We used to set the TOPOSORT flag of commits during the topological
sorting, but we can just as well use the member "indegree" for it:
indegree is now incremented by 1 in the cases where the commit used
to have the TOPOSORT flag.
This is the same behavior as before, since indegree could not be
non-zero when TOPOSORT was unset.
Incidentally, this fixes the bug in show-branch where the 8th column
was not shown: show-branch sorts the commits in topological order,
assuming that all the commit flags are available for show-branch's
private matters.
But this was not true: TOPOSORT was identical to the flag corresponding
to the 8th ref. So the flags for the 8th column were unset by the
topological sorting.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The -u option to override the remote system's path to git-upload-pack was
being ignored by "git clone"; caused by a missing call to
transport_set_option to set TRANS_OPT_UPLOADPACK. Presumably this crept in
when git-clone was converted from shell to C.
Signed-off-by: Steve Haslam <shaslam@lastminute.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The diff.external examples pass a flag to gnu-diff, but GNU diff
does not follow the GIT_EXTERNAL_DIFF interface.
Signed-off-by: Anders Melchiorsen <mail@cup.kalibalik.dk>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
A wildcard refspec is internally parsed into a refspec structure with src
and dst strings. Many parts of the code assumed that these do not include
the trailing "/*" when matching the wildcard pattern with an actual ref we
see at the remote. What this meant was that we needed to make sure not
just that the prefix matched, and also that a slash followed the part that
matched.
But a codepath that scans the result from ls-remote and finds matching
refs forgot to check the "matching part must be followed by a slash" rule.
This resulted in "refs/heads/b1" from the remote side to mistakenly match
the source side of "refs/heads/b/*:refs/remotes/b/*" refspec.
Worse, the refspec crafted internally by "git-clone", and a hardcoded
preparsed refspec that is used to implement "git-fetch --tags", violated
this "parsed widcard refspec does not end with slash" rule; simply adding
the "matching part must be followed by a slash" rule then would have
broken codepaths that use these refspecs.
This commit changes the rule to require a trailing slash to parsed
wildcard refspecs. IOW, "refs/heads/b/*:refs/remotes/b/*" is parsed as
src = "refs/heads/b/" and dst = "refs/remotes/b/". This allows us to
simplify the matching logic because we only need to do a prefixcmp() to
notice "refs/heads/b/one" matches and "refs/heads/b1" does not.
Acked-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If a user passes "--template=", then our template parameter
is blank. Unfortunately, copy_templates() assumes it has at
least one character, and does all sorts of bad things like
reading from template[-1] and then proceeding to link all of
'/' into the .git directory.
This patch just checks for that condition in copy_templates
and aborts. As a side effect, this means that --template=
now has the meaning "don't copy any templates."
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Makefile records paths to a few programs in GIT-BUILD-OPTIONS file. These
paths need to be quoted twice: once to protect specials from the shell
that runs the generated GIT-BUILD-OPTIONS file, and again to protect them
(and the first level of quoting itself) from the shell that runs the
"echo" inside the Makefile.
You can test this by trying:
$ ln -s /bin/tar "$HOME/Tes' program/tar"
$ make TAR="$HOME/Tes' program/tar" test
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When index-pack completes a thin pack it appends objects to the pack.
Since the commit 92392b4(index-pack: Honor core.deltaBaseCacheLimit when
resolving deltas) such an object can be pruned in case of memory
pressure, and will be read back again by get_data_from_pack(). For this
to work, the fields in object_entry structure need to be initialized
properly.
Noticed by Pierre Habouzit.
Signed-off-by: Björn Steinbrink <B.Steinbrink@gmx.de>
Acked-by: Nicolas Pitre <nico@cam.org>
Acked-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 8291db6 (git-send-email: add charset header if we add encoded 'From',
2007-11-16), "$1" is used from a regexp without using () to capture
anything in $1. Later, when that value was used, it causes a warning about
a variable being undefined, instead of using the correct value for
comparison (not that it makes difference in the current code that does not
do actual re-encoding).
