* 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
Don't use "<unknown>" for placeholders and suppress printing of empty user formats.
This changes the interporate() to replace entries with NULL values
by the empty string, and uses it to interpolate missing fields in
custom format output used in git-log and friends. It is most useful
to avoid <unknown> output from %b format for a commit log message
that lack any body text.
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>
rebase -i: create .dotest-merge after validating options.
Creating .dotest-merge before validating the options prevents both
--continue and --interactive from working if the options are invalid,
so only create it after validating the options.
[jc: however, just moving the creation of DOTEST breaks output]
Signed-off-by: Matt Kraai <kraai@ftbfs.org> Acked-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The tinyurl is incorrect -- it attempts to go to groups.osdl.org,
which is gone. Either use the full URL (in patch) or create a new
tinyurl for this URL.
Is the web page (where I first saw this problem) generated from
this txt file?
http://www.kernel.org/pub/software/scm/git/docs/core-tutorial.html
If not, it needs to be updated also.
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
When doing an "edit" on a commit, editing and git-adding some files,
"git rebase -i" complained about a missing "author-script". The idea was
that the user would call "git commit --amend" herself.
But we can be nice and do that for the user.
Noticed by Dmitry Potapov.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
gitattributes.txt: Be more to the point in the filter driver description.
The description was meant to emphasizes that the project should remain
usable even if the filter driver was not used. This makes it more explicit
and removes the "here is rope to hang yourself" paraphrase.
Signed-off-by: Johannes Sixt <johannes.sixt@telecom.at> Signed-off-by: Junio C Hamano <gitster@pobox.com>
It might be a sign of source code management gone bad, but when two branches
has diverged almost beyond recognition and time has come for the branches to
merge, the user is going to need all the help his tool can give him. Honoring
diff.renamelimit has great potential as a painkiller in such situations.
The painkiller effect could have been achieved by e.g. 'merge.renamelimit',
but the flexibility gained by a separate option is questionable: our user
would probably expect git to detect renames equally good when merging as
when diffing (I known I did).
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
convert-objects was needed to convert from an old-style repository,
which hashed the compressed contents and used a different date format.
Such repositories are presumably no longer common and, if such
conversions are necessary, should be done by writing a frontend for
git-fast-import.
Linus, the original author, is OK with moving it to contrib.
Signed-off-by: Matt Kraai <kraai@ftbfs.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
user-manual: Explain what submodules are good for.
Rework the introduction to the Submodules section to explain why
someone would use them, and fix up submodule references from the
tree-object and todo sections.
Signed-off-by: Michael Smith <msmith@cbnco.com> Acked-by: J. Bruce Fields <bfields@citi.umich.edu> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-submodule - allow a relative path as the subproject url
This allows a subproject's location to be specified and stored as relative
to the parent project's location (e.g., ./foo, or ../foo). This url is
stored in .gitmodules as given. It is resolved into an absolute url by
appending it to the parent project's url when the information is written
to .git/config (i.e., during submodule add for the originator, and
submodule init for a downstream recipient). This allows cloning of the
project to work "as expected" if the project is hosted on a different
server than when the subprojects were added.
Signed-off-by: Mark Levedahl <mdl123@verizon.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint:
Do not over-quote the -f envelopesender value.
unexpected Make output (e.g. from --debug) causes build failure
Fixed minor typo in t/t9001-send-email.sh test command line.
Without this, the value passed to sendmail would have an extra set of
single quotes. At least exim's sendmail emulation would object to that:
exim: bad -f address "'list-addr@example.org'": malformed address: ' \
may not follow 'list-addr@example.org
error: hooks/post-receive exited with error code 1
Signed-off-by: Jim Meyering <jim@meyering.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Prevent send-pack from segfaulting when a branch doesn't match
If `git push url foo` can't find a local branch named foo we can't
match it to any remote branch as the local branch is NULL and its
name is probably at position 0x34 in memory. On most systems that
isn't a valid address for git-send-pack's virtual address space
and we segfault.
If we can't find a source match and we have no destination we
need to abort the match function early before we try to match the
destination against the remote.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This simple change makes the body of "case 0" easier to read; no
matter what the value of matched_src is we want to break out of
the switch and not fall through. We only want to display an error
if matched_src is NULL, as this indicates there is no local branch
matching the input.
