git-fetch: fail if specified refspec does not match remote.
'git-fetch remote no-such-ref' succeeded without fetching any
ref from the remote. Detect such case and report an error.
Note that this makes 'git-fetch remote master master' to fail,
because the remote branch 'master' matches the first refspec,
and the second refspec is left unmatched, which is detected by
the error checking logic. This is somewhat unintuitive, but
giving the same refspec more than once to git-fetch is useless
in any case so it should not be much of a problem. I'd accept a
patch to change this if somebody cares enough, though.
Josef Weidendorfer points out that git-clone documentation does not
mention the initial copying of remote branch heads into corresponding
local branches. Also clarify the purpose of the ref mappings description
in the "remotes" file and recommended workflow.
We checked the result of patch application for full permission bits,
when the only thing we cared about was to make sure the executable
bit was correctly set.
If we let cpio to create the leading directories implicitly,
it ends up having funny perm bits (GNU cpio 2.5 and 2.6, at least).
This leaves .git/object/?? directories readable only by the owner.
We do not accept multiple <refspecs> on one Pull:/Push: line
right now (we could lift this tentative workaround for the
broken refnames), but we have always accepted multiple such
lines, so use that form in the examples and discussion.
Also explicitly mention that Octopus is made only with an
explicit command line request and never from Pull: lines.
Clarified and added notes for pull/push refspecs.
Converted to back-ticks for literal text examples.
[jc: Also fixed git-pull description that still talked about its
calling git-resolve or git-octopus (we do not anymore; instead
we just call git-merge). BTW, I am reasonably impressed by how
well "git-am -3" applied this patch, which had some conflicts
because I've updated the documentation somewhat.]
Signed-off-by: Jon Loeliger <jdl@freescale.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
Having the temporary files in the working tree root when making
tags is not as bad because it does not involve 'git status' as
the git-commit case, but this makes things more consistent.
Remove the temp file if it is empty after the request has failed
After using cg-update to pull, empty files named *.temp are left in
the various subdirectories of .git/objects/. These are created by
git-http-fetch to hold data as it's being fetched from the remote
repository. They are left behind after a transfer error so that the
next time git-http-fetch runs it can pick up where it left off. If
they're empty though, it would make more sense to delete them rather
than leaving them behind for the next attempt.
Signed-off-by: Nick Hengeveld <nickh@reactrix.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-merge expects this check to be done appropriately by the
merge strategy backends. In the case of merge-ours strategy,
the resulting tree comes what we have in the index file, so it
must match the current HEAD; otherwise it would not be "ours"
merge.
This adds the coolest merge strategy ever, "ours". It can take
arbitrary number of foreign heads and merge them into the
current branch, with the resulting tree always taken from our
branch head, hence its name.
What this means is that you can declare that the current branch
supersedes the development histories of other branches using
this merge strategy.
With --no-commit flag, git-pull will perform the merge but pretends as
if the merge needed a hand resolve even if automerge cleanly resolves,
to give the user a chance to add further changes and edit the commit
message.
One caller of deref_tag() was not careful enough to make sure
what deref_tag() returned was not NULL (i.e. we found a tag
object that points at an object we do not have). Fix it, and
warn about refs that point at such an incomplete tag where
needed.
For everyone cursed by dos/windows line endings (aka CRLF):
The code reading the .gitignore files (excludes and excludes per
directory) leaves \r in the patterns, which causes fnmatch to fail for
no obvious reason. Just remove a "\r" preceding a "\n"
unconditionally.
"git-checkout -b frotz/nitfol master" failed to create
$GIT_DIR/refs/heads/frotz/nitfol but went ahead and updated
$GIT_DIR/HEAD to point at it, resulting in a corrupt repository.
Exit when we cannot create the new branch with an error status.
While we are at it, there is no reason to forbid subdirectories
in refs/heads, so make sure we handle that correctly.
The newly cloned repository by default had .git/remotes/origin
set up to track the remote master to origin, but forgot to
create the origin branch ourselves. Also it hardcoded the
assumption that the remote HEAD points at "master", which may
not always be true.
