
Usually we show deletion as a big hunk deleting all of the
file's text. However, for files with no content, the diff
shows just the 'deleted file mode ...' line. This patch
cause "add -p" (and related commands) to recognize that line
and explicitly ask about deleting the file.
We only add the "stage this deletion" hunk for empty files,
since other files will already ask about the big content
deletion hunk. We could also change those files to simply
display "stage this deletion", but showing the actual
deleted content is probably what an interactive user wants.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jan Krüger [Tue, 27 Oct 2009 14:58:14 +0000 (15:58 +0100)]
rebase -i: more graceful handling of invalid commands
Currently, when there is an invalid command, the rest of the line is
still treated as if the command had been valid, i.e. rebase -i attempts
to produce a patch, using the next argument as a SHA1 name. If there is
no next argument or an invalid one, very confusing error messages
appear (the line was '.'; path to git-rebase-todo substituted):
Unknown command: .
fatal: ambiguous argument 'Please fix this in the file $somefile.':
unknown revision or path not in the working tree.
Use '--' to separate paths from revisions
fatal: Not a valid object name Please fix this in the file $somefile.
fatal: bad revision 'Please fix this in the file $somefile.'
Instead, verify the validity of the remaining line and error out earlier
if necessary.
Signed-off-by: Jan Krüger <jk@jk.gs> Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Jeff King [Tue, 27 Oct 2009 01:10:24 +0000 (21:10 -0400)]
gitignore: root most patterns at the top-level directory
Our gitignore doesn't use a preceding "/" to root its
patterns in the top of the repository. This means that if
you add a file or directory called "git" (for example)
inside a subdirectory, it will be erroneously ignored.
This patch was done mechanically with "s/^[^*]/\/&/" with
one exception: instead of ignoring gitk-wish, we should
gitk-git/gitk-wish (arguably, this should be done in
gitk-git/.gitignore, but because that is a subtree merge
from elsewhere, this is easier).
Acked-by: Sverre Rabbelier <srabbelier@gmail.com> Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Tue, 27 Oct 2009 23:07:41 +0000 (16:07 -0700)]
Merge git://git.bogomips.org/git-svn
* git://git.bogomips.org/git-svn:
git-svn: convert SVN 1.5+ / svnmerge.py svn:mergeinfo props to parents
git-svn: add test data for SVN 1.5+ merge, with script.
git-svn: convert SVK merge tickets to extra parents
git-svn: allow test setup script to support PERL env. var
git-svn: add test data for SVK merge, with script.
git svn: fix fetch where glob is on the top-level URL
Junio C Hamano [Tue, 27 Oct 2009 21:55:37 +0000 (14:55 -0700)]
Merge git://repo.or.cz/git-gui
* git://repo.or.cz/git-gui:
git-gui: adjust the minimum height of diff pane for shorter screen height
git-gui: fix use of uninitialized variable
git-gui: store wm state and fix wm geometry
git-gui: Ensure submodule path is quoted properly
git-gui: fix diff for partially staged submodule changes
git-gui: Update russian translation
git-gui: Limit display to a maximum number of files
git-gui: remove warning when deleting correctly merged remote branch
git-gui: Added Greek translation & glossary
git-gui: display summary when showing diff of a submodule
Many build scripts use `git describe` to produce a version number based on
the description of HEAD (on which the work tree is based) + saying that if
the build contains uncommitted changes. This patch helps the writing of
such scripts since `git describe --dirty` does directly the intended thing.
Three possiblities were considered while discussing this new feature:
1. Describe the work tree by default and describe HEAD only if "HEAD" is
explicitly specified
Pro: does the right thing by default (both for users and for scripts)
Pro: other git commands that works on the work tree by default
Con: breaks existing scripts used by the Linux kernel and other projects
2. Use --worktree instead of --dirty
Pro: does what it says: "git describe --worktree" describes the work tree
Con: other commands do not require a --worktree option when working
on the work tree (it often is the default mode for them)
Con: unusable with an optional value: "git describe --worktree=.mod"
is quite unintuitive.
3. Use --dirty as in this patch
Pro: makes sense to specify an optional value (what the dirty mark is)
Pro: does not have any of the big cons of previous alternatives
* does not break scripts
* is not inconsistent with other git commands
This patch takes the third approach.