Also modified the default case's error message so it uses one less
line of text. Even at 8 column per tab indentation we still don't
break 80 columns with this new formatting.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix pool handling in git-svnimport to avoid memory leaks.
- Create an explicit one-and-only root pool.
- Closely follow examples in SVN::Core man page.
Before calling a subversion function, create a subpool of our
root pool and make it the new default pool.
- Create a subpool for looping over svn revisions and clear
this subpool (i.e. it mark for reuse, don't decallocate it)
at the start of the loop instead of allocating new memory
with each iteration.
See http://marc.info/?l=git&m=118554191513822&w=2 for a detailed
explanation of the issue.
Signed-off-by: Stefan Sperling <stsp@elego.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>
In 'commitdiff' view, for the merge commit, there is an extra header
for the difftree table, with links to commitdiffs to individual
parents. Do not show such header when there is nothing to show, for
trivial merges.
This means that for trivial merge you have to go to 'commit' view
to get links to diffs to each parent.
gitweb: Remove parse_from_to_diffinfo code from git_patchset_body
In commit 90921740bd00029708370673fdc537522aa48e6f
"gitweb: Split git_patchset_body into separate subroutines"
a part of git_patchset_body code was separated into parse_from_to_diffinfo
subroutine. But instead of replacing the separated code by the call to
mentioned subroutine, the call to subroutine was placed before the separated
code. This patch removes parse_from_to_diffinfo code from git_patchset_body
subroutine.
These new options can be used to control the policy for fast-forward
merges: --ff allows it (this is the default) while --no-ff will create
a merge commit.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-merge: add support for --commit and --no-squash
These options can be used to override --no-commit and --squash, which is
needed since --no-commit and --squash now can be specified as default merge
options in $GIT_DIR/config.
The change also introduces slightly different behavior for --no-commit:
when specified, it explicitly overrides --squash. Earlier,
'git merge --squash --no-commit' would result in a squashed merge (i.e. no
$GIT_DIR/MERGE_HEAD was created) but with this patch the command will
behave as if --squash hadn't been specified.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-merge: add support for branch.<name>.mergeoptions
This enables per branch configuration of merge options. Currently, the most
useful options to specify per branch are --squash, --summary/--no-summary
and possibly --strategy, but all options are supported.
Note: Options containing whitespace will _not_ be handled correctly. Luckily,
the only option which can include whitespace is --message and it doesn't
make much sense to give that option a default value.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Only the first 'remote' head is currently specified as an argument to 'git
log' when generating a SQUSH_MSG, which makes the generated message fail
to mention every commit involved in the merge. This fixes the problem.
Noticed-by: Junio C Hamano <gitster@pobox.com> Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
This test-script excercises the porcelainish aspects of git-merge, and
does it thoroughly enough to detect a small bug already noticed by Junio:
squashing an octopus generates a faulty .git/SQUASH_MSG.
Signed-off-by: Lars Hjemli <hjemli@gmail.com> Signed-off-by: Junio C Hamano <gitster@pobox.com>
* maint:
git-svn: don't attempt to spawn pager if we don't want one
Supplant the "while case ... break ;; esac" idiom
User Manual: add a chapter for submodules
user-manual: don't assume refs are stored under .git/refs
Detect exec bit in more cases.
Conjugate "search" correctly in the git-prune-packed man page.
Move the paragraph specifying where the .idx and .pack files should be
Documentation/git-lost-found.txt: drop unnecessarily duplicated name.
git-svn: don't attempt to spawn pager if we don't want one
Even though config_pager() unset the $pager variable, we were
blindly calling exec() on it through run_pager().
Noticed-by: Chris Moore <christopher.ian.moore@gmail.com> Signed-off-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
A lot of shell scripts contained stuff starting with
while case "$#" in 0) break ;; esac
and similar. I consider breaking out of the condition instead of the
body od the loop ugly, and the implied "true" value of the
non-matching case is not really obvious to humans at first glance. It
happens not to be obvious to some BSD shells, either, but that's
because they are not POSIX-compliant. In most cases, this has been
replaced by a straight condition using "test". "case" has the
advantage of being faster than "test" on vintage shells where "test"
is not a builtin. Since none of them is likely to run the git
scripts, anyway, the added readability should be worth the change.
A few loops have had their termination condition expressed
differently.