-P:: <cvsps-output-file>
Instead of calling cvsps, read the provided cvsps output file. Useful
for debugging or when cvsps is being handled outside cvsimport.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz> Signed-off-by: Junio C Hamano <junkio@cox.net>
cvsimport: catch error condition where cvs host disappears
Add error handling for cases where the cvs server goes away unexpectedly.
While I don't know why the cvs server is so erratic, we should definitely
exit here before committing bogus files.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz> Signed-off-by: Junio C Hamano <junkio@cox.net>
Do not install backward compatibility links anymore.
This is a companion patch to 4f9dcf7e5cf6c82455925102d315daf3b833e6d6
which stops mentioning the old command names. As promised, we do not
install symlinks to let people use backward compatibility names anymore.
cmd-rename.sh script is still shipped to help people who installed
previous git by hand to clean up the leftover symlinks.
Needed because generating a target paths will add another slash.
This fixes e.g. "git-mv file dir/", which removed "file" from
version control by renaming it to "dir//file", as
git-update-index does not accept such paths.
Thanks goes to Ben Lau for noting this bug.
Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
That notice was added by me for the emergency documentation, but Junio
already expanded it to a full-fledged manual page. This patch removes
the notice.
Signed-off-by: Petr Baudis <pasky@suse.cz> Signed-off-by: Junio C Hamano <junkio@cox.net>
Remove 'Previously this command was known as ...' messages.
For a 1.0 release, there is no need to maintain the
historical "Previously this command was known as..."
information on the doc splash page. It is noise;
command names should stand on their own now.
Signed-off-by: Jon Loeliger <jdl@freescale.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
Document the use of "current directory" as pull source.
The repository to pull from can be a local repository, and as a
special case the current directory can be specified to perform
merges across local branches.
Add examples for git-log documentation and others.
I don't think people really follow the links or think very abstractly at
all in the first place.
So I was thinking more of some explicit examples. I actually think every
command should have an example in the man-page, and hey, here's a patch to
start things off.
Of course, I'm not exactly "Mr Documentation", and I don't know that this
is the prettiest way to do this, but I checked that the resulting html and
man-page seems at least reasonable.
And hey, if the examples look like each other, that's just because I'm
also not "Mr Imagination".
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
The require statement at the top of git-svnimport seems to confuse
rpmbuild dependency generation. It uses the newer notation "v5.8.0",
and rpm ends up requiring "perl(v5.8.0)", while we would want it to
say something like "perl >= 0:5.008".
Ryan suggests old-style "require 5.008" might fix this problem, so
here it is.
* clone request over git native protocol from a repository with
too many refs did not work; this has been fixed.
* git-daemon got safer for kernel.org use [HPA].
* Extended SHA1 parser was not enforcing uniqueness for
abbreviated SHA1; this has been fixed.
* http transport does not barf on funny characters in URL.
* The ref naming restrictions have been formalized and the
coreish refuses to create funny refs; we still need to audit
importers. See git-check-ref-format(1).
New Features and Commands
~~~~~~~~~~~~~~~~~~~~~~~~~
* .git/config file as a per-repository configuration mechanism,
and some commands understand it [Linus]. See
git(7).
* The core.filemode configuration item can be used to make us a
bit more FAT friendly. See git(7).
* The extended SHA1 notation acquired Peel-the-onion operator
^{type} and ^{}. See git-rev-parse(1).
* SVN importer [Matthias]. See git-svnimport(1).
* .git/objects/[0-9a-f]{2} directories are created on demand,
and removed when becomes empty after prune-packed [Linus].
* Filenames output from various commands without -z option are
quoted when they embed funny characters (TAB and LF) using
C-style quoting within double-quotes, to match the proposed
GNU diff/patch notation [me, but many people contributed in
the discussion].
* git-mv is expected to be a better replacement for git-rename.
While the latter has two parameter restriction, it acts more
like the regular 'mv' that can move multiple things to one
destinatino directory [Josef Weidendorfer].
