Code

git.git
19 years agoAdd some missing commands to the git.txt commands list
Petr Baudis [Tue, 25 Oct 2005 22:03:21 +0000 (00:03 +0200)]
Add some missing commands to the git.txt commands list

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd usage string to git-update-index
Petr Baudis [Tue, 25 Oct 2005 15:26:25 +0000 (17:26 +0200)]
Add usage string to git-update-index

This patch adds usage string to git-update-index, can be printed by the -h
or --help parameter.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDocumentation for git-shell
Petr Baudis [Tue, 25 Oct 2005 21:43:30 +0000 (23:43 +0200)]
Documentation for git-shell

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoCheck another error condition in git-mv
Josef Weidendorfer [Tue, 25 Oct 2005 12:20:45 +0000 (14:20 +0200)]
Check another error condition in git-mv

When moving multiple files at once, it can happen that
files get the same target name, like in

git-mv a/foo b/foo destdir

Both a/foo and b/foo target destdir/foo.

Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agofix daemon.c to compile on OpenBSD
Randal L. Schwartz [Tue, 25 Oct 2005 23:29:09 +0000 (16:29 -0700)]
fix daemon.c to compile on OpenBSD

I can confirm that the following patch lets the current origin
compile on OpenBSD.  If you could apply this until you sort out the
rest of the namespace issue, I would be happy.  Thanks.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRevert recent fetch-pack/upload-pack updates.
Junio C Hamano [Tue, 25 Oct 2005 21:55:24 +0000 (14:55 -0700)]
Revert recent fetch-pack/upload-pack updates.

Let's have it simmer a bit longer in the proposed updates branch
and shake the problems out.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoupload-pack: fix thinko in common-commit finder code.
Junio C Hamano [Sun, 23 Oct 2005 10:02:18 +0000 (03:02 -0700)]
upload-pack: fix thinko in common-commit finder code.

The code to check if we have the object the other side has was bogus
(my fault).

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-fetch-pack: Implement client part of the multi_ack extension
Johannes Schindelin [Sun, 23 Oct 2005 01:40:13 +0000 (03:40 +0200)]
git-fetch-pack: Implement client part of the multi_ack extension

This patch concludes the series, which makes
git-fetch-pack/git-upload-pack negotiate a potentially better set of
common revs. It should make a difference when fetching from a repository
with a few branches.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-fetch-pack: Do not use git-rev-list
Johannes Schindelin [Sun, 23 Oct 2005 01:39:08 +0000 (03:39 +0200)]
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>
19 years agogit-upload-pack: Support sending multiple ACK messages
Johannes Schindelin [Sun, 23 Oct 2005 01:37:45 +0000 (03:37 +0200)]
git-upload-pack: Support sending multiple ACK messages

The current fetch/upload protocol works like this:

- client sends revs it wants to have via "want" messages
- client sends a flush message (message with len 0)
- client sends revs it has via "have" messages
- after one window (32 revs), a flush is sent
- after each subsequent window, a flush is sent, and an ACK/NAK is received.
        (NAK means that server does not have any of the transmitted revs;
         ACK sends also the sha1 of the rev server has)
 - when the first ACK is received, client sends "done", and does not expect
        any further messages

One special case, though:

- if no ACK is received (only NAK's), and client runs out of revs to send,
        "done" is sent, and server sends just one more "NAK"

A smarter scheme, which actually has a chance to detect more than one
common rev, would be to send more than just one ACK. This patch implements
the server side of the following extension to the protocol:

- client sends at least one "want" message with "multi_ack" appended, like

        "want 1234567890123456789012345678901234567890 multi_ack"

- if the server understands that extension, it will send ACK messages for all
        revs it has, not just the first one

- server appends "continue" to the ACK messages like

        "ACK 1234567890123456789012345678901234567890 continue"

        until it has MAX_HAS-1 revs. In this manner, client knows when to
        stop sending revs by checking for the substring "continue" (and
        further knows that server understands multi_ack)

In this manner, the protocol stays backwards compatible, since both client
must send "want ... multi_ack" and server must answer with "ACK ...
continue" to enable the extension.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-upload-pack: More efficient usage of the has_sha1 array
Johannes Schindelin [Sun, 23 Oct 2005 01:36:06 +0000 (03:36 +0200)]
git-upload-pack: More efficient usage of the has_sha1 array

This patch is based on Junio's proposal. It marks parents of common revs
so that they do not clutter up the has_sha1 array.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd git-shell.
Linus Torvalds [Sun, 23 Oct 2005 21:30:45 +0000 (14:30 -0700)]
Add git-shell.

This adds a very git specific restricted shell, that can be
added to /etc/shells and set to the pw_shell in the /etc/passwd
file, to give users ability to push into repositories over ssh
without giving them full interactive shell acount.

[jc: I updated Linus' patch to match what the current sq_quote()
 does.]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoClarify git status output.
Junio C Hamano [Mon, 24 Oct 2005 22:11:47 +0000 (15:11 -0700)]
Clarify git status output.