Signed-off-by: Peter Valdemar Mørch <peter@morch.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
This fixes an issue when you use:
$ git checkout -- <path1> [<paths>...]
and that <path1> can also be understood as a reference. git-checkout
mistakenly understands this as the same as:
$ git checkout <path1> -- [<paths>...]
because parse-options was eating the '--' and the argument parser thought
he was parsing:
$ git checkout <path1> [<paths>...]
Where there indeed is an ambiguity
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Without this patch, git-grep gives confusing usage information:
$ git grep --confused
usage: git grep <option>* <rev>* [-e] <pattern> [<path>...]
$ git grep HEAD pattern
fatal: ambiguous argument 'pattern': unknown revision or path no
t in the working tree.
Use '--' to separate paths from revisions
So put <pattern> before the <rev>s, in accordance with actual correct
usage. While we're changing the usage string, we might as well include
the "--" separating revisions and paths, too.
Signed-off-by: Jonathan Nieder <jrnieder@uchicago.edu>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
5fdeacb (Teach update-index about --ignore-submodules, 2008-05-14) added a
new refresh option flag but did not assign a unique bit for it correctly,
and broke "update-index --ignore-missing".
This fixes it.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When hold_locked_index() is called with a relative git_dir and you are
outside the work tree, the lock file become relative to the current
directory. So when later setup_work_tree() change the current directory
it breaks lock file path and commit_locked_index() fails.
This patch move index locking code after setup_work_tree() call to make
lock file relative to the working tree as it should be and add a test
case.
Noticed by Nick Andrew.
Signed-off-by: Olivier Marin <dkr@freesurf.fr>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* sp/maint-index-pack:
index-pack: Honor core.deltaBaseCacheLimit when resolving deltas
index-pack: Track the object_entry that creates each base_data
index-pack: Chain the struct base_data on the stack for traversal
index-pack: Refactor base arguments of resolve_delta into a struct
Nick Andrew noticed that rev-list lets --quiet option to be parsed by
underlying diff_options parser but did not pick up the result. This
resulted in --quiet option to become effectively a no-op.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Replace "run_command_v_opt_dir" by "run_command_v_opt_cd".
Signed-off-by: Stephan Beyer <s-beyer@gmx.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If the repo is empty, it is obvious that there are no common commits
when fetching from _anywhere_.
So there is no use in saying it in that case, and it can even be
annoying. Therefore suppress the message unilaterally if the repository
is empty prior to the fetch.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
* sp/maint-bash-completion-optim:
bash completion: Resolve git show ref:path<tab> losing ref: portion
bash completion: Append space after file names have been completed
bash completion: Don't offer "a.." as a completion for "a."
bash completion: Improve responsiveness of git-log completion
When "rebase -i -p" tries to preserve merges of unrelated branches, it
lost some parents:
- When you have more than two parents, the commit in the new history
ends up with fewer than expected number of parents and this breakage
goes unnoticed;
- When you are rebasing a merge with two parents and one is lost, the
command tries to cherry-pick the original merge commit, and the command
fails.
Signed-off-by: Stephan Beyer <s-beyer@gmx.net>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
It used plain 'if git merge ...', which hides a segfault. The test does not pass.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If PATH_MAX on your system is smaller than a path stored in the git repo,
it may cause the buffer overflow in prepare_attr_stack.
Signed-off-by: Dmitry Potapov <dpotapov@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If PATH_MAX on your system is smaller than a path stored, it may cause
buffer overflow and stack corruption in diff_addremove() and diff_change()
functions when running git-diff
Signed-off-by: Dmitry Potapov <dpotapov@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If PATH_MAX on your system is smaller than any path stored in the git
repository, that can cause memory corruption inside of the grep_tree
function used by git-grep.
Signed-off-by: Dmitry Potapov <dpotapov@gmail.com>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-cvsserver.perl contained a single call to a nonexistant function
cleanupWorkDir(). This was obviously a typo for cleanupWorkTree().
