Avoid a dup2(2) in apply_filter() - start_command() can do it for us.
When apply_filter() runs the external (clean or smudge) filter program, it
needs to pass the writable end of a pipe as its stdout. For this purpose,
it used to dup2(2) the file descriptor explicitly to stdout. Now we use
the facilities of start_command() to do it for us.
Furthermore, the path argument of a subordinate function, filter_buffer(),
was not used, so here we replace it to pass the fd instead.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
t0021-conversion.sh: Test that the clean filter really cleans content.
This test uses a rot13 filter, which is its own inverse. It tested only
that the content was the same as the original after both the 'clean' and
the 'smudge' filter were applied. This way it would not detect whether
any filter was run at all. Hence, here we add another test that checks
that the repository contained content that was processed by the 'clean'
filter.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
upload-pack: Run rev-list in an asynchronous function.
This gets rid of an explicit fork().
Since upload-pack has to coordinate two processes (rev-list and
pack-objects), we cannot use the normal finish_async(), but have to monitor
the process explicitly. Hence, there are no changes at this front.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
upload-pack: Move the revision walker into a separate function.
This allows us later to use start_async() with this function, and at
the same time is a nice cleanup that makes a long function
(create_pack_file()) shorter.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Use the asyncronous function infrastructure in builtin-fetch-pack.c.
We run the sideband demultiplexer in an asynchronous function.
Note that earlier there was a check in the child process that closed
xd[1] only if it was different from xd[0]; this test is no longer needed
because git_connect() always returns two different file descriptors
(see ec587fde0a76780931c7ac32474c8c000aa45134).
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Add infrastructure to run a function asynchronously.
This adds start_async() and finish_async(), which runs a function
asynchronously. Communication with the caller happens only via pipes.
For this reason, this implementation forks off a child process that runs
the function.
[sp: Style nit fixed by removing unnecessary block on if condition
inside of start_async()]
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
upload-pack: Use start_command() to run pack-objects in create_pack_file().
This gets rid of an explicit fork/exec.
Since upload-pack has to coordinate two processes (rev-list and
pack-objects), we cannot use the normal finish_command(), but have to
monitor the processes explicitly. Hence, the waitpid() call remains.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Have start_command() create a pipe to read the stderr of the child.
This adds another stanza that allocates a pipe that is connected to the
child's stderr and that the caller can read from. In order to request this
pipe, the caller sets cmd->err to -1.
The implementation is not exactly modeled after the stdout case: For stdout
the caller can supply an existing file descriptor, but this facility is
nowhere needed in the stderr case. Additionally, the caller is required to
close cmd->err.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Change git_connect() to return a struct child_process instead of a pid_t.
This prepares the API of git_connect() and finish_connect() to operate on
a struct child_process. Currently, we just use that object as a placeholder
for the pid that we used to return. A follow-up patch will change the
implementation of git_connect() and finish_connect() to make full use
of the object.
Old code had early-return-on-error checks at the calling sites of
git_connect(), but since git_connect() dies on errors anyway, these checks
were removed.
[sp: Corrected style nit of "conn == NULL" to "!conn"]
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Define compat version of mkdtemp for systems lacking it
Solaris 9 doesn't have mkdtemp() so we need to emulate it for the
rsync transport implementation. Since Solaris 9 is lacking this
function we can also reasonably assume it is not available on
Solaris 8 either. The new Makfile definition NO_MKDTEMP can be
set to enable the git compat version.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Avoid scary errors about tagged trees/blobs during git-fetch
This is the same bug as 42a32174b600f139b489341b1281fb1bfa14c252.
The warning "Object $X is a tree, not a commit" is bogus and is
not relevant here. If its not a commit we just need to make sure
we don't mark it for merge as we fill out FETCH_HEAD.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
fetch: if not fetching from default remote, ignore default merge
When doing "git fetch <remote>" on a remote that does not have the
branch referenced in branch.<current-branch>.merge, git fetch failed.
It failed because it tried to add the "merge" ref to the refs to be
fetched.
Fix that. And add a test case.
