fast-import: Fail if a non-existant commit is used for merge
Johannes Sixt noticed during one of his own imports that fast-import
did not fail if a non-existant commit is referenced by SHA-1 value
as an argument to the 'merge' command. This allowed the user to
unknowingly create commits that would fail in fsck, as the commit
contents would not be completely reachable.
A side effect of this bug was that a frontend process could mark
any SHA-1 object (blob, tree, tag) as a parent of a merge commit.
This should also fail in fsck, as the commit is not a valid commit.
We now use the same rule as the 'from' command. If a commit is
referenced in the 'merge' command by hex formatted SHA-1 then the
SHA-1 must be a commit or a tag that can be peeled back to a commit,
the commit must already exist, and must be readable by the core Git
infrastructure code. This requirement means that the commit must
have existed prior to fast-import starting, or the commit must have
been flushed out by a prior 'checkpoint' command.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Johannes Sixt noticed that a 'reset' command applied to a branch that
is already active in the branch LRU cache can cause fast-import to
relink the same branch into the LRU cache twice. This will cause
the LRU cache to contain a cycle, making unload_one_branch run in an
infinite loop as it tries to select the oldest branch for eviction.
I have trivially fixed the problem by adding an active bit to
each branch object; this bit indicates if the branch is already
in the LRU and allows us to avoid trying to add it a second time.
Converting the pack_id field into a bitfield makes this change take
up no additional memory.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
user-manual: how to replace commits older than most recent
"Modifying" an old commit by checking it out, --amend'ing it, then
rebasing on top of it, is a slightly cumbersome technique, but I've
found it useful frequently enough to make it seem worth documenting.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
The generated user manual is rather hard to read thanks to the lack of
the css that's supposed to be included from docbook-xsl.css.
I'm totally ignorant of the toolchain; grubbing through xmlto and
related scripts, the easiest way I could find to ensure that the
generated html links to the stylesheet is by calling xsltproc directly.
Maybe there's some better way.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
This should only be set based on the capability of your
compiler/library to support c99 format specifiers. In this
case the version of gcc/newlib and indirectly the version
of Cygwin. It should probably only be set in your config.mak
file.
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
In particular, the second parameter in the call to iconv() will
cause this warning if your library declares iconv() with the
second (input buffer pointer) parameter of type const char **.
This is the old prototype, which is none-the-less used by the
current version of newlib on Cygwin. (It appears in old versions
of glibc too).
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
Fix some "comparison is always true/false" warnings.
On Cygwin the wchar_t type is an unsigned short (16-bit) int.
This results in the above warnings from the return statement in
the wcwidth() function (in particular, the expressions involving
constants with values larger than 0xffff). Simply replace the
use of wchar_t with an unsigned int, typedef-ed as ucs_char_t.
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
The function at issue being initgroups() from the <grp.h> header
file. On Cygwin, setting _XOPEN_SOURCE suppresses the definition
of initgroups(), which causes the warning while compiling daemon.c.
Signed-off-by: Ramsay Jones <ramsay@ramsay1.demon.co.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
Documentation/build-docdep.perl: Fix dependencies for included asciidoc files
Adding dependencies on included files to the generated man pages is
wrong - includes are processed by asciidoc, therefore the intermediate
Docbook XML files really depend on included files. Because of these
wrong dependencies the man pages were not rebuilt properly if the
intermediate XML files were left in the tree.
Signed-off-by: Sergey Vlasov <vsu@altlinux.ru> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-archimport: support empty summaries, put summary on a single line.
Don't fail if the summary line in an arch commit is empty. In this case,
try to use the first line in the commit message followed by an ellipsis.
In addition, if the summary is multi-line, it is joined on a single line.
http-push.c::lock_remote(): validate all remote refs.
Starting from offset 11 might have been good back when it was
only used for updating "refs/heads/*", but it is used to update
"info/refs" and "refs/tags/*" as well.
git-cvsexportcommit: don't cleanup .msg if not yet committed to cvs.
Unless the -c option is given, and the commit to cvs was successful,
.msg shouldn't be deleted to be able to run the command suggested by
git-cvsexportcommit.
See http://bugs.debian.org/412732
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
index-pack: Loop over pread until data loading is complete.
A filesystem might not be able to completely supply our pread
request in one system call, such as if we are reading data from a
network file system and the requested length is just simply huge.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Correct ordering in git-cvsimport's option documentation
A pair of commits on January 8th added option documentation (for -a,
-S and -L) in the middle of the documentation for the -A option. This
makes -A's documentation contiguous again.