* git-checkout can take filenames to revert the changes to
them. See git-checkout(1)
* The new program git-am is a replacement for git-applymbox that
has saner command line options and a bit easier to use when a
patch does not apply cleanly.
* git-ls-remote can show unwrapped onions using ^{} notation, to
help Cogito to track tags.
* git-merge-recursive backend can merge unrelated projects.
* git-clone over native transport leaves the result packed.
* git-http-fetch issues multiple requests in parallel when
underlying cURL library supports it [Nick and Daniel].
* git-fetch-pack and git-upload-pack try harder to figure out
better common commits [Johannes].
* git-read-tree -u removes a directory when it makes it empty.
* git-diff-* records abbreviated SHA1 names of original and
resulting blob; this sometimes helps to apply otherwise an
unapplicable patch by falling back to 3-way merge.
* git-format-patch now takes series of from..to rev ranges and
with '-m --stdout', writes them out to the standard output.
This can be piped to 'git-am' to implement cheaper
cherry-picking.
* git-tag takes '-u' to specify the tag signer identity [Linus].
* git-rev-list can take optional pathspecs to skip commits that
do not touch them (--dense) [Linus].
* Comes with new and improved gitk [Paulus and Linus].
The latest init-db does not create .git/objects/??/ directories
anymore and expects the users of the repository to create them
as they are needed. local-fetch was not taught about it, which
broke local cloning with Cogito.
On Linux, "mktemp tmp-XXXX" will not work. Also, redirect stderr on which,
so it does not complain too loudly. After all, this test should only be
executed when old binaries are available.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
This implements three things (trying very hard to be backwards
compatible):
It sends the "multi_ack" capability via the mechanism proposed by
Sergey Vlasov.
When the client sends "multi_ack" with at least one "want", multi_ack
is enabled.
When multi_ack is enabled, "continue" is appended to each "ACK" until
either the server can not store more refs, or "done" is received.
In contrast to the original protocol, as long as "continue" is sent,
flushes are answered by a "NAK" (not just until an "ACK" was sent),
and if "continue" was sent at least once, the last message is an
"ACK" without "continue".
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
This patch implements the client side of backward compatible upload-pack
protocol extension, <20051027141619.0e8029f2.vsu@altlinux.ru> by Sergey.
The updated server can append "server_capabilities" which is supposed
to be a string containing space separated features of the server, after
one of elements in the initial list of SHA1-refname line, hidden with
an embedded NUL.
After get_remote_heads(), check if the server supports the feature like
if (server_supports("multi_ack"))
do_something();
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
Implement an interoperability test for fetch-pack/upload-pack
The next patches will extend the pack protocol. This test assures that this
extension is compatible to earlier versions of git-fetch-pack/git-upload-pack.
All you need to do to take advantage of this test, is to install older
known-to-be-working binaries in the path as "old-git-fetch-pack" and
"old-git-upload-pack".
Note that the warning when testing with old-git-fetch-pack is to be
expected (it just says that the old version was not taking advantage
of all the information which the server sent).
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
When git-fetch-pack gets the remote refs, it does not need to filter them
right away, but it can see which refs are common (taking advantage of the
patch which makes git-fetch-pack not use git-rev-list).
This means that we ask get_remote_heads() to return all remote refs,
including the funny refs, and filtering them with a separate function later.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
Subject: [PATCH] git-fetch-pack: Do not use git-rev-list
The code used to call git-rev-list to enumerate the local revisions.
A disadvantage of that method was that git-rev-list, lacking a
control apart from the command line, would happily enumerate
ancestors of acknowledged common commits, which was just taking
unnecessary bandwidth.
Therefore, do not use git-rev-list on the fetching side, but rather
construct the list on the go. Send the revisions starting from the
local heads, ignoring the revisions known to be common.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
The new option, --numstat, shows number of inserted and deleted
lines for each path. It is similar to --stat output but is
meant to be more machine friendly by giving number of added and
deleted lines and unabbreviated paths.
- we don't allow "." and ".." as components of a refname. Thus get_sha1()
will not accept "./refname" as being the same as "refname" any more.