What we list as "Ignored files" are not "ignored".  Rather, it
is the list of "not listed in the to-be-ignored files, but
exists -- you may be forgetting to add them".

Pointed out by Daniel.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRequire zlib >= 1.2 for RPM.
Andreas Ericsson [Sat, 22 Oct 2005 17:22:58 +0000 (19:22 +0200)]
Require zlib >= 1.2 for RPM.

git-update-index requires zlib >= 1.2, which introduced the *Bound
functions.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd git-mv
Josef Weidendorfer [Sun, 23 Oct 2005 16:15:34 +0000 (18:15 +0200)]
Add git-mv

It supersedes git-rename by adding functionality to move multiple
files, directories or symlinks into another directory.  It also
provides according documentation.

The implementation renames multiple files, using the arguments from
the command line to produce an array of sources and destinations.  In
a first pass, all requested renames are checked for errors, and
overwriting of existing files is only allowed with '-f'.  The actual
renaming is done in a second pass.  This ensures that any error
condition is checked before anything is changed.

Signed-off-by: Josef Weidendorfer <Josef.Weidendorfer@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSilence confusing and false-positive curl error message
Petr Baudis [Fri, 21 Oct 2005 16:18:46 +0000 (18:18 +0200)]
Silence confusing and false-positive curl error message

git-http-fetch spits out curl 404 error message when unable to fetch an object,
but that's confusing since no error really happened and the object is usually
found in a pack it tries right after that. And if the object still cannot be
retrieved, it will say another error message anyway. OTOH other HTTP errors
(403 etc) are likely fatal and the user should be still informed about them.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'fixes'
Junio C Hamano [Sun, 23 Oct 2005 08:20:41 +0000 (01:20 -0700)]
Merge branch 'fixes'

19 years agogit-show-branch: Fix off-by-one error.
Junio C Hamano [Sun, 23 Oct 2005 08:18:42 +0000 (01:18 -0700)]
git-show-branch: Fix off-by-one error.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-rev-list: add "--dense" flag
Linus Torvalds [Fri, 21 Oct 2005 23:40:54 +0000 (16:40 -0700)]
git-rev-list: add "--dense" flag

This is what the recent git-rev-list changes have all been gearing up for.

When we use a path filter to git-rev-list, the new "--dense" flag asks
git-rev-list to compress the history so that it _only_ contains commits
that change files in the path filter.  It also rewrites the parent
information so that tools like "gitk" will see the result as a dense
history tree.

For example, on the current kernel archive:

[torvalds@g5 linux]$ git-rev-list HEAD | wc -l
9904
[torvalds@g5 linux]$ git-rev-list HEAD -- kernel | wc -l
5442
[torvalds@g5 linux]$ git-rev-list --dense HEAD -- kernel | wc -l
356

which shows that while we have almost ten thousand commits, we can prune
down the work to slightly more than half by only following the merges
that are interesting. But further, we can then compress the history to
just 356 entries that actually make changes to the kernel subdirectory.

To see this in action, try something like

gitk --dense -- gitk

to see just the history that affects gitk.  Or, to show that true
parallel development still remains parallel, do

gitk --dense -- daemon.c

which shows some parallel commits in the current git tree.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoTeach git-rev-list to follow just a specified set of files
Linus Torvalds [Fri, 21 Oct 2005 04:25:09 +0000 (21:25 -0700)]
Teach git-rev-list to follow just a specified set of files

This is the first cut at a git-rev-list that knows to ignore commits that
don't change a certain file (or set of files).

NOTE! For now it only prunes _merge_ commits, and follows the parent where
there are no differences in the set of files specified. In the long run,
I'd like to make it re-write the straight-line history too, but for now
the merge simplification is much more fundamentally important (the
rewriting of straight-line history is largely a separate simplification
phase, but the merge simplification needs to happen early if we want to
optimize away unnecessary commit parsing).

If all parents of a merge change some of the files, the merge is left as
is, so the end result is in no way guaranteed to be a linear history, but
it will often be a lot /more/ linear than the full tree, since it prunes
out parents that didn't matter for that set of files.

As an example from the current kernel:

[torvalds@g5 linux]$ git-rev-list HEAD | wc -l
9885
[torvalds@g5 linux]$ git-rev-list HEAD -- Makefile | wc -l
4084
[torvalds@g5 linux]$ git-rev-list HEAD -- drivers/usb | wc -l
5206

and you can also use 'gitk' to more visually see the pruning of the
history tree, with something like

gitk -- drivers/usb

showing a simplified history that tries to follow the first parent in a
merge that is the parent that fully defines drivers/usb/.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSplit up tree diff functions into tree-diff.c library
Linus Torvalds [Fri, 21 Oct 2005 04:05:05 +0000 (21:05 -0700)]
Split up tree diff functions into tree-diff.c library

This makes the tree diff functionality independent of the "git-diff-tree"
program, by splitting the core functionality up into a library file.

This will be needed for when we teach git-rev-list to only follow a
specified set of pathnames, rather than the global revision history.