Incidentally, this unconvered a bug in our own test suite, where
"git pull <some-path>" was expected to merge the ref given in the
defaults, even if not pulling from the default remote.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If the end-user requested a dry-run push we need to pass that flag
over to http-push and additionally make sure it does not actually
upload any changes to the remote server.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If the end-user requested a dry-run push we should pass that flag
though to rsync so that the rsync command can show what it would do
(or not do) if push was to be executed without the --dry-run flag.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The builtin-fetch topic changed push's handling of --all to setting
the new TRANSPORT_PUSH_ALL flag before starting the push subroutine
for the given transport. Unfortunately not all references within
builtin-push were changed to test this flag therefore allowing push
to incorrectly accept refspecs and --all.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
There's a number of tricky conflicts between master and
this topic right now due to the rewrite of builtin-push.
Junio must have handled these via rerere; I'd rather not
deal with them again so I'm pre-merging master into the
topic. Besides this topic somehow started to depend on
the strbuf series that was in next, but is now in master.
It no longer compiles on its own without the strbuf API.
* master: (184 commits)
Whip post 1.5.3.4 maintenance series into shape.
Minor usage update in setgitperms.perl
manual: use 'URL' instead of 'url'.
manual: add some markup.
manual: Fix example finding commits referencing given content.
Fix wording in push definition.
Fix some typos, punctuation, missing words, minor markup.
manual: Fix or remove em dashes.
Add a --dry-run option to git-push.
Add a --dry-run option to git-send-pack.
Fix in-place editing functions in convert.c
instaweb: support for Ruby's WEBrick server
instaweb: allow for use of auto-generated scripts
Add 'git-p4 commit' as an alias for 'git-p4 submit'
hg-to-git speedup through selectable repack intervals
git-svn: respect Subversion's [auth] section configuration values
gtksourceview2 support for gitview
fix contrib/hooks/post-receive-email hooks.recipients error message
Support cvs via git-shell
rebase -i: use diff plumbing instead of porcelain
...
manual: Fix example finding commits referencing given content.
If I'm handed a file, then it typically lives outside the
working directory. git-log only operates on in-tree files,
so the first 'filename' should be an in-tree one, or it should
look at all files. This patch does the latter, so it would
also find renamed files. However, it is also slower.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
The default behaviour of git-push is potentially confusing
for new users, since it will push changes that are not on
the current branch. Publishing patches that were still
cooking on a development branch is hard to undo.
It would also be nice to be able to verify the expansion
of refspecs if you've edited them, so that you know
what branches matched on the server.
Adding a --dry-run flag allows the user to experiment
safely and learn how to use git-push properly. Originally
suggested by Steffen Prohaska.
Signed-off-by: Brian Ewins <brian.ewins@gmail.com> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Implement support for --dry-run, so that it can be used
in calls from git-push. With this flag set, git-send-pack
will not send any updates to the server.
Signed-off-by: Brian Ewins <brian.ewins@gmail.com> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
There were a few places which did not cope well without curl. This
fixes all of them. We still need to link against the walker.o part
of the library as some parts of transport.o still call into there
even though we don't have HTTP support enabled.
If compiled with NO_CURL=1 we now get the following useful error
message:
$ git-fetch http://www.example.com/git
error: git was compiled without libcurl support.
fatal: Don't know how to fetch from http://www.example.com/git
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Don't grow the buffer if there is enough space in the first place.
As a side effect, when the editing is done "in place", we don't grow, so
the buffer pointer doesn't changes, and `src' isn't invalidated anymore.
Thanks to Bernt Hansen for the bug report.
* apply_filter:
Fix memory leak due to fake in-place editing that didn't collected the
old buffer when the filter succeeds. Also a cosmetic fix.
Signed-off-by: Pierre Habouzit <madcoder@debian.org> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
running the webrick server with git requires Ruby and Ruby's YAML and
Webrick libraries (both of which come standard with Ruby). nice for
single-user standalone invocations.
the --httpd=webrick option generates a ruby script on the fly to read
httpd.conf options and invoke the web server via library call. this
script is placed in the .git/gitweb directory. it also generates a
shell script in a feeble attempt to invoke ruby in a portable manner,
which assumes that 'ruby' is in the user's $PATH.