- git-rev-parse stops doing revision translation after seeing a pathname,
to match the brhaviour of all the tools (once we see a pathname,
everything else will also be parsed as a pathname).
Basically, if you did
git log *
and "gitk" was somewhere in the "*", we don't want to replace the filename
"gitk" with the SHA1 of the branch with the same name.
Of course, if there is any change of ambiguity, you should always use "--"
to make it explicit what are filenames and what are revisions, but this
makes the normal cases sane. The refname rule also means that instead of
the "--", you can do the same thing we're used to doing with filenames
that start with a slash: use "./filename" instead, and now it's a
filename, not an option (and not a revision).
So "git log ./*.c" is now actually a perfectly valid thing to do, even if
the first C-file might have the same name as a branch.
where the "./gitk" isn't seen as a revision, and the second "gitk" is a
filename simply because we've seen filenames already, and thus stopped
doing revision parsing.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
The git philosophy when it comes to disk accesses is "Laugh in the face of
danger".
Notably, since we never modify an existing object, we don't really care
that deeply about flushing things to disk, since even if the machine
crashes in the middle of a git operation, you can never really have lost
any old work. At most, you'd need to figure out the proper heads (which
git-fsck-objects can do for you) and re-do the operation.
However, there's two exceptions to this: pruning and repacking. Those
operations will actually _delete_ old objects that they know about in
other ways (ie that they just repacked, or that they have found in other
places).
However, since they actually modify old state, we should thus be a bit
more careful about them. If the machine crashes and the duplicate new
objects haven't been flushed to disk, you can actually be in trouble.
This is trivially stupid about it by calling "sync" before removing the
objects. Not very smart, but we're talking about special operations than
are usually done once a week if that.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Documentation changes to recursive option for git-diff-tree
Update docs and usages regarding '-r' recursive option for git-diff-tree.
Remove '-r' from common diff options, mention it only for git-diff-tree.
Remove one extraneous use of '-r' with git-diff-files in get-merge.sh.
Sync the synopsis and usage string for git-diff-tree.
Signed-off-by: Chris Shoemaker <c.shoemaker at cox.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
Pavel Roskin wondered what the SHA1 output at the beginning of
git-diff-tree was about. The only consumer of that information
so far is this git-patch-id command, which was inadequately
documented.
This removes the unoptimization. The previous round does not mind
missing fan-out directories, but still makes sure they exist, lest
older versions choke on a repository created/packed by it.
This round does not play that nicely anymore -- empty fan-out
directories are not created by init-db, and will stay removed by
prune-packed. The prune command also removes empty fan-out directories.
[PATCH] Make "gitk" work better with dense revlists
To generate the diff for a commit, gitk used to do
git-diff-tree -p -C $p $id
(and same thing to generate filenames, except using just "-r" there) which
does actually generate the diff from the parent to the $id, exactly like
it meant to do.
However, that really sucks with --dense, where the "parent" information
has all been rewritten to point to the previous commit. The diff actually
works exactly right, but now it's the diff of the _whole_ sequence of
commits all the way to the previous commit that last changed the file(s)
that we are looking at.
And that's really not what we want 99.9% of the time, even if it may be
perfectly sensible. Not only will the diff not actually match the commit
message, but it will usually be _huge_, and all of it will be totally
uninteresting to us, since we were only interested in a particular set of
files.
It also doesn't match what we do when we write the patch to a file.
So this makes gitk just show the diff of _that_ commit.
We might even want to have some way to limit the diff to only the
filenames we're interested in, but it's often nice to see what else
changed at the same time, so that's secondary.
The merge diff handling is left alone, although I think that should also
be changed to only look at what that _particular_ merge did, not what it
did when compared to the faked-out parents.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Paul Mackerras <paulus@samba.org>
What happens is that the new logic decides that if it can't look up a
commit reference (ie "get_commit_reference()" returns NULL), the thing
must be a pathname.
Fair enough.
But wrong.
The thing is, it may be a perfectly fine ref that _isn't_ a commit. In
git, you have a tag that points to your PGP key, and in the kernel, I have
a tag that points to a tree (and a direct ref that points to that tree
too, for that matter).