Signed-off-by: David Kastrup <dak@gnu.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Signed-off-by: Michael Smith <msmith@cbnco.com> Signed-off-by: Miklos Vajna <vmiklos@frugalware.org> Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
user-manual: don't assume refs are stored under .git/refs
The scripts taken from Tony Luck's howto assume all refs can be found
under .git/refs, but this is not necessarily true, especially since
git-gc runs git-pack-refs.
Also add a note warning of this in the chapter that introduces refs, and
fix the same incorrect assumption in one other spot.
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
gitk: Simplify highlighting interface and combine with Find function
This effectively coaelesces the highlighting function and the search
function. Instead of separate highlight and find controls, there is
now one set of interface elements that controls both. The main
selector is a drop-down menu that controls whether commits are
highlighted and searched for on the basis of text in the commit
(i.e. the commit object), files affected by the commit or strings
added/removed by the commit.
The functions to highlight by membership of a view or by ancestor/
descendent relation to the selected commit are gone, as is the
move to next/previous highlighted commit (shift-up/down) function.
git-gui: Ensure .git/info/exclude is honored in Cygwin workdirs
If we are using Cygwin and the git repository is actually a
workdir (by way of git-new-workdir) but this Tcl process is
a native Tcl/Tk and not the Cygwin Tcl/Tk then we are unable
to traverse the .git/info path as it is a Cygwin symlink and
not a standard Windows directory.
So we actually need to start a Cygwin process that can do the
path translation for us and let it test for .git/info/exclude
so we know if we can include that file in our git-ls-files or
not.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Handle starting on mapped shares under Cygwin
I really cannot explain Cygwin's behavior here but if we start
git-gui through Cygwin on a local drive it appears that Cygwin
is leaving $env(PATH) in Unix style, even if it started a native
(non-Cygwin) Tcl/Tk process to run git-gui. Yet starting that
same git-gui and Tcl/Tk combination through Cygwin on a network
share causes it to automatically convert $env(PATH) into Windows
style, which broke our internal "which" implementation.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
git-gui: Display message box when we cannot find git in $PATH
If we cannot find the git executable in the user's $PATH then
we cannot function correctly. Because we need that to get the
version so we can load our library correctly we cannot rely on
the library function "error_popup" here, as this is all running
before the library path has been configured, so error_popup is
not available to us.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Documentation/git-lost-found.txt: drop unnecessarily duplicated name.
I only did this back when I wanted to make sure git-log and gitk work
properly with non Occidental characters. There is really no reason to
keep it around.
sq_quote_argv and add_to_string rework with strbuf's.
* sq_quote_buf is made public, and works on a strbuf.
* sq_quote_argv also works on a strbuf.
* make sq_quote_argv take a "maxlen" argument to check the buffer won't grow
too big.
Signed-off-by: Pierre Habouzit <madcoder@debian.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Previously, if you passed a revision and a path to svn cp, it meant to look
back at that revision and select that path. New behaviour is to get the
path then go back to the revision (like other commands that accept @REV
or -rREV do). The more consistent syntax is not supported by the old
tools, so we have to try both in turn.
Signed-off-by: Sam Vilain <sam.vilain@catalyst.net.nz> Acked-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
git-svn: fix test for trunk svn (commit message not needed)
The 'svn mv -m "rename to thunk"' was a local operation, therefore not
needing a commit message, it was silently ignored. Newer svn clients will
instead raise an error.
Signed-off-by: Sam Vilain <sam.vilain@catalyst.net.nz> Acked-by: Eric Wong <normalperson@yhbt.net> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Full rework of quote_c_style and write_name_quoted.
* quote_c_style works on a strbuf instead of a wild buffer.
* quote_c_style is now clever enough to not add double quotes if not needed.
* write_name_quoted inherits those advantages, but also take a different
set of arguments. Now instead of asking for quotes or not, you pass a
"terminator". If it's \0 then we assume you don't want to escape, else C
escaping is performed. In any case, the terminator is also appended to the
stream. It also no longer takes the prefix/prefix_len arguments, as it's
seldomly used, and makes some optimizations harder.
* write_name_quotedpfx is created to work like write_name_quoted and take
the prefix/prefix_len arguments.
Thanks to those API changes, diff.c has somehow lost weight, thanks to the
removal of functions that were wrappers around the old write_name_quoted
trying to give it a semantics like the new one, but performing a lot of
allocations for this goal. Now we always write directly to the stream, no
intermediate allocation is performed.