Signed-off-by: Mike Dalessio <mike@csa.net> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
this patch allows scripts that reside in $fqgitdir/gitweb to be used
for firing up an instaweb server. this lays the groundwork for
extending instaweb support to non-standard web servers, which may
require a script for proper invocation.
Signed-off-by: Mike Dalessio <mike@csa.net> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Add 'git-p4 commit' as an alias for 'git-p4 submit'
Given that git uses 'commit', git-p4's 'sumbit' was a bit confusing at times;
often making me do 'git submit' and 'git-p4 commit' instead.
Signed-off-by: Marius Storm-Olsen <marius@trolltech.com> Acked-By: Simon Hausmann <simon@lst.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Parameters 'store-passwords' and 'store-auth-creds' from Subversion's
configuration (~/.subversion/config) were not respected. This was
fixed: the default values for these parameters are set to 'yes' to
follow Subversion behaviour.
Signed-off-by: Eygene Ryabinkin <rea-git@codelabs.ru> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Acked-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
This adds cvs support to the git-shell; You can now give new users
a restricted git-shell and they still can commit via git's cvs
emulator.
Note that either the gecos information must be accurate, or you must
provide a $HOME/.gitconfig with the appropriate user credentials.
Since the git-shell is too restricted to allow the user to do it
(on purpose!), it is up to the administrator to take care of that.
Based on an idea by Jan Wielemaker.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
When diff drivers are installed, calling "git diff <tree1>..<tree2>"
calls those drivers. This borks the patch generation of rebase -i.
So use "git diff-tree -p" instead, which does not call diff drivers.
Noticed by Johannes Sixt.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Before this patch the clean target has removed the
configure script that comes with Git tar file.
That made compiling Git for different architectures
inconvenient.
This patch excludes configure from the files to be
deleted by 'make clean' and adds new target 'distclean'
to preserve old functionality.
Signed-off-by: Mathias Megyei <mathias@mnet-mail.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-config: handle --file option with relative pathname properly
When calling git-config not from the top level directory of a repository,
it changes directory before trying to open the config file specified
through the --file option, which then fails if the config file was
specified by a relative pathname. This patch adjusts the pathname to
the config file if applicable.
The problem was noticed by Joey Hess, reported through
http://bugs.debian.org/445208
Signed-off-by: Gerrit Pape <pape@smarden.org> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Before this patch, clear_commit_marks() recursed for each parent. This
could be potentially very expensive in terms of stack space. Probably
the only reason that this did not lead to problems is the fact that we
typically call clear_commit_marks() after marking a relatively small set
of commits.
Use (sort of) a tail recursion instead: first recurse on the parents
other than the first one, and then continue the loop with the first
parent.
Noticed by Shawn Pearce.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git add -i: Fix parsing of abbreviated hunk headers
The unified diff format allows one-line ranges to be abbreviated
by omiting the size. The hunk header "@@ -10,1 +10,1 @@" can be
expressed as "@@ -10 +10 @@", but this wasn't properly parsed in
all cases.
Such abbreviated hunk headers are generated when a one-line change
(add, remove or modify) appears without context; for example
because the file is a one-liner itself or because GIT_DIFF_OPTS
was set to '-u0'. If the user then runs 'git add -i' and enters
the 'patch' command for that file, perl complains about undefined
variables.
Signed-off-by: Jean-Luc Herren <jlh@gmx.ch> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-config: don't silently ignore options after --list
Error out if someone gives options after --list since that is
not a valid syntax.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gc: by default use safer "-A" option to repack when not --prune'ing
This makes use of repack's new "-A" option which does not drop packed
unreachable objects. This makes git-gc safe to call at any time,
particularly when a repository is referenced as an alternate by
another repository.
git-gc --prune will use the "-a" option to repack instead of "-A", so
that packed unreachable objects will be removed.