Signed-off-by: Jean Privat <jean@pryen.org> Acked-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Vietor Liu [Fri, 16 Oct 2009 09:41:26 +0000 (17:41 +0800)]
git-gui: adjust the minimum height of diff pane for shorter screen height
When the main window is maximized, if the screen height is shorter (e.g.
Netbook screen 1024x600), both the partial commit pane and the status bar
are hidden. The diff pane is resizable, so that it can use less vertical
height, allowing the overall window to be shorter and still display both
the entire commit pane and status bar.
Signed-off-by: Vietor Liu <vietor@vxwo.org> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Sam Vilain [Tue, 20 Oct 2009 02:42:03 +0000 (15:42 +1300)]
git-svn: convert SVN 1.5+ / svnmerge.py svn:mergeinfo props to parents
This feature is long overdue; convert SVN's merge representation to git's
as revisions are imported. This works by converting the list of revisions
in each line of the svn:mergeinfo into git revision ranges, and then
checking the latest of each of these revision ranges for A) being new and
B) now being completely merged.
Signed-off-by: Sam Vilain <sam.vilain@catalyst.net.nz> Acked-by: Eric Wong <normalperson@yhbt.net>
Sam Vilain [Tue, 20 Oct 2009 02:42:01 +0000 (15:42 +1300)]
git-svn: convert SVK merge tickets to extra parents
SVK is a simple case to start with, as its idea of merge parents
matches git's one. When a svk:merge ticket is encountered, check each
of the listed merged revisions to see if they are in the history of
this commit; if not, then we have encountered a merge - record it.
[ew: minor formatting cleanups]
Signed-off-by: Sam Vilain <sam.vilain@catalyst.net.nz> Acked-by: Eric Wong <normalperson@yhbt.net>
Eric Wong [Fri, 23 Oct 2009 06:39:04 +0000 (23:39 -0700)]
git svn: fix fetch where glob is on the top-level URL
In cases where the top-level URL we're tracking is the path we
glob against, we can once again track odd repositories that keep
branches/tags at the top level. This regression was introduced
in commit 6f5748e14cc5bb0a836b649fb8e2d6a5eb166f1d.
Thanks to Daniel Cordero for the original bug report and
bisection.
Jeff King [Sun, 25 Oct 2009 19:15:22 +0000 (15:15 -0400)]
push: always load default config
This is needed because we want to use the
advice.pushnonfastforward variable.
Previously, we would load the config on demand only when we
needed to look at push.default. Which meant that "git push"
would load it, but "git push remote" would not, leading to
differing behavior.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Mon, 26 Oct 2009 01:40:20 +0000 (18:40 -0700)]
Merge branch 'jc/receive-pack-auto'
* jc/receive-pack-auto:
receive-pack: run "gc --auto --quiet" and optionally "update-server-info"
gc --auto --quiet: make the notice a bit less verboase
Andreas Schwab [Sat, 24 Oct 2009 13:06:57 +0000 (15:06 +0200)]
Work around option parsing bug in the busybox tar implementation
The first argument of the tar command is interpreted as a bundle of
letters specifying the mode of operation and additional options, with
any option arguments taken from subsequent words on the command line
as needed. The implementation of tar in busybox treats this bundle
as if preceded by a dash and then parses it by getopt rules, which
mishandles 'tar xfo -'. Use 'tar xof -' instead to work this around.
Signed-off-by: Andreas Schwab <schwab@linux-m68k.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We don't use crypto, but rather require libeay32 and
ssleay32. handle it in both the Makefile msvc linker
script, and the buildsystem generator.
Signed-off-by: Marius Storm-Olsen <mstormo@gmail.com> Signed-off-by: Erik Faye-Lund <kusmabite@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
René Scharfe [Sat, 17 Oct 2009 21:04:19 +0000 (23:04 +0200)]
Implement wrap format %w() as if it is a mode switch
I always considered line wrapping to be more similar to a colour, i.e. a
state that one can change and that is applied to all following text until
the next state change, except that it's always reset at the end of the
format string.