As a side effect of the refactor in builtin-apply.c, the length of the bar
graphs in diffstats are not affected anymore by the fact that the path was
clipped.
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
If the gain is not obvious in the diffstat, the resulting code is more
readable, _and_ in checkout-index/update-index we now reuse the same buffer
to unquote strings instead of always freeing/mallocing.
This also is more coherent with the next patch that reworks quoting
functions.
The quoting function is also made more efficient scanning for backslashes
and treating portions of strings without a backslash at once.
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Add strbuf_remove, change strbuf_insert:
As both are special cases of strbuf_splice, implement them as such.
gcc is able to do the math and generate almost optimal code this way.
Add strbuf_swap:
Exchange the values of its arguments.
Use it in fast-import.c
Also fix spacing issues in strbuf.h
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
nfv?asprintf are broken without va_copy, workaround them.
* drop nfasprintf.
* move nfvasprintf into imap-send.c back, and let it work on a 8k buffer,
and die() in case of overflow. It should be enough for imap commands, if
someone cares about imap-send, he's welcomed to fix it properly.
* replace nfvasprintf use in merge-recursive with a copy of the strbuf_addf
logic, it's one place, we'll live with it.
To ease the change, output_buffer string list is replaced with a strbuf ;)
* rework trace.c to call vsnprintf itself. It's used to format strerror()s
and git command names, it should never be more than a few octets long, let
it work on a 8k static buffer with vsnprintf or die loudly.
Signed-off-by: Pierre Habouzit <madcoder@debian.org>
Fix memory leaks when disconnecting transport instances
Most transport implementations tend to allocate a data buffer
in the struct transport instance during transport_get() so we
need to free that data buffer when we disconnect it.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The only way to configure the unpacking limit is currently through
the .git/config (or ~/.gitconfig) mechanism as we have no existing
command line option interface to control this threshold on a per
invocation basis. This was intentional by design as the storage
policy of the repository should be a repository-wide decision and
should not be subject to variations made on individual command
executions.
Earlier builtin-fetch was bypassing the unpacking limit chosen by
the user through the configuration file as it did not reread the
configuration options through fetch_pack_config if we called the
internal fetch_pack() API directly. We now ensure we always run the
config file through fetch_pack_config at least once in this process,
thereby setting our unpackLimit properly.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Always obtain fetch-pack arguments from struct fetch_pack_args
Copying the arguments from a fetch_pack_args into static globals
within the builtin-fetch-pack module is error-prone and may lead
rise to cases where arguments supplied via the struct from the
new fetch_pack() API may not be honored by the implementation.
Here we reorganize all of the static globals into a single static
struct fetch_pack_args instance and use memcpy() to move the data
from the caller supplied structure into the globals before we
execute our pack fetching implementation. This strategy is more
robust to additions and deletions of properties.
As keep_pack is a single bit we have also introduced lock_pack to
mean not only download and store the packfile via index-pack but
also to lock it against repacking by creating a .keep file when
the packfile itself is stored. The caller must remove the .keep
file when it is safe to do so.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Refactor struct transport_ops inlined into struct transport
Aside from reducing the code by 20 lines this refactoring removes
a level of indirection when trying to access the operations of a
given transport "instance", making the code clearer and easier to
follow.
It also has the nice effect of giving us the benefits of C99 style
struct initialization (namely ".fetch = X") without requiring that
level of language support from our compiler. We don't need to worry
about new operation methods being added as they will now be NULL'd
out automatically by the xcalloc() we use to create the new struct
transport we supply to the caller.
This pattern already exists in struct walker, so we already have
a precedent for it in Git. We also don't really need to worry
about any sort of performance decreases that may occur as a result
of filling out 4-8 op pointers when we make a "struct transport".
The extra few CPU cycles this requires over filling in the "struct
transport_ops" is killed by the time it will take Git to actually
*use* one of those functions, as most transport operations are
going over the wire or will be copying object data locally between
two directories.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Rename remote.uri to remote.url within remote handling internals
Anyplace we talk about the address of a remote repository we always
refer to it as a URL, especially in the configuration file and
.git/remotes where we call it "remote.$n.url" or start the first
line with "URL:". Calling this value a uri within the internal C
code just doesn't jive well with our commonly accepted terms.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
These options are all strictly boolean (true/false). Its easier to
document this implicitly by making their storage type a single bit.