Signed-off-by: Brandon Casey <casey@nrlssc.navy.mil> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Fix a crash in ls-remote when refspec expands into nothing
Originally-by: Väinö Järvelä <v@pp.inet.fi> Signed-off-by: Alex Riesen <raa.lkml@gmail.com> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If multiple refspecs matched the same ref, the update would be
processed multiple times. Now having the same destination for the same
source has no additional effect, and having the same destination for
different sources is an error.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Clean up "git log" format with DIFF_FORMAT_NO_OUTPUT
This fixes an unnecessary empty line that we add to the log message when
we generate diffs, but don't actually end up printing any due to having
DIFF_FORMAT_NO_OUTPUT set.
This can happen with pickaxe or with rename following. The reason is that
we normally add an empty line between the commit and the diff, but we do
that even for the case where we've then suppressed the actual printing of
the diff.
This also updates a couple of tests that assumed the extraneous empty
line would exist at the end of output.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
It turns out that I completely broke "git log --follow" with my recent
patch to revision.c ("Fix revision log diff setup, avoid unnecessary diff
generation", commit b7bb760d5ed4881422673d32f869d140221d3564).
Why? Because --follow obviously requires the diff machinery to function,
exactly the same way pickaxe does.
So everybody is away right now, but considering that nobody even noticed
this bug, I don't think it matters. But for the record, here's the trivial
one-liner fix (well, two, since I also fixed the comment).
Because of the nature of the bug, if you ask for patches when following
(which is one of the things I normally do), the bug is hidden, because
then the request for diff output will automatically also enable the diffs
themselves.
So while "git log --follow <filename>" didn't work, adding a "-p"
magically made it work again even without this fix.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
In most cases of branching, the tree is copied unmodified from the trunk
to the branch. When that is done, we can simply start with the parent's
index and apply the changes on the branch as usual.
[ew: rewritten from Steven's original to use SVN::Client instead
of the command-line svn client.
Since SVN::Client connects separately, we'll share our
authentication providers array between our usages of
SVN::Client and SVN::Ra, too. Bypassing the high-level
SVN::Client library can avoid this, but the code will be
much more complex. Regardless, any implementation of this
seems to require restarting a connection to the remote
server.
Also of note is that SVN 1.4 and later allows a more
efficient diff_summary to be done instead of a full diff,
but since this code is only to support SVN < 1.4.4, we'll
ignore it for now.]
Signed-off-by: Steven Walter <stevenrwalter@gmail.com> Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* ph/strbuf: (44 commits)
Make read_patch_file work on a strbuf.
strbuf_read_file enhancement, and use it.
strbuf change: be sure ->buf is never ever NULL.
double free in builtin-update-index.c
Clean up stripspace a bit, use strbuf even more.
Add strbuf_read_file().
rerere: Fix use of an empty strbuf.buf
Small cache_tree_write refactor.
Make builtin-rerere use of strbuf nicer and more efficient.
Add strbuf_cmp.
strbuf_setlen(): do not barf on setting length of an empty buffer to 0
sq_quote_argv and add_to_string rework with strbuf's.
Full rework of quote_c_style and write_name_quoted.
Rework unquote_c_style to work on a strbuf.
strbuf API additions and enhancements.
nfv?asprintf are broken without va_copy, workaround them.
Fix the expansion pattern of the pseudo-static path buffer.
builtin-for-each-ref.c::copy_name() - do not overstep the buffer.
builtin-apply.c: fix a tiny leak introduced during xmemdupz() conversion.
Use xmemdupz() in many places.
...
* lh/merge:
git-merge: add --ff and --no-ff options
git-merge: add support for --commit and --no-squash
git-merge: add support for branch.<name>.mergeoptions
git-merge: refactor option parsing
git-merge: fix faulty SQUASH_MSG
Add test-script for git-merge porcelain
* jc/autogc:
git-gc --auto: run "repack -A -d -l" as necessary.
git-gc --auto: restructure the way "repack" command line is built.
git-gc --auto: protect ourselves from accumulated cruft
git-gc --auto: add documentation.
git-gc --auto: move threshold check to need_to_gc() function.
repack -A -d: use --keep-unreachable when repacking
pack-objects --keep-unreachable
Export matches_pack_name() and fix its return value
Invoke "git gc --auto" from commit, merge, am and rebase.