Most of it is a fairly straightforward code move, but it also involves
some calling convention cleanup, and moving some of the static variables
from diff-tree.c into the options structure.

The actual tree change callback routines also become paramterized by the
diff_options structure, allowing the library functionality to do something
else than just show the diff on stdout.

Right now the only user of this functionality remains git-diff-tree
itself.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAllow git-merge not to commit.
Junio C Hamano [Sat, 22 Oct 2005 11:45:15 +0000 (04:45 -0700)]
Allow git-merge not to commit.

Martin Langhoff wants to use git-merge from outside git-pull and wants
to do further processing; for this, he wants git-merge no to commit
even when it cleanly merges.  I think other script writers would want
something like that as well, so here it is.

Instead of the "merge commit message" parameter (which usually is made
for you by "git-pull" which calls this command), you pass an empty
string to it.  Then it will not update your HEAD -- you can do whatever
you want with the resulting index file, which contains the merge results.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoupload-pack: Increase MAX_HAS.
Junio C Hamano [Sat, 22 Oct 2005 09:28:27 +0000 (02:28 -0700)]
upload-pack: Increase MAX_HAS.

Later round would further improve fetch-pack not to send useless "have",
but in the meantime, increase it to help upload-pack to find more common
commits, as discussed on the list.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix malformatted git-am documentation.
Junio C Hamano [Sat, 22 Oct 2005 03:57:34 +0000 (20:57 -0700)]
Fix malformatted git-am documentation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years ago[PATCH 3/3] Allow running requests to finish after a pull error
Nick Hengeveld [Fri, 21 Oct 2005 19:06:27 +0000 (12:06 -0700)]
[PATCH 3/3] Allow running requests to finish after a pull error

Allow running requests to finish after a pull error

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years ago[PATCH 2/3] Switched back to loading alternates as needed
Nick Hengeveld [Fri, 21 Oct 2005 19:06:20 +0000 (12:06 -0700)]
[PATCH 2/3] Switched back to loading alternates as needed

Switched back to loading alternates as needed

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years ago[PATCH 1/3] Clean up CURL handles in unused request slots
Nick Hengeveld [Fri, 21 Oct 2005 19:06:10 +0000 (12:06 -0700)]
[PATCH 1/3] Clean up CURL handles in unused request slots

Clean up CURL handles in unused request slots

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'fixes'
Junio C Hamano [Fri, 21 Oct 2005 06:21:50 +0000 (23:21 -0700)]
Merge branch 'fixes'

19 years agodaemon.c: remove trailing whitespace.
Junio C Hamano [Fri, 21 Oct 2005 06:19:36 +0000 (23:19 -0700)]
daemon.c: remove trailing whitespace.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix git-daemon argument-parsing bug
H. Peter Anvin [Fri, 21 Oct 2005 01:34:58 +0000 (18:34 -0700)]
Fix git-daemon argument-parsing bug

Fix stupid bug in parsing the --init-timeout option.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate git-daemon's documentation wrt. new options
Petr Baudis [Fri, 21 Oct 2005 00:28:45 +0000 (02:28 +0200)]
Update git-daemon's documentation wrt. new options

New options --timeout, --init-timeout, --export-all and whitelist support
were added to git-daemon, but noone bothered to also add the proper
documentation. This patch aims to fix that.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFinish git-am documentation.
Junio C Hamano [Fri, 21 Oct 2005 05:14:14 +0000 (22:14 -0700)]
Finish git-am documentation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoBrief documentation for the mysterious git-am script
Petr Baudis [Fri, 21 Oct 2005 00:28:42 +0000 (02:28 +0200)]
Brief documentation for the mysterious git-am script

The git-am script is nowhere called and nowhere (including itself)
explained, and the name isn't helpful either. For those like me who will
wonder what is it about, add some documentation stub for it to the
documentation.

I probably got something wrong and I don't feel like investigating all the
options - this is just kind of "emergency" docs.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-rev-parse: pass on "--" flag when required
Linus Torvalds [Fri, 21 Oct 2005 00:16:30 +0000 (17:16 -0700)]
git-rev-parse: pass on "--" flag when required

If rev-parse output includes both flags and files, we should pass on any
"--" marker we see, so that the end result can also tell the difference
between a flag and a filename that begins with '-'.

[jc: merged a later one liner updates from Linus]

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUse sensible domain name (the DNS one) when guessing ident information
Petr Baudis [Fri, 21 Oct 2005 01:57:39 +0000 (03:57 +0200)]
Use sensible domain name (the DNS one) when guessing ident information

Currently, the code would use getdomainname() call, which however returns
something usually unset and not necessarily related at all to the DNS
domain name (it seems to be mostly some scary NIS/YP thing).

This patch changes the code to actually use the DNS domain name, which is
also what tends to be used in emails, and we aim at emails with our ident
code.

Signed-off-by: Petr Baudis <pasky@suse.cz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMake git-cherry-pick in target "all"
Johannes Schindelin [Thu, 20 Oct 2005 15:13:24 +0000 (17:13 +0200)]
Make git-cherry-pick in target "all"

Since git-cherry-pick is simply a copy of git-revert, it can be created
before installing (so that it can be used without installing, too).