Signed-off-by: Lars Noschinski <lars@public.noschinski.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The manual page of git-cherry-pick and git-revert asserts that -n works
primarily on the working tree, while in fact the primary object it operates
on is the index, and the changes only "accidentally" propagate to the
working tree. This e.g. leads innocent #git IRC folks to believe that you
can use -n to prepare changes for git-add -i staging.
Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we are trying to resolve deltas for a long delta chain composed
of multi-megabyte objects we can easily run into requiring 500M+
of memory to hold each object in the chain on the call stack while
we recurse into the dependent objects and resolve them.
We now use a simple delta cache that discards objects near the
bottom of the call stack first, as they are the most least recently
used objects in this current delta chain. If we recurse out of a
chain we may find the base object is no longer available, as it was
free'd to keep memory under the deltaBaseCacheLimit. In such cases
we must unpack the base object again, which will require recursing
back to the root of the top of the delta chain as we released that
root first.
The astute reader will probably realize that we can still exceed
the delta base cache limit, but this happens only if the most
recent base plus the delta plus the inflated dependent sum up to
more than the base cache limit. Due to the way patch_delta is
currently implemented we cannot operate in less memory anyway.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we free the data stored within a base_data we need the struct
object_entry to get the data back again for use with another dependent
delta. Storing the object_entry* in base_data makes it simple to call
get_data_from_pack() to recover the compressed information.
This however means that we must add the missing base object to the end of
our packfile prior to calling resolve_delta() on each of the dependent
deltas. Adding the base first ensures we can read the base back from the
pack we are indexing, as if it had been included by the remote side.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We need to release earlier inflated base objects when memory gets
low, which means we need to be able to walk up or down the stack
to locate the objects we want to release, and free their data.
The new link/unlink routines allow inserting and removing the struct
base_data during recursion inside resolve_delta, and the global
base_cache gives us the head of the chain (bottom of the stack)
so we can traverse it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
We need to discard base objects which are not recently used if our
memory gets low, such as when we are unpacking a long delta chain
of a very large object.
To support tracking the available base objects we combine the
pointer and size into a struct. Future changes would allow the
data pointer to be free'd and marked NULL if memory gets low.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Linus reported that the bash completion for git show often dropped
the ref portion of the argument (stuff before the :) when trying
to complete a file name of a file in another branch or tag.
Björn Steinbrink tracked it down to the gvfs completion script
which comes standard on many Fedora Core based systems. That is
removing : from COMP_WORDBREAKS, making readline treat the entire
argument (including the ref) as the name that must be completed.
When the git completion routines supplied a completion of just the
filename, readline replaced everything.
Since Git users often need to use "ref:path" or "ref:ref" sort of
arguments, and expect completion support on both sides of the :
we really want the : in COMP_WORDBREAKS to provide a good user
experience. This is also the default that ships with bash as it
can be useful in other contexts, such as rcp/scp.
We now try to add : back to COMP_WORDBREAKS if it has been removed
by a script that loaded before us. However if this doesn't work
(as the : is stripped after we load) we fallback in the completion
routines to include "ref:" as part of the prefix for completions,
allowing readine to fully insert the argument the user wanted.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When completing `git show origin/maint:Makef<tab>` we should add a
space after the filename has been completed, so that the user can
immediately begin the next argument.
I also added a special case for the symlink variant so we treat it
just like a normal blob, as there are no items below it in the Git
tree structure.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
"Subject: " isn't in the static array "header", and thus
memcmp("Subject:", header[i], 7) will never match.
Even if it did so, hdr_data[] may not have been allocated if there weren't
a "Subject: " in-body when we process "[PATCH]" in the affected codepath.