Signed-off-by: Michael Poole <mdpoole@troilus.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
that's an illogical thing to do, since "git show" is defined to be a
non-revision-walking action, which means the range operator be pointless
and wrong. The fact that we happily accept it (and then _only_ show
v1.5.0, which is the positive end of the range) is quite arguably not very
logical.
We should complain, and say that you can only do "no_walk" with positive
refs. Negative object refs really don't make any sense unless you walk
the obejct list (or you're "git diff" and know about ranges explicitly).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-apply: do not fix whitespaces on context lines.
Internal function apply_line() is called to copy both context lines
and added lines to the output buffer, while possibly fixing the
whitespace breakages depending on --whitespace=strip settings.
However, it did its fix-up on both context lines and added lines.
This resulted in two symptoms:
(1) The number of lines reported to have been fixed up included
these context lines.
(2) However, the lines actually shown were limited to the added
lines that had whitespace breakages.
diff --cc: integer overflow given a 2GB-or-larger file
Few of us use git to compare or even version-control 2GB files,
but when we do, we'll want it to work.
Reading a recent patch, I noticed two lines like this:
int len = st.st_size;
Instead of "int", that should be "size_t". Otherwise, in the
non-symlink case, with 64-bit size_t, if the file's size is 2GB,
the following xmalloc will fail:
result = xmalloc(len + 1);
trying to allocate 2^64 - 2^31 + 1 bytes (assuming sign-extension
in the int-to-size_t promotion). And even if it didn't fail, the
subsequent "result[len] = 0;" would be equivalent to an unpleasant
"result[-2147483648] = 0;"
The other nearby "int"-declared size variable, sz, should also be of
type size_t, for the same reason. If sz ever wraps around and becomes
negative, xread will corrupt memory _before_ the "result" buffer.
Signed-off-by: Jim Meyering <jim@meyering.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
mailinfo: do not get confused with logical lines that are too long.
It basically considers all the continuation lines to be lines of their
own, and if the total line is bigger than what we can fit in it, we just
truncate the result rather than stop in the middle and then get confused
when we try to parse the "next" line (which is just the remainder of the
first line).
[jc: added test, and tightened boundary a bit per list discussion.]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-remote: support remotes with a dot in the name
[jc: the original from Pavel was limiting the variable names to only
fetch and url, but I loosened it to take valid variable names.]
[jc: cherry-picked from 'master', since people seem to be reinventing
this many times.]
Signed-off-by: Pavel Roskin <proski@gnu.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
diff --cc: fix display of symlink conflicts during a merge.
"git-diff-files --cc" to show conflicts during merge did not pass
the correct mode information for the working tree down, and showed
bogus combined diff.
merge-recursive: fix longstanding bug in merging symlinks
Commit 3af244ca added unlink(2) before running symlink(2) to
update the working tree with the merge result, but it was
unlinking a wrong path. This resulted in loss of the path
pointed by a symlink.
merge-index: fix longstanding bug in merging symlinks
Ancient commit e2b6a9d0 added code to pass "file modes" from
merge-index to merge-one-file, and then later commit 54dd99a1
wanted to make sure we do not end up creating a nonsense symlink
that points at a path whose name contains conflict markers.
However, nobody noticed that the code in merge-index added by e2b6a9d0 were stripping the S_IFMT bits and the code in 54dd99a1
was meaningless. This fixes it.
Allow arbitrary number of arguments to git-pack-objects
If a repository ever gets in a situation where there are too many
packs (more than 60 or so), perhaps because of frequent use of
git-fetch -k or incremental git-repack, then it becomes impossible to
fully repack the repository with git-repack -a. That command just
dies with the cryptic message
fatal: too many internal rev-list options
This message comes from git-pack-objects, which is passed one command
line option like --unpacked=pack-<SHA1>.pack for each pack file to be
repacked. However, the current code has a static limit of 64 command
line arguments and just aborts if more arguments are passed to it.
Fix this by dynamically allocating the array of command line
arguments, and doubling the size each time it overflows.
Signed-off-by: Roland Dreier <roland@digitalvampire.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
We should always avoid rewriting a built file during `make install`
if nothing has changed since `make all`. This is to help support
the typical installation process of compiling a package as yourself,
then installing it as root.
Forcing CREDITS-FILE to be always be rebuilt in the Makefile means
that CREDITS-GEN needs to check for a change and only update
CREDITS-FILE if the file content actually differs. After all,
content is king in Git.
Signed-off-by: Junio C Hamano <junkio@cox.net> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
When the blobs recorded on the index lines in the patch as pre-image
blobs are not found in the repository, "git-am" punted saying
that the index line does not record anything useful. This was not
clear enough -- the index line does have something useful but the
problem was that it was not useful in _that_ repository.
Badly formatted commits may have very long comments. This causes
git-format-patch to fail. To avoid that, truncate the filename
to a value we believe will always work.