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix missing exports in git-am
Junio C Hamano [Fri, 21 Oct 2005 05:31:56 +0000 (22:31 -0700)]
Fix missing exports in git-am

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-daemon poll() spinning out of control
Jens Axboe [Thu, 20 Oct 2005 07:52:32 +0000 (00:52 -0700)]
git-daemon poll() spinning out of control

With the '0' timeout given to poll, it returns instantly without any
events on my system, causing git-daemon to consume all the CPU time. Use
-1 as the timeout so poll() only returns in case of EINTR or actually
events being available.

Signed-off-by: Jens Axboe <axboe@suse.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge /pub/scm/git/git to recover lost side branch
Junio C Hamano [Fri, 21 Oct 2005 00:06:15 +0000 (17:06 -0700)]
Merge /pub/scm/git/git to recover lost side branch

Sorry for the mistake of rewinding something already pushed out.
This recovers the side branch lost by that mistake, specifically
ea5a65a59916503d2a14369c46b1023384d51645 commit.

Signed-off-by: Junio C Hamano <junio@hera.kernel.org>
19 years agoMake sure we barf on ref^{type} failure.
Junio C Hamano [Thu, 20 Oct 2005 05:48:16 +0000 (22:48 -0700)]
Make sure we barf on ref^{type} failure.

Martin Langhoff noticed that ref^0 barfed correctly when we did not
have the commit in a broken repository, but ref^{commit} didn't.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoBe more careful tangling object chains while marking commits.
Junio C Hamano [Thu, 20 Oct 2005 04:55:49 +0000 (21:55 -0700)]
Be more careful tangling object chains while marking commits.

Also Johannes noticed we use parse_object to look up if we know that
object already -- we should just ask the in-core object registry with
lookup_object() for that.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-fetch/push/pull: documentation.
Junio C Hamano [Thu, 20 Oct 2005 04:25:39 +0000 (21:25 -0700)]
git-fetch/push/pull: documentation.

The documentation was lazily sharing the argument description across these
commands.

Lazy may be a way of life, but that does not justify confusing others ;-).

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDo not feed rev-list an invalid SHA1 expression.
Junio C Hamano [Thu, 20 Oct 2005 01:28:17 +0000 (18:28 -0700)]
Do not feed rev-list an invalid SHA1 expression.

The previous round to optimize fetch-pack has a small bug that
feeds SHA1^ ("parent commit") before making sure SHA1 is
actually a commit (or a tag that eventually dereferences to a
commit).  Also it did not help culling the known-to-be-common
parents if the common one was a merge.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years ago[PATCH] Do not send "want" lines for complete objects
Johannes Schindelin [Wed, 19 Oct 2005 23:14:34 +0000 (16:14 -0700)]
[PATCH] Do not send "want" lines for complete objects

It was all good and well to check if all remote refs are complete (local
refs or descendants thereof), but we can just as easily use the same
information to avoid sending "want" lines just for the complete objects in
the case that not all remote refs are complete (or their names differ).

Also, git-fetch-pack does not have to ask for descendants of remote refs
which are complete (for now, git-rev-list is told to ignore only the first
parent). That change also eliminates a code path where a popen()ed handle
was not pclose()ed.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agocount-objects: squelch error from find on sparse object directory.
Junio C Hamano [Wed, 19 Oct 2005 22:01:50 +0000 (15:01 -0700)]
count-objects: squelch error from find on sparse object directory.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-clone: always keep pack sent from remote (documentation).
Junio C Hamano [Wed, 19 Oct 2005 21:43:43 +0000 (14:43 -0700)]
git-clone: always keep pack sent from remote (documentation).

This adjusts the documentation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-clone: always keep pack sent from remote.
Junio C Hamano [Wed, 19 Oct 2005 21:43:43 +0000 (14:43 -0700)]
git-clone: always keep pack sent from remote.

This deprecates --keep and -q flags and always keeps the pack
sent from the remote site.  Corresponding configuration
variables are also removed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDo not ask for objects known to be complete.
Junio C Hamano [Wed, 19 Oct 2005 21:27:02 +0000 (14:27 -0700)]
Do not ask for objects known to be complete.

On top of optimization by Linus not to ask refs that already match, we
can walk our refs and not issue "want" for things that are known to be
reachable from them.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSupport for HTTP transfer timeouts based on transfer speed
Nick Hengeveld [Wed, 19 Oct 2005 21:27:01 +0000 (14:27 -0700)]
Support for HTTP transfer timeouts based on transfer speed

Add configuration settings to abort HTTP requests if the transfer rate
drops below a threshold for a specified length of time.  Environment
variables override config file settings.

Signed-off-by: Nick Hengeveld <nickh@reactrix.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-daemon: timeout, eliminate double DWIM
H. Peter Anvin [Wed, 19 Oct 2005 21:27:01 +0000 (14:27 -0700)]
git-daemon: timeout, eliminate double DWIM

It turns out that not only did git-daemon do DWIM, but git-upload-pack
does as well.  This is bad; security checks have to be performed *after*
canonicalization, not before.