Implement git gc --auto
* ap/dateformat:
Add a test script for for-each-ref, including test of date formatting
dateformat: parse %(xxdate) %(yydate:format) correctly
Make for-each-ref's grab_date() support per-atom formatting
Make for-each-ref allow atom names like "<name>:<something>"
parse_date_format(): convert a format name to an enum date_mode
This tests basic functionality and also exercises a bug noticed
by Keith Packard, (prune_cache followed by add_index_entry can
trigger an attempt to realloc a pointer into the middle of an
allocated buffer).
Must not modify the_index.cache as it may be passed to realloc at some point.
The index cache is not static, growing as new entries are added. If
entries are added after prune_cache is called, cache will no longer
point at the base of the allocation, and realloc will not be happy.
I verified that this was the only place in the current source which
modified any index_state.cache elements aside from the alloc/realloc
calls in read-cache by changing the type of the element to 'struct
cache_entry ** const cache' and recompiling.
A more efficient patch would create a separate 'cache_base' value to
track the allocation and then fix things up when reallocation was
necessary, instead of the brute-force memmove used here.
This adds a verbosity level below 0 for suppressing default messages
with --quiet, and makes the default for http be verbose instead of
quiet. This matches the behavior of the shell script version of git-fetch.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
We find rename candidates by computing a fingerprint hash of
each file, and then comparing those fingerprints. There are
inherently O(n^2) comparisons, so it pays in CPU time to
hoist the (rather expensive) computation of the fingerprint
out of that loop (or to cache it once we have computed it once).
Previously, we didn't keep the filespec information around
because then we had the potential to consume a great deal of
memory. However, instead of keeping all of the filespec
data, we can instead just keep the fingerprint.
This patch implements and uses diff_free_filespec_data_large
to accomplish that goal. We also have to change
estimate_similarity not to needlessly repopulate the
filespec data when we already have the hash.
Practical tests showed 4.5x speedup for a 10% memory usage
increase.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The string value of %(numparent) was not returned correctly.
Also %(parent) misbehaved for the root commits (returned garbage)
and merge commits (returned first parent, followed by a space).
We rely on TMP_INDEX variable to decide if we are doing a partial commit,
as it is only set in the partial commit codepath. But the variable is
never initialized. A stray environment variable from outside could
ruin the day.
We lost rsync support when transitioning from shell to C. Support it
again (even if the transport is technically deprecated, some people just
do not have any chance to use anything else).
Also, add a test to t5510. Since rsync transport is not configured by
default on most machines, and especially not such that you can write to
rsync://127.0.0.1$(pwd)/, it is disabled by default; you can enable it by
setting the environment variable TEST_RSYNC.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint:
Whip post 1.5.3.3 maintenance series into shape.
git stash: document apply's --index switch
post-receive-hook: Remove the From field from the generated email header so that the pusher's name is used
post-receive-hook: Remove the From field from the generated email header so that the pusher's name is used
Using the name of the committer of the revision at the tip of the
updated ref is not sensible. That information is available in the email
itself should it be wanted, and by supplying a "From", we were
effectively hiding the person who performed the push - which is useful
information in itself.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
There was a function called remove_empty_dir_recursive() buried
in refs.c. Expose a slightly enhanced version in dir.h: it can now
optionally remove a non-empty directory.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Currently alloc_ref() expects the length of the refname plus 1
as its parameter, prepares that much space and returns a "ref"
structure for the caller to fill the refname. One caller in
transport.c::get_refs_from_bundle() however allocated one byte
less.
It may be a good idea to change the calling convention to give
alloc_ref() the length of the refname, but that clean-up can be
done in a separate patch. This patch only fixes the bug and
makes all callers consistent.
There was also one overallocation in connect.c, which would not
hurt but was wasteful. This patch fixes it as well.