There is no compelling memory space reduction reason for this change,
it just makes the structure definition slightly more readable.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Avoid printing unnecessary warnings during fetch and push
If a transport doesn't support an option we already are telling
the higher level application (fetch or push) that the option is not
valid by sending back a >0 return value from transport_set_option
so there's not a strong motivation to have the function perform the
output itself. Instead we should let the higher level application
do the output if it is necessary. This avoids always telling the
user that depth isn't supported on HTTP urls even when they did
not pass a --depth option to git-fetch.
If the user passes an option and the option value is invalid we now
properly die in git-fetch instead of just spitting out a message
and running anyway. This mimics prior behavior better where
incorrect/malformed options are not accepted by the process.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Correct handling of branch.$name.merge in builtin-fetch
My prior bug fix for git-push titled "Don't configure remote "." to
fetch everything to itself" actually broke t5520 as we were unable
to evaluate a branch configuration of:
as remote "." did not have a "remote...fetch" configuration entry to
offer up refs/heads/master as a possible candidate available to be
fetched and merged. In shell script git-fetch and prior to the above
mentioned commit this was hardcoded for a url of "." to be the set of
local branches.
Chasing down this bug led me to the conclusion that our prior behavior
with regards to branch.$name.merge was incorrect. In the shell script
based git-fetch implementation we only fetched and merged a branch if
it appeared both in branch.$name.merge *and* in remote.$r.fetch, where
$r = branch.$name.remote. In other words in the following config file:
Attempting to run `git pull` while on branch "pu" would always give
the user "Already up-to-date" as git-fetch did not fetch pu and thus
did not mark it for merge in .git/FETCH_HEAD. The configured merge
would always be ignored and the user would be left scratching her
confused head wondering why merge did not work on "pu" but worked
fine on "master".
If we are using the "default fetch" specification for the current
branch and the current branch has a branch.$name.merge configured
we now union it with the list of refs in remote.$r.fetch. This
way the above configuration does what the user expects it to do,
which is to fetch only "master" by default but when on "pu" to
fetch both "master" and "pu".
This uncovered some breakage in the test suite where old-style Cogito
branches (.git/branches/$r) did not fetch the branches listed in
.git/config for merging and thus did not actually merge them if the
user tried to use `git pull` on that branch. Junio and I discussed
it on list and felt that the union approach here makes more sense to
DWIM for the end-user than silently ignoring their configured request
so the test vectors for t5515 have been updated to include for-merge
lines in .git/FETCH_HEAD where they have been configured for-merge
in .git/config.
Since we are now performing a union of the fetch specification and
the merge specification and we cannot allow a branch to be listed
twice (otherwise it comes out twice in .git/FETCH_HEAD) we need to
perform a double loop here over all of the branch.$name.merge lines
and try to set their merge flag if we have already schedule that
branch for fetching by remote.$r.fetch. If no match is found then
we must add new specifications to fetch the branch but not store it
as no local tracking branch has been designated.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Don't attempt to merge non-existant remotes in t5515
This was actually reverted in 756373da by Junio. We no longer
support merging the right hand side of a fetchspec in a branch's
branch.$name.merge configuration setting as we interpret these
names as being only those published by the remote we are going to
fetch from.
The older shell based implementation of git-fetch did not report an
error when branch.$name.merge was referencing a branch that does
not exist on the remote and we are running `git fetch` for the
current branch. The new builtin-fetch does notice this failure
and aborts the fetch, thus breaking the tests.
Junio and I kicked it around on #git earlier today and decided that
the best approach here is to error out and tell the user that their
configuration is wrong, as this is likely more user friendly than
silently ignoring the user's request. Since the new builtin-fetch
is already issuing the error there is no code change required, we
just need to remove the bad configuration from our test.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
If we are fetching something and were configured to do a forced
fetch and have no local ref to store the fetched object into we
cannot mark the local ref as having a forced update. Instead we
should just silently discard the + request.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Don't configure remote "." to fetch everything to itself
When we are talking about a remote URI of "." we are really talking
about *this* repository that we are fetching into or pushing out of.
There are no matching tracking branches for this repository; we
do not attempt to map a ref back to ourselves as this would either
create an infinite cycle (for example "fetch = +refs/*:refs/mine/*")
or it causes problems when we attempt to push back to ourselves.