Additionally, the current git-daemon can be trivially DoSed by spewing
SYNs at the target port.

This patch adds a --strict option to git-upload-pack to disable all
DWIM, a --timeout option to git-daemon and git-upload-pack, and an
--init-timeout option to git-daemon (which is typically set to a much
lower value, since the initial request should come immediately from the
client.)

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDo not ask for objects known to be complete.
Junio C Hamano [Wed, 19 Oct 2005 01:42:19 +0000 (18:42 -0700)]
Do not ask for objects known to be complete.

On top of optimization by Linus not to ask refs that already match, we
can walk our refs and not issue "want" for things that are known to be
reachable from them.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoEven when overwriting tags, report if they are changed or not.
Junio C Hamano [Wed, 19 Oct 2005 01:42:14 +0000 (18:42 -0700)]
Even when overwriting tags, report if they are changed or not.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoOptimize common case of git-rev-list
Linus Torvalds [Wed, 19 Oct 2005 01:29:17 +0000 (18:29 -0700)]
Optimize common case of git-rev-list

I took a look at webgit, and it looks like at least for the "projects"
page, the most common operation ends up being basically

git-rev-list --header --parents --max-count=1 HEAD

Now, the thing is, the way "git-rev-list" works, it always keeps on
popping the parents and parsing them in order to build the list of
parents, and it turns out that even though we just want a single commit,
git-rev-list will invariably look up _three_ generations of commits.

It will parse:
 - the commit we want (it obviously needs this)
 - it's parent(s) as part of the "pop_most_recent_commit()" logic
 - it will then pop one of the parents before it notices that it doesn't
   need any more
 - and as part of popping the parent, it will parse the grandparent (again
   due to "pop_most_recent_commit()".

Now, I've strace'd it, and it really is pretty efficient on the whole, but
if things aren't nicely cached, and with long-latency IO, doing those two
extra objects (at a minimum - if the parent is a merge it will be more) is
just wasted time, and potentially a lot of it.

So here's a quick special-case for the trivial case of "just one commit,
and no date-limits or other special rules".

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agorevised^2: git-daemon extra paranoia, and path DWIM
H. Peter Anvin [Wed, 19 Oct 2005 01:26:52 +0000 (18:26 -0700)]
revised^2: git-daemon extra paranoia, and path DWIM

This patch adds some extra paranoia to the git-daemon filename test.  In
particular, it now rejects pathnames containing //; it also adds a
redundant test for pathname absoluteness (belts and suspenders.)

A single / at the end of the path is still permitted, however, and the
.git and /.git append DWIM stuff is now handled in an integrated manner,
which means the resulting path will always be subjected to pathname checks.

Signed-off-by: H. Peter Anvin <hpa@zytor.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRemove unused include.
Junio C Hamano [Tue, 18 Oct 2005 22:41:43 +0000 (15:41 -0700)]
Remove unused include.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-fetch-pack: avoid unnecessary zero packing
Linus Torvalds [Tue, 18 Oct 2005 18:35:17 +0000 (11:35 -0700)]
git-fetch-pack: avoid unnecessary zero packing

If everything is up-to-date locally, we don't need to even ask for a
pack-file from the remote, or try to unpack it.

This is especially important for tags - since the pack-file common commit
logic is based purely on the commit history, it will never be able to find
a common tag, and will thus always end up re-fetching them.

Especially notably, if the tag points to a non-commit (eg a tagged tree),
the pack-file would be unnecessarily big, just because it cannot any most
recent common point between commits for pruning.

Short-circuiting the case where we already have that reference means that
we avoid a lot of these in the common case.

NOTE! This only matches remote ref names against the same local name,
which works well for tags, but is not as generic as it could be. If we
ever need to, we could match against _any_ local ref (if we have it, we
have it), but this "match against same name" is simpler and more
efficient, and covers the common case.

Renaming of refs is common for branch heads, but since those are always
commits, the pack-file generation can optimize that case.

In some cases we might still end up fetching pack-files unnecessarily, but
this at least avoids the re-fetching of tags over and over if you use a
regular

git fetch --tags ...

which was the main reason behind the change.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoNo funny names on cygwin...
Johannes Schindelin [Tue, 18 Oct 2005 18:35:17 +0000 (11:35 -0700)]
No funny names on cygwin...

On FAT/NTFS, filenames cannot contain tabs. So t3300-funny-names would
reliably fail already when trying to create such files.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoIgnore more generated files
Johannes Schindelin [Tue, 18 Oct 2005 18:35:17 +0000 (11:35 -0700)]
Ignore more generated files

Since git-status now shows the "other" files, too, bring .gitignore
up-to-date.

Signed-off-by: Johannes Schindelin <johannes.schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFix cvsimport warning when called without --no-cvs-direct
Johannes Schindelin [Tue, 18 Oct 2005 18:35:16 +0000 (11:35 -0700)]
Fix cvsimport warning when called without --no-cvs-direct

Perl was warning that $opt_p was undefined in that case.

Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-checkout: revert specific paths to either index or a given tree-ish.
Junio C Hamano [Tue, 18 Oct 2005 08:29:27 +0000 (01:29 -0700)]
git-checkout: revert specific paths to either index or a given tree-ish.

When extra paths arguments are given, git-checkout reverts only those
paths to either the version recorded in the index or the version
recorded in the given tree-ish.

This has been on the TODO list for quite a while.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoTeach git-add and git-commit to handle filenames starting with '-'.
Junio C Hamano [Tue, 18 Oct 2005 07:27:50 +0000 (00:27 -0700)]
Teach git-add and git-commit to handle filenames starting with '-'.

Recent '--' fixes to "git diff" by Linus made it possible to specify
filenames that start with '-'.  But in order to do that, you need to
be able to add and commit such file to begin with.

Teach git-add and git-commit to honor the same '--' convention.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoHandle "-" at beginning of filenames, part 3
Linus Torvalds [Tue, 18 Oct 2005 07:16:45 +0000 (00:16 -0700)]
Handle "-" at beginning of filenames, part 3

This fixes the default built-in exec() of "diff" to add a "--" before the
filenames, so that if a filename starts with a "-", the diff program won't
think it's an option.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoTeach "git diff" to handle filenames starting with '-'
Linus Torvalds [Tue, 18 Oct 2005 07:16:45 +0000 (00:16 -0700)]
Teach "git diff" to handle filenames starting with '-'

It adds "--" to the git-diff.sh scripts, to keep any filenames that start
with a "-" from being confused with an option.

But in order to do that, it needs to teach git-diff-files to honor "--".

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAvoid ambiguity between refname and filename in rev-parse
Linus Torvalds [Tue, 18 Oct 2005 07:16:45 +0000 (00:16 -0700)]
Avoid ambiguity between refname and filename in rev-parse

Although it really is very convenient, not requiring explicit
'-r' option to name revs is sometimes ambiguous.

Usually we allow a "--" to say where a filename starts when it
_is_ ambiguous.  However, we fail that at times. In particular,
git-rev-parse fails it.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoForward port the "funny ref avoidance" in clone and fetch from maint branch.
Junio C Hamano [Tue, 18 Oct 2005 04:47:06 +0000 (21:47 -0700)]
Forward port the "funny ref avoidance" in clone and fetch from maint branch.

Somehow I forgot to forward port these fixes.  "git clone" from a
repository prepared with the latest update-server-info would fail
without this patch.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdjust tests for not quoting SP.
Junio C Hamano [Mon, 17 Oct 2005 20:32:14 +0000 (13:32 -0700)]
Adjust tests for not quoting SP.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDo not quote SP.
Junio C Hamano [Mon, 17 Oct 2005 20:32:03 +0000 (13:32 -0700)]
Do not quote SP.

Follow the "encode minimally" principle -- our tools, including
git-apply and git-status, can handle pathnames with embedded SP just
fine.  The only problematic ones are TAB and LF, and we need to quote
the metacharacters introduced for quoting.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-apply: remove unused --show-files flag.
Junio C Hamano [Mon, 17 Oct 2005 05:06:16 +0000 (22:06 -0700)]
git-apply: remove unused --show-files flag.

Linus says he does not use it (and the thinking behind its initial
introduction), and neither Cogito nor StGIT uses it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoupdate-index --index-info: adjust for funny-path quoting.
Junio C Hamano [Sun, 16 Oct 2005 07:39:07 +0000 (00:39 -0700)]
update-index --index-info: adjust for funny-path quoting.

Although the sole current user uses -z to read this, we should be
prepared for somebody to feed non-z format to the command.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoAdd tests for funny pathnames.
Junio C Hamano [Sat, 15 Oct 2005 23:15:49 +0000 (16:15 -0700)]
Add tests for funny pathnames.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate documentation for C-style quoting.
Junio C Hamano [Sat, 15 Oct 2005 04:56:46 +0000 (21:56 -0700)]
Update documentation for C-style quoting.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate git-status to new git-diff-* and git-ls-files output.
Junio C Hamano [Sat, 15 Oct 2005 04:56:46 +0000 (21:56 -0700)]
Update git-status to new git-diff-* and git-ls-files output.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate git-diff-* to use C-style quoting for funny pathnames.
Junio C Hamano [Sat, 15 Oct 2005 04:56:46 +0000 (21:56 -0700)]
Update git-diff-* to use C-style quoting for funny pathnames.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoImprove "git add" again.
Junio C Hamano [Sat, 15 Oct 2005 04:56:46 +0000 (21:56 -0700)]
Improve "git add" again.

This makes it possible to add paths that have funny characters (TAB
and LF) in them, and makes adding many paths more efficient in
general.

New flag "--stdin" to update-index was initially added for different
purpose, but it turns out to be a perfect match for feeding "ls-files
--others -z" output to improve "git add".