Err out if the patch file cannot be created.
Signed-off-by: Robin Rosenberg <robin.rosenberg@dewire.com> Acked-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
The Makefile for the git-gui subproject will fail to execute if run
outside of a git.git directory, such as when building from a .tar.gz
or .tar.bz2. This is because it is looking for the credits file,
which was created but omitted from the tarball by the toplevel
Makefile.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Merge branch 'master' of git://repo.or.cz/git-gui into maint
* 'master' of git://repo.or.cz/git-gui:
git-gui: Don't crash in citool mode on initial commit.
git-gui: Remove TODO list.
git-gui: Include browser in our usage message.
git-gui: Change summary of git-gui.
git-gui: Display all authors of git-gui.
git-gui: Use mixed path for docs on Cygwin.
git-gui: Correct crash when saving options in blame mode.
git-gui: Expose the browser as a subcommand.
git-gui: Create new branches from a tag.
git-gui: Prefer version file over git-describe.
git-gui: Print version on the console.
git-gui: More consistently display the application name.
git-gui: Permit merging tags into the current branch.
git-gui: Basic version check to ensure git 1.5.0 or later is used.
git-gui: Refactor 'exec git subcmd' idiom.
Not everyone has gzcat or bzcat installed on their system, but
gunzip -c and bunzip2 -c perform the same task and are available
if the user has installed gzip support or bzip2 support.
Signed-off-by: Michael Loeffler <zvpunry@zvpunry.de> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Don't crash in citool mode on initial commit.
Attempting to use `git citool` to create an initial commit caused
git-gui to crash with a Tcl error as it tried to add the newly
born branch to the non-existant branch menu. Moving this code
to after the normal commit cleanup logic resolves the issue, as
we only have a branch menu if we are not in singlecommit mode.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
I'm apparently not very good at keeping my own TODO file current.
I its also somewhat strange to keep the TODO list as part of the
software branch, as its meta-information that is not directly
related to the code. I'm pulling the TODO list from git-gui and
moving it into a seperate branch.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Now that git-gui has been released to the public as part of Git 1.5.0
I am starting to see some work from other people beyond myself and
Paul. Consequently the copyright for git-gui is not strictly the
two of us anymore, and these others deserve to have some credit
given to them.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The Firefox browser requires that a URL use / to delimit directories.
This is instead of \, as \ gets escaped by the browser into its hex
escape code and then relative URLs are incorrectly resolved, Firefox
no longer sees the directories for what they are. Since we are
handing the browser a true URL, we better use the standard / for
directories.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Check for PRIuMAX rather than NO_C99_FORMAT in fast-import.c.
Thanks to Simon 'corecode' Schubert <corecode@fs.ei.tum.de> for
the clean-up. Defining the C99 standard PRIuMAX when necessary
replaces UM_FMT and the awkward UM10_FMT. There are no direct
C99 translations for other uses of NO_C99_FORMAT in git, alas.
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
Define UM_FMT and UM10_FMT and use in place of %ju and %10ju,
respectively. Both format as unsigned long long, so this
assumes the compiler supports long long.
Signed-off-by: Jason Riedy <jason@acm.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Solaris 8 was pre-c99, and they weren't willing to commit to
the strtoumax definition according to /usr/include/inttypes.h.
This adds NO_STRTOUMAX and NO_STRTOULL for ancient systems.
If NO_STRTOUMAX is defined, the routine in compat/strtoumax.c
will be used instead. That routine passes its arguments to
strtoull unless NO_STRTOULL is defined. If NO_STRTOULL, then
the routine uses strtoul (unsigned long).
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu> Acked-by: Shawn O Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Documentation advertises the new `--depth <n>' parameter with an equal
sign, while the usage notes (shown after `git-clone --help') do not. If I
understood git-clone's source code correctly, the version without the
equal sign is correct, which is why this patch syncs documentation to the
usage note.
Signed-off-by: Christian Schlotter <schlotter@users.sourceforge.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-gui: Correct crash when saving options in blame mode.
Martin Waitz noticed that git-gui crashed while saving the user's
options out if the application was started in blame mode. This
was caused by the do_save_config procedure invoking reshow_diff
incase the number of context lines was modified by the user.
Because we bypassed main window UI setup to enter blame mode we
did not set many of the globals which were accessed by reshow_diff,
and reading unset variables is an error in Tcl.
Aside from moving the globals to be set earlier, I also modified
reshow_diff to not invoke clear_diff if there is no path currently
in the diff viewer. This way reshow_diff does not crash when in
blame mode due to the $ui_diff command not being defined.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
* git://git.kernel.org/pub/scm/gitk/gitk:
Make gitk save and restore window pane position on Linux and Cygwin.