So the rule is (as for all the other programs that mix revs and pathnames)
not that we only accept commit references, but _any_ valid object ref.
If the object then isn't a commit ref, git-rev-list will either ignore it,
or add it to the list of non-commit objects (if using "--objects").
The solution is to move the "get_sha1()" out of get_commit_reference(),
and into the callers. In fact, we already _have_ the SHA1 in the case of
the handle_all() loop, since for_each_ref() will have done it for us, so
this is the correct thing to do anyway.
This patch (on top of the original one) does exactly that.
git-rev-list: make --dense the default (and introduce "--sparse")
This actually does three things:
- make "--dense" the default for git-rev-list. Since dense is a no-op if
no filenames are given, this doesn't actually change any historical
behaviour, but it's logically the right default (if we want to prune on
filenames, do it fully. The sparse "merge-only" thing may be useful,
but it's not what you'd normally expect)
- make "git-rev-parse" show the default revision control before it shows
any pathnames.
This was a real bug, but nobody would ever have noticed, because
the default thing tends to only make sense for git-rev-list, and
git-rev-list didn't use to take pathnames.
- it changes "git-rev-list" to match the other commands that take a mix
of revisions and filenames - it no longer requires the "--" before
filenames (although you still need to do it if a filename could be
confused with a revision name, eg "gitk" in the git archive)
This all just makes for much more pleasant and obvous usage. Just doing a
gitk t/
does the obvious thing: it will show the history as it concerns the "t/"
subdirectory.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-name-rev tries to find nice symbolic names for commits. It does so by
walking the commits from the refs. When the symbolic name is ambiguous, the
following heuristic is applied: Try to avoid too many ~'s, and if two ambiguous
names have the same count of ~'s, take the one whose last number is smaller.
With "--tags", the names are derived only from tags.
With "--stdin", the stdin is parsed, and after every sha1 for which a name
could be found, the name is appended. (Try "git log | git name-rev --stdin".)
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-pack-objects can reuse pack files stored in $GIT_DIR/pack-cache
directory, when a necessary pack is found. This is hopefully useful
when upload-pack (called from git-daemon) is expected to receive
requests for the same set of objects many times (e.g full cloning
request of any project, or updates from the set of heads previous day
to the latest for a slow moving project).
Currently git-pack-objects does *not* keep pack files it creates for
reusing. It might be useful to add --update-cache option to it,
which would allow it store pack files it created in the pack-cache
directory, and prune rarely used ones from it.
Fix what to do and how to detect when hardlinking fails
Recent FAT workaround caused compilation trouble on OpenBSD;
different platforms use different error codes when we try to
hardlink the temporary file to its final location. Existing
Coda hack also checks its own error code, but the thing is,
the case we care about is if link failed for a reason other
than that the final file has already existed (which would be
normal, or it could mean collision). So just check the error
code against EEXIST.
upload-pack would set create_full_pack=1 if nr_has==0, but would ask later
if nr_needs<MAX_NEEDS. If that proves true, it would ignore create_full_pack,
and arguments would be written into unreserved memory.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
This makes sure what the other end asks for are among what we
offered to give them. Otherwise we would end up running
git-rev-list with 20-byte nonsense, only to find it either die
(because the object was not found) or waste time (because we
ended up serving that phony 'client').
Also avoid wasting needs_sha1 pool to record duplicates, and
detect cloning requests better.
[this used to be on top of Johannes fetch-pack enhancements,
which we are rewinding it for further testing for now, so
the commit is rebased.]
Work around missing hard links on FAT formatted media
FAT -- like Coda -- does not like cross-directory hard links. To be
precise, FAT does not like links at all. But links are not needed either.
So get rid of them.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>
create_symref: if symlink fails, fall back to writing a "symbolic ref"
There are filesystems out there which do not understand symlinks, even if
the OS is perfectly capable of writing them. So, do not fail right away,
but try to write a symbolic ref first. If that fails, you can die().
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de> Signed-off-by: Junio C Hamano <junkio@cox.net>