It also adds "--verbose" flag to update-index for use with "git add"
command.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate ls-files and ls-tree to use C-style quoting for funny pathnames.
Junio C Hamano [Sat, 15 Oct 2005 04:56:46 +0000 (21:56 -0700)]
Update ls-files and ls-tree to use C-style quoting for funny pathnames.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUpdate git-apply to use C-style quoting for funny pathnames.
Junio C Hamano [Sat, 15 Oct 2005 04:54:52 +0000 (21:54 -0700)]
Update git-apply to use C-style quoting for funny pathnames.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoFunctions to quote and unquote pathnames in C-style.
Junio C Hamano [Sat, 15 Oct 2005 04:54:47 +0000 (21:54 -0700)]
Functions to quote and unquote pathnames in C-style.

Following the list discussion, define two functions, quote_c_style and
unquote_c_style, to help adopting the proposed way for quoting funny
pathname letters for GNU patch.  The rule is described in:

    http://marc.theaimsgroup.com/?l=git&m=112927316408690&w=2

Currently we do not support the leading '!', but we probably should
barf upon seeing it.  Rule B4. is interpreted to require always 3
octal digits in \XYZ notation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'fixes'
Junio C Hamano [Tue, 18 Oct 2005 00:41:37 +0000 (17:41 -0700)]
Merge branch 'fixes'

19 years agogit-checkout-index: documentation updates.
Junio C Hamano [Tue, 18 Oct 2005 00:38:09 +0000 (17:38 -0700)]
git-checkout-index: documentation updates.

Now the behaviour of '-a' has been straightened out, document it.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agomake checkout-index '-a' flag saner.
Linus Torvalds [Tue, 18 Oct 2005 00:32:12 +0000 (17:32 -0700)]
make checkout-index '-a' flag saner.

The original semantics of pretending as if all files were
specified where '-a' appeared and using only the flags given so
far was too confusing.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoref-format documentation.
Junio C Hamano [Mon, 17 Oct 2005 05:41:59 +0000 (22:41 -0700)]
ref-format documentation.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSparse-directory safety fix.
Junio C Hamano [Sun, 16 Oct 2005 21:09:50 +0000 (14:09 -0700)]
Sparse-directory safety fix.

This will be removed when merging the second phase of Linus' "Create
object subdirectories on demand" change anyway, but the code to
recreate the empty .git/objects/??/ directory was confused.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'fixes'
Junio C Hamano [Sun, 16 Oct 2005 19:23:59 +0000 (12:23 -0700)]
Merge branch 'fixes'

19 years agoWe do not depend on patch.
Junio C Hamano [Sun, 16 Oct 2005 18:59:13 +0000 (11:59 -0700)]
We do not depend on patch.

Deb packaging claim we depend on patch, but I think we use git-apply
where it matters.  When a patch does not apply with git-apply, using
GNU patch still is helpful sometimes.  So demote it from "Depends" to
"Suggests".

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'svn' of http://netz.smurf.noris.de/git/git
Junio C Hamano [Sun, 16 Oct 2005 18:55:35 +0000 (11:55 -0700)]
Merge branch 'svn' of netz.smurf.noris.de/git/git

[jc: I have my pre-commit hook enabled to catch trailing whitespaces,
 and fixed them up while merging.]

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agosvn commit: re-word the exit-due-to-memory-leak message
Matthias Urlichs [Sun, 16 Oct 2005 17:57:38 +0000 (19:57 +0200)]
svn commit: re-word the exit-due-to-memory-leak message

Reworded the exit message, as per Kalle Valo's suggestion (but shorter).

Signed-Off-By: Matthias Urlichs <smurf@smurf.noris.de>
19 years agoMakefile entry for git-svnimport contained a small typo.
Kalle Valo [Sun, 16 Oct 2005 17:37:25 +0000 (19:37 +0200)]
Makefile entry for git-svnimport contained a small typo.

Signed-Off-By: Matthias Urlichs <smurf@smurf.noris.de>
19 years agoSquelch compiler warnings from connect.c
Junio C Hamano [Sun, 16 Oct 2005 07:25:26 +0000 (00:25 -0700)]
Squelch compiler warnings from connect.c

Forgot to include necessary header file to get the function declaration.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoDisable hooks during tests.
Junio C Hamano [Sun, 16 Oct 2005 07:24:34 +0000 (00:24 -0700)]
Disable hooks during tests.

Individual tests for hooks would want to have their own tests when
written.  Also we should not pick up from random templates the user
happens to have.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSparse fixes for http-fetch
Peter Hagervall [Sun, 16 Oct 2005 07:01:08 +0000 (00:01 -0700)]
Sparse fixes for http-fetch

This patch cleans out all sparse warnings from http-fetch.c

I'm a bit uncomfortable with adding extra #ifdefs to avoid either
'mixing declaration with code' or 'unused variable' warnings, but I
figured that since those functions are already littered with #ifdefs I
might just get away with it. Comments?

[jc: I adjusted Peter's patch to address uncomfortableness issues.]