Signed-off-by: Lukas Sandström <lukass@etek.chalmers.se>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
If the user is trying to complete "v1.5.3.<tab>" to see all of
the available maintenance releases for 1.5.3 we should not give
them an extra dot as the completion. Instead if the user wants
a ".." or a "..." operator they should key the two dots out on
their own. Its the same number of keystrokes either way.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Junio noticed the bash completion has been taking a long time lately.
Petr Baudis tracked it down to 72e5e989b ("bash: Add space after
unique command name is completed."). Tracing the code showed
we spent significant time inside of this loop within __gitcomp,
due to the string copying overhead.
[28.146109654] _git common over
[28.164791148] gitrefs in
[28.280302268] gitrefs dir out
[28.300939737] gitcomp in
[28.308378112] gitcomp pre-case
* [28.313407453] gitcomp iter in
* [28.701270296] gitcomp iter out
[28.713370786] out normal
Since __git_refs avoids this string copying by forking and using
echo we use the same trick here when we need to finish generating
the names for the caller.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The test created an initial commit, made .git/objects unwritable and then
exercised various codepaths to create loose commit, tree and blob objects
to make sure the commands notice failures from these attempts.
However, the initial commit was not preceded with test_tick, which made
its object name depend on the timestamp. The names of all the later tree
and blob objects the test tried to create were static. If the initial
commit's object name happened to begin with the same two hexdigits as the
tree or blob objects the test later attempted to create, the fan-out
directory in which these tree or blob would be created is already created
when the initial commit was made, and the object creation succeeds, and
commands being tested should not notice any failure --- in short, the test
was bogus.
This makes the fan-out directories also unwritable, and adds test_tick
before the commit object creation to make the test repeatable.
The contents of the file to create a blob from "a" to "60" is to force the
name of the blob object to begin with "1b", which shares the fan-out
directory with the initial commit that is created with the test. This was
useful when diagnosing the breakage of this test.
Signed-off-by: Junio C Hamano <gitster@pobox.com>
The pretty format %an does not respect .mailmap, but gives the exact
author name recorded in the commit. Sometimes it is more desirable,
however, to look if the email has another name mapped to it in .mailmap.
This commit adds %aN (and %cN for the committer name) to do exactly that.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
06cbe85 (Make core.sharedRepository more generic, 2008-04-16) broke the
traditional setting of core.sharedRepository to true, which was to make
the repository group writable: with umask 022, it would clear the
permission bits for 'other'. (umask 002 did not exhibit this behaviour
since pre-chmod() check in adjust_shared_perm() fails in that case.)
The call to adjust_shared_perm() should only loosen the permission.
If the user has umask like 022 or 002 that allow others to read, the
resulting files should be made readable and writable by group, without
restricting the readability by others.
This patch fixes the adjust_shared_perm() mode tweak based on Junio's
suggestion and adds the appropriate tests to t/t1301-shared-repo.sh.
Cc: Heikki Orsila <heikki.orsila@iki.fi>
Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
When recursing to unpack a delta base we must unuse_pack() so that
the pack window for the current object does not remain pinned in
memory while the delta base is itself being unpacked and materialized
for our use.
On a long delta chain of 50 objects we may need to access 6 different
windows from a very large (>3G) pack file in order to obtain all
of the delta base content. If the process ulimit permits us to
map/allocate only 1.5G we must release windows during this recursion
to ensure we stay within the ulimit and transition memory from pack
cache to standard malloc, or other mmap needs.
Inserting an unuse_pack() call prior to the recursion allows us to
avoid pinning the current window, making it available for garbage
collection if memory runs low.
This has been broken since at least before 1.5.1-rc1, and very
likely earlier than that. Its fixed now. :)
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <gitster@pobox.com>
Johannes Schindelin
Junio C Hamano
Steve Haslam
Anders Melchiorsen
Jeff King
Stephan Beyer
Björn Steinbrink
Peter Valdemar Mørch
Pierre Habouzit
Jonathan Nieder
Olivier Marin
Johannes Sixt
Dmitry Potapov
Lars Noschinski
Petr Baudis
Shawn O. Pearce
Lukas Sandström