So we really cannot setup a remote like this:
[remote "."]
url = .
fetch = +refs/*:refs/*
In the case of `git push . B:T` to fast-forward branch T to B's
current commit git-send-pack will update branch T to B, assuming that
T is the remote tracking branch for B. This update is performed
immediately before git-send-pack asks git-receive-pack to perform
the same update, and git-receive-pack then fails because T is not
where git-send-pack told it to expect T to be at.
In the case of `git fetch .` we really should do the same thing as
`git fetch $otherrepo`, that is load .git/FETCH_HEAD with the commit
of HEAD, so that `git pull .` will report "Already up-to-date".
We have always behaved like this before on this insane request and
we should at least continue to behave the same way. With the above
(bad) remote configuration we were instead getting fetch errors
about funny refs, e.g. "refs/stash".
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we are running fetch in a repository that has a detached HEAD
then there is no current_branch available. In such a case any ref
that the fetch might update by definition cannot also be the current
branch so we should always bypass the "don't update HEAD" test.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Remove unnecessary 'fetch' argument from transport_get API
We don't actually need to know at the time of transport_get if the
caller wants to fetch, push, or do both on the returned object.
It is easier to just delay the initialization of the HTTP walker
until we know we will need it by providing a CURL specific fetch
function in the curl_transport that makes sure the walker instance
is initialized before use.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Cleanup duplicate initialization code in transport_get
We always allocate and return a struct transport* right now as every
URL is considered to be a native Git transport if it is not rsync,
http/https/ftp or a bundle. So we can simplify the initialization
of a new transport object by performing one xcalloc call and filling
in only the attributes required.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Don't bother passing ref log details to walker in builtin-fetch
When using the walker API within builtin-fetch we don't allow
it to update refs locally; instead that action is reserved for
builtin-fetch's own main loop once the objects have actually
been downloaded.
Passing NULL here will bypass the unnecessary malloc/free of a
string buffer within the walker API. That buffer is never used
because the prior argument (the refs to update) is also NULL.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Properly cleanup in http_cleanup so builtin-fetch does not segfault
Junio and I both noticed that the new builtin-fetch was segfaulting
immediately on http/https/ftp style URLs (those that went through
libcurl and the commit walker). Although the builtin-fetch changes
in this area were really just minor refactorings there was one major
change made: we invoked http_init(), http_cleanup() then http_init()
again in the same process.
When we call curl_easy_cleanup() on each active_request_slot we
are telling libcurl we did not want that buffer to be used again.
Unfortunately we did not also deallocate the active_request_slot
itself nor did we NULL out active_queue_head. This lead us to
attempt to reuse these cleaned up libcurl handles when we later tried
to invoke http_init() a second time to reactivate the curl library.
The next file get operation then immediately segfaulted on most
versions of libcurl.
Properly freeing our own buffers and clearing the list causes us to
reinitialize the curl buffers again if/when we need to use libcurl
from within this same process.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Backup the array passed to fetch_pack so we can free items
fetch_pack() can call remove_duplicates() on its input array and
this will possibly overwrite an earlier entry with a later one if
there are any duplicates in the input array. In such a case the
caller here might then attempt to free an item multiple times as
it goes through its cleanup.
I also forgot to free the heads array we pass down into fetch_pack()
when I introduced the allocation of it in this function during my
builtin-fetch cleanup series. Better free it while we are here
working on related memory management fixes.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix builtin-fetch memory corruption by not overstepping array
A long time ago Junio added this line to always ensure that the
output array created by remove_duplicates() had a NULL as its
terminating node. Today none of the downstream consumers of this
array care about a NULL terminator; they only pay attention to the
size of the array (as indicated by nr_heads). In (nearly?) all
cases passing a NULL element will cause SIGSEGV failures. So this
NULL terminal is not actually necessary.
Unfortunately we cannot continue to NULL terminate the array at
this point as the array may only have been allocated large enough
to match the input of nr_heads. If there are no duplicates than
we would be trying to store NULL into heads[nr_heads] and that may
be outside of the array.
My recent series to cleanup builtin-fetch changed the allocation of
the heads array from 256 entries to exactly nr_heads thus ensuring
we were always overstepping the array and causing memory corruption.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Always ensure the pack.keep file is removed by git-fetch
If we are using a native transport and the transport chose to
save the packfile it may have created a .keep file to protect
the packfile from a concurrently running git-repack process.