Here's a patch to implement this behaviour, using Dscho's
strbuf_add_wrapped_text()
Junio C Hamano [Mon, 19 Oct 2009 06:40:35 +0000 (23:40 -0700)]
Teach --wrap to only indent without wrapping
When a zero or negative width is given to "shortlog -w<width>,<in1>,<in2>"
and --format=%[wrap(w,in1,in2)...%], just indent the text by in1 without
wrapping.
Thomas Rast [Thu, 22 Oct 2009 08:19:06 +0000 (10:19 +0200)]
Quote ' as \(aq in manpages
The docbook/xmlto toolchain insists on quoting ' as \'. This does
achieve the quoting goal, but modern 'man' implementations turn the
apostrophe into a unicode "proper" apostrophe (given the right
circumstances), breaking code examples in many of our manpages.
Quote them as \(aq instead, which is an "apostrophe quote" as per the
groff_char manpage.
Unfortunately, as Anders Kaseorg kindly pointed out, this is not
portable beyond groff, so we add an extra Makefile variable GNU_ROFF
which you need to enable to get the new quoting.
Thanks also to Miklos Vajna for documentation.
Signed-off-by: Thomas Rast <trast@student.ethz.ch> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio C Hamano [Thu, 22 Oct 2009 06:06:14 +0000 (23:06 -0700)]
Fix incorrect error check while reading deflated pack data
The loop in get_size_from_delta() feeds a deflated delta data from the
pack stream _until_ we get inflated result of 20 bytes[*] or we reach the
end of stream.
Side note. This magic number 20 does not have anything to do with the
size of the hash we use, but comes from 1a3b55c (reduce delta head
inflated size, 2006-10-18).
The loop reads like this:
do {
in = use_pack();
stream.next_in = in;
st = git_inflate(&stream, Z_FINISH);
curpos += stream.next_in - in;
} while ((st == Z_OK || st == Z_BUF_ERROR) &&
stream.total_out < sizeof(delta_head));
This git_inflate() can return:
- Z_STREAM_END, if use_pack() fed it enough input and the delta itself
was smaller than 20 bytes;
- Z_OK, when some progress has been made;
- Z_BUF_ERROR, if no progress is possible, because we either ran out of
input (due to corrupt pack), or we ran out of output before we saw the
end of the stream.
The fix b3118bd (sha1_file: Fix infinite loop when pack is corrupted,
2009-10-14) attempted was against a corruption that appears to be a valid
stream that produces a result larger than the output buffer, but we are
not even trying to read the stream to the end in this loop. If avail_out
becomes zero, total_out will be the same as sizeof(delta_head) so the loop
will terminate without the "fix". There is no fix from b3118bd needed for
this loop, in other words.
The loop in unpack_compressed_entry() is quite a different story. It
feeds a deflated stream (either delta or base) and allows the stream to
produce output up to what we expect but no more.
do {
in = use_pack();
stream.next_in = in;
st = git_inflate(&stream, Z_FINISH);
curpos += stream.next_in - in;
} while (st == Z_OK || st == Z_BUF_ERROR)
This _does_ risk falling into an endless interation, as we can exhaust
avail_out if the length we expect is smaller than what the stream wants to
produce (due to pack corruption). In such a case, avail_out will become
zero and inflate() will return Z_BUF_ERROR, while avail_in may (or may
not) be zero.
But this is not a right fix:
do {
in = use_pack();
stream.next_in = in;
st = git_inflate(&stream, Z_FINISH);
+ if (st == Z_BUF_ERROR && (stream.avail_in || !stream.avail_out)
+ break; /* wants more input??? */
curpos += stream.next_in - in;
} while (st == Z_OK || st == Z_BUF_ERROR)
as Z_BUF_ERROR from inflate() may be telling us that avail_in has also run
out before reading the end of stream marker. In such a case, both avail_in
and avail_out would be zero, and the loop should iterate to allow the end
of stream marker to be seen by inflate from the input stream.
The right fix for this loop is likely to be to increment the initial
avail_out by one (we allocate one extra byte to terminate it with NUL
anyway, so there is no risk to overrun the buffer), and break out if we
see that avail_out has become zero, in order to detect that the stream
wants to produce more than what we expect. After the loop, we have a
check that exactly tests this condition:
So here is a patch (without my previous botched attempts) to fix this
issue. The first hunk reverts the corresponding hunk from b3118bd, and
the second hunk is the same fix proposed earlier.
projects / git.git / log