Make gitk save and restore the user set window position.
[PATCH] gitk: Use show-ref instead of ls-remote
[PATCH] Make gitk work reasonably well on Cygwin.
[PATCH] gitk - remove trailing whitespace from a few lines.
Change git repo-config to git config
Since `git add` is the approved porcelain for an end-user to invoke
when they want to manipulate the index, porcelain documentation
should steer the user to this command rather than the pure plumbing
update-index.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Attempt to improve git-rebase lead-in description.
It was mentioned on #git this morning that the lead-in description
of git-rebase is very confusing. Too many branch this and branch
that in a very short run of text.
This new description attempts to walk the user through the command
syntax, while also describing exactly what git-rebase is doing to
their repository.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Do not take mode bits from index after type change.
When we do not trust executable bit from lstat(2), we copied
existing ce_mode bits without checking if the filesystem object
is a regular file (which is the only thing we apply the "trust
executable bit" business) nor if the blob in the index is a
regular file (otherwise, we should do the same as registering a
new regular file, which is to default non-executable).
The 3rd branch in builtin-blame.c should also check for lacking
arguments. Running that in top dir does not trigger the problem
because the 'prefix' is NULL.
Signed-off-by: Tommi Kyntola <tommi.kyntola@ray.fi> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-merge: minor fix for no_trivial_merge_strategies.
The shell loop to determine if we should skip the trivial
in-index merge stage based on what strategy is given was not
prepared to have more than one strategy listed in the variable
$no_trivial_merge_strategies.
This does not trigger unless you use a modified git but the fix
is simple and straightforward, so let's fix it before 1.5.0.1.
Some users may find being able to browse around an arbitrary
branch to be handy, so we now expose our graphical browser
through `git gui browse <committish>`.
Yes, I'm being somewhat lazy and making the user give us
the name of the branch to browse. They can always enter
HEAD.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
When somebody else extracts git tarball inside a larger project,
'git describe' would reported the version number of that upper
level project.
Sometimes, using the consistent versioning across subdirectories
of a larger project is useful, but it may not always be the
right thing to do.
This changes the script to check ./vertion file first, and then
fall back to "git describe". This way, by default, tarball
distribution will get our own version. If the upper level wants
to use consistent versioning across its subdirectories, its
Makefile can overwrite ./version file to force whatever version
number they want to give us before descending into us.
Make gitk save and restore window pane position on Linux and Cygwin.
Subtle bugs remained on both Cygwin and Linux that caused the various
window panes to be restored in positions different than where the user
last placed them. Sergey Vlasov posed a pair of suggested fixes to this,
what is done here is slightly different. The basic fix here involves
a) explicitly remembering and restoring the sash positions for the upper
window, and b) using paneconfigure to redundantly set height and width of
other elements. This redundancy is needed as Cygwin Tcl has a nasty habit
of setting pane sizes to zero if their slaves are not configured with a
specific size, but Linux Tcl does not honor the specific size given.
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
Make gitk save and restore the user set window position.
gitk was saving widget sizes and positions when the main window was
destroyed, which is after all child widgets are destroyed. The cure
is to trap the WM_DELETE_WINDOW event before the gui is torn down. Also,
the saved geometry was captured using "winfo geometry .", rather than
"wm geometry ." Under Linux, these two return different answers and the
latter one is correct.
[jc: credit goes to Brett Schwarz for suggesting the use of "wm protocol";
I also squashed the follow-up patch to remove extraneous -0
from expressions.]
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
It used to be ls-remote on self was the only easy way to grab
the ref information. Now we have show-ref which does not
involve fork and IPC, so use it.
Signed-off-by: Junio C Hamano <junkio@cox.net> Signed-off-by: Paul Mackerras <paulus@samba.org>
The gitk gui layout was completely broken on Cygwin. If gitk was started
without previous geometry in ~/.gitk, the user could drag the window sashes
to get a useable layout. However, if ~/.gitk existed, this was not possible
at all.
The fix was to rewrite makewindow, changing the toplevel containers and
the particular geometry information saved between sessions. Numerous bugs
in both the Cygwin and the Linux Tk versions make this a delicate
balancing act: the version here works in both but many subtle variants
are competely broken in one or the other environment.
Three user visible changes result:
1 - The viewer is fully functional under Cygwin.
2 - The search bar moves from the bottom to the top of the lower left
pane. This was necessary to get around a layout problem on Cygwin.
3 - The window size and position is saved and restored between sessions.
Again, this is necessary to get around a layout problem on Cygwin.
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Junio C Hamano <junkio@cox.net> Signed-off-by: Paul Mackerras <paulus@samba.org>