In such a case the git-fetch process should make sure it will
unlink the .keep file even if it fails to update any refs as
otherwise the newly downloaded packfile's diskspace will never
be reclaimed if the objects are not actually referenced.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we are using a native packfile to perform a git-fetch invocation
and the received packfile contained more than the configured limits
of fetch.unpackLimit/transfer.unpackLimit then index-pack will output
a single line saying "keep\t$sha1\n" to stdout. This line needs to
be captured and retained so we can delete the corresponding .keep
file ("$GIT_DIR/objects/pack/pack-$sha1.keep") once all refs have
been safely updated.
This trick has long been in use with git-fetch.sh and its lower level
helper git-fetch--tool as a way to allow index-pack to save the new
packfile before the refs have been updated and yet avoid a race with
any concurrently running git-repack process. It was unfortunately
lost when git-fetch.sh was converted to pure C and fetch--tool was
no longer being invoked.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Refactor index-pack "keep $sha1" handling for reuse
There is a subtle (but important) linkage between receive-pack and
index-pack that allows index-pack to create a packfile but protect
it from being deleted by a concurrent `git repack -a -d` operation.
The linkage works by having index-pack mark the newly created pack
with a ".keep" file and then it passes the SHA-1 name of that new
packfile to receive-pack along its stdout channel.
The receive-pack process must unkeep the packfile by deleting the
.keep file, but can it can only do so after all elgible refs have
been updated in the receiving repository. This ensures that the
packfile is either kept or its objects are reachable, preventing
a concurrent repacker from deleting the packfile before it can
determine that its objects are actually needed by the repository.
The new builtin-fetch code needs to perform the same actions if
it choose to run index-pack rather than unpack-objects, so I am
moving this code out to its own function where both receive-pack
and fetch-pack are able to invoke it when necessary. The caller
is responsible for deleting the returned ".keep" and freeing the
path if the returned path is not NULL.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Commit walkers need to know the SHA-1 name of any objects they
have been asked to fetch while the native pack transport only
wants to know the names of the remote refs as the remote side
must do the name->SHA-1 translation.
Since we only have three fetch implementations and one of them
(bundle) doesn't even need the name information we can reduce
the code required to perform a fetch by having just one function
and passing of the filtered list of refs to be fetched. Each
transport can then obtain the information it needs from that ref
array to construct its own internal operation state.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Conflicts:
transport.c Signed-off-by: Junio C Hamano <gitster@pobox.com>
Replace custom memory growth allocator with ALLOC_GROW
The ALLOC_GROW macro is a shorter way to implement an array that
grows upon demand as additional items are added to it. We have
mostly standardized upon its use within git and transport.c is
not an exception.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The older git-fetch client did not produce all of this debugging
information to stdout. Most end-users and Porcelain (e.g. StGIT,
git-gui, qgit) do not want to see these low-level details on the
console so they should be removed.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Fix off by one bug in reflog messages written by builtin-fetch
We are adding a space between each argument in the sprintf above
so we must account for this as we update our position within the
reflog message and append in any remaining arguments.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
If we are fetching to a local reference (the so called peer_ref) and
the refspec that created this ref/peer_ref association had started
with '+' we are supposed to allow a non-fast-forward update during
fetch, even if --force was not supplied on the command line. The
builtin-fetch implementation was not honoring this setting as it
was copied from the wrong struct ref instance.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
Thanks to Johannes Schindelin for review and fixes, and Julian
Phillips for the original C translation.
This changes a few small bits of behavior:
branch.<name>.merge is parsed as if it were the lhs of a fetch
refspec, and does not have to exactly match the actual lhs of a
refspec, so long as it is a valid abbreviation for the same ref.
branch.<name>.merge is no longer ignored if the remote is configured
with a branches/* file. Neither behavior is useful, because there can
only be one ref that gets fetched, but this is more consistant.
Also, fetch prints different information to standard out.
Signed-off-by: Daniel Barkalow <barkalow@iabervon.org> Signed-off-by: Junio C Hamano <gitster@pobox.com>
The transport specific stuff was moved into libgit.a, and the
bundle specific stuff will not be left behind.
This is a big code move, with one exception: the function
unbundle() no longer outputs the list of refs. You have to call
list_bundle_refs() yourself for that.
Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de> Signed-off-by: Junio C Hamano <gitster@pobox.com>