Signed-off-by: Peter Hagervall <hager@cs.umu.se>
Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agowhatchanged: document -m option from git-diff-tree.
Junio C Hamano [Sun, 16 Oct 2005 06:49:27 +0000 (23:49 -0700)]
whatchanged: document -m option from git-diff-tree.

The documentation for git-whatchanged is meant to describe only
the most frequently used options from git-diff-tree.  Because "why
doesn't it show merges" was asked more than once, we'd better
describe '-m' option there.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoShow peeled onion from upload-pack and server-info.
Junio C Hamano [Fri, 14 Oct 2005 01:57:40 +0000 (18:57 -0700)]
Show peeled onion from upload-pack and server-info.

This updates git-ls-remote to show SHA1 names of objects that are
referred by tags, in the "ref^{}" notation.

This would make git-findtags (without -t flag) almost trivial.

    git-peek-remote . |
    sed -ne "s:^$target "'refs/tags/\(.*\)^{}$:\1:p'

Also Pasky could do:

    git-ls-remote --tags $remote |
    sed -ne 's:\( refs/tags/.*\)^{}$:\1:p'

to find out what object each of the remote tags refers to, and
if he has one locally, run "git-fetch $remote tag $tagname" to
automatically catch up with the upstream tags.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoIntroduce notation "ref^{type}".
Junio C Hamano [Fri, 14 Oct 2005 01:57:40 +0000 (18:57 -0700)]
Introduce notation "ref^{type}".

Existing "tagname^0" notation means "dereference tag zero or more
times until you cannot dereference it anymore, and make sure it is a
commit -- otherwise barf".  But tags do not necessarily reference
commit objects.

This commit introduces a bit more generalized notation, "ref^{type}".
Existing "ref^0" is a shorthand for "ref^{commit}".  If the type
is empty, it just dereferences tags until it hits a non-tag object.

With this, "git-rev-parse --verify 'junio-gpg-pub^{}'" shows the blob
object name -- there is no need to manually read the tag object and
find out the object name anymore.

"git-rev-parse --verify 'HEAD^{tree}'" can be used to find out the
tree object name of the HEAD commit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoIgnore funny refname sent from remote
Junio C Hamano [Fri, 14 Oct 2005 01:57:40 +0000 (18:57 -0700)]
Ignore funny refname sent from remote

This allows the remote side (most notably, upload-pack) to show
additional information without affecting the downloader.  Peek-remote
does not ignore them -- this is to make it useful for Pasky's
automatic tag following.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoRefuse to create funny refs in clone-pack, git-fetch and receive-pack.
Junio C Hamano [Fri, 14 Oct 2005 01:57:39 +0000 (18:57 -0700)]
Refuse to create funny refs in clone-pack, git-fetch and receive-pack.

Using git-check-ref-format, make sure we do not create refs with
funny names when cloning from elsewhere (clone-pack), fast forwarding
local heads (git-fetch), or somebody pushes into us (receive-pack).

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agogit-check-ref-format: reject funny ref names.
Junio C Hamano [Fri, 14 Oct 2005 01:57:39 +0000 (18:57 -0700)]
git-check-ref-format: reject funny ref names.

Update check_ref_format() function to reject ref names that:

 * has a path component that begins with a ".", or
 * has a double dots "..", or
 * has ASCII control character, "~", "^", ":" or SP, anywhere, or
 * ends with a "/".

Use it in 'git-checkout -b', 'git-branch', and 'git-tag' to make sure
that newly created refs are well-formed.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoMerge branch 'fixes'
Junio C Hamano [Sat, 15 Oct 2005 18:19:09 +0000 (11:19 -0700)]
Merge branch 'fixes'

19 years agoShow curl error a bit better.
Junio C Hamano [Sat, 15 Oct 2005 18:13:55 +0000 (11:13 -0700)]
Show curl error a bit better.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoSome curl versions lack curl_easy_duphandle()
Johannes Schindelin [Sat, 15 Oct 2005 18:10:46 +0000 (11:10 -0700)]
Some curl versions lack curl_easy_duphandle()

Hi,

On Fri, 14 Oct 2005, Junio C Hamano wrote:

> Johannes Schindelin <Johannes.Schindelin@gmx.de> writes:
>
> > This patch looks bigger than it really is: The code to get the
> > default handle was refactored into a function, and is called
> > instead of curl_easy_duphandle() if that does not exist.
>
> I'd like to take Nick's config file patch first, which
> unfortunately interferes with your patch.  I'd hate to ask you
> this, but could you rebase it on top of Nick's patch, [...]

No need to hate it. Here comes the rebased patch, and this time, I
actually tested it a bit.

Signed-off-by: Junio C Hamano <junkio@cox.net>
19 years agoUnlocalized isspace and friends
Linus Torvalds [Thu, 13 Oct 2005 18:03:18 +0000 (11:03 -0700)]
Unlocalized isspace and friends

Do our own ctype.h, just to get the sane semantics: we want
locale-independence, _and_ we want the right signed behaviour. Plus we
only use a very small subset of ctype.h anyway (isspace, isalpha,
isdigit and isalnum).

Signed-off-by: Junio C Hamano <junkio@cox.net>