Don't leak file descriptors from unavailable pack files.
If open_packed_git failed it may have been because the packfile
actually exists and is readable, but some sort of verification
did not pass. In this case open_packed_git left pack_fd filled
in, as the file descriptor is valid. We don't want to leak the
file descriptor, nor do we want to allow someone in the future
to use this packed_git.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Don't find objects in packs which aren't available anymore.
Matthias Lederhofer identified a race condition where a Git reader
process was able to locate an object in a packed_git index, but
was then preempted while a `git repack -a -d` ran and completed.
By the time the reader was able to seek in the packfile to get the
object data, the packfile no longer existed on disk.
In this particular case the reader process did not attempt to
open the packfile before it was deleted, so it did not already
have the pack_fd field popuplated. With the packfile itself gone,
there was no way for the reader to open it and fetch the data.
I'm fixing the race condition by teaching find_pack_entry to ignore
a packed_git whose packfile is not currently open and which cannot
be opened. If none of the currently known packs can supply the
object, we will return 0 and the caller will decide the object is
not available. If this is the first attempt at finding an object,
the caller will reprepare_packed_git and try again. If it was
the second attempt, the caller will typically return NULL back,
and an error message about a missing object will be reported.
This patch does not address the situation of a reader which is
being starved out by a tight sequence of `git repack -a -d` runs.
In this particular case the reader will try twice, probably fail
both times, and declare the object in question cannot be found.
As it is highly unlikely that a real world `git repack -a -d` can
complete faster than a reader can open a packfile, so I don't think
this is a huge concern.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Because I want to reuse open_packed_git in a context where I don't
want the process to die if the packfile in question is bogus, I'm
changing its behavior to return error("...") rather than die("...")
when it detects something is wrong with the packfile it was given.
Right now we still must die out of use_pack should open_packed_git
fail, as none of use_pack's callers are prepared to handle a failure
from that function.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
After staring at the comment and the associated for loop, I
realized the comment was completely bogus. The section of
code its talking about is trying to avoid duplicate mapping
of the same packfile.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Cleanup prepare_packed_git_one to reuse install_packed_git.
There is little point in having the linked list insertion code
appearing in install_packed_git, and then again just 30 lines
further down in the same file.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Teach 'git remote' how to cleanup stale tracking branches.
Since it can be annoying to manually cleanup 40 tracking branches
which were removed by the remote system, 'git remote prune <n>'
can now be used to delete any tracking branches under <n> which
are no longer available on the remote system.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
I want to reuse the stale branch detection to implement a new
'git remote prune' subcommand. Easiest way to do that is to use
the same logic that 'git remote show' uses to determine the stale
tracking branches, then delete those.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
prevent HEAD reflog to be interpreted as current branch reflog
The work in progress to enable separate reflog for HEAD will make it
independent from reflog of any branch HEAD might be pointing to. In
the mean time disallow HEAD@{...} until that work is completed. Otherwise
people might get used to the current behavior which makes HEAD@{...} an
alias for <current_branch>@{...} which won't be the case later.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Converts one use of git-checkout in git-bisect not to say "switching
to branch". It looks like all the other cases it is friendlier to
give notice to the end user.
This updates "git-checkout" to report which branch you are
switching to. Especially for people who do not use __git_ps1
from contrib/completion/git-completion.bash this would give a
friendlier feedback of what is going on, and should make the
reminder message much less scary.
Here is a sample session (the prompt tells which branch I am on).
* I have some local modification and realize that the change deserves
to be on its own new topic branch.
* So I switch to a new branch. I get a listing of local modifications
and assuring "Switched to a new branch" message.
[git.git (master)]$ git checkout -b jc/checkout
M git-checkout.sh
Switched to a new branch "jc/checkout"
* If I switch back to "master", I get essentially the same.
[git.git (jc/checkout)]$ git checkout master
M git-checkout.sh
Switched to branch "master"
* Detaching head would say which commit I am at and reminds me that
I am not on any branch (not that I would detach my HEAD while keeping
precious local changes around in any real-world workflow -- this is
just a sample session).
[git.git (master)]$ git checkout master^
M git-checkout.sh
Note: you are not on any branch and are at commit "master^"
If you want to create a new branch from this checkout, you may do so
(now or later) by using -b with the checkout command again. Example:
git checkout -b <new_branch_name>
* Coming back to an attached state can lose the detached HEAD, so
I get warned and stopped.
[git.git]$ git checkout master
You are not on any branch and switching to branch 'master'
may lose your changes. At this point, you can do one of two things:
(1) Decide it is Ok and say 'git checkout -f master';
(2) Start a new branch from the current commit, by saying
'git checkout -b <branch-name>'.
Leaving your HEAD detached; not switching to branch 'master'.
* Moving around while my HEAD is detached is Ok. I still get the list
of local modifications.
[git.git]$ git checkout master^0
M git-checkout.sh
* The previous step that switched to the tip commit is an obscure but
useful trick. My HEAD is still detached but now it is pointed at by
an existing ref, so I can come back safely.
[git.git]$ git checkout master
M git-checkout.sh
Switched to branch "master"
Merge branch 'master' of git://linux-nfs.org/~bfields/git
This is in the hope of giving JBF's user-manual wider exposure.
I am not very happy with trailing whitespaces in the new
document, but let's not worry too much about the formatting
issues for now, but concentrate more on the structure and the
contents.
t9200: do not test -x bit if the filesystem does not support it.
The last test in t9200 wants to see if executable bit is
retained, which has no chance of succeeding on a filesystem that
does not handle executable bit correctly.
For the purpose of this test we do not really care if the paths
are in latin-1, but people on Cygwin seem to be having problem
on foreign-looking pathnames that do not play well with their
locale.
Let's try to re-code them in UTF-8 and see who screams,
thanks, or reports no-improvements.
Recent commit ae1dffcb28ee89a23f8d2747be65e17c8eab1690 by Junio
changed the way --upload-pack was passed around between clone,
fetch and ls-remote and modified the handling of the command
line parameter parsing.
Unfortunately FreeBSD 6.1 insists that the expression
Apparently if we are unable to parse an object update-server-info
coredumps, as it doesn't bother to check the return value of its
call to parse_object.
Instead of coredumping, skip the ref.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
This is not meant to frighten people or even to suggest they might be
doing something wrong, but rather to notify them of a state change and
provide a likely option in the case this state was entered by mistake.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
blameview: Use git-cat-file to read the file content.
Fix blameview to use git-cat-file to read the file content.
This make sure we show the right content when we have modified
file in the working directory which is not committed.
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
lock_any_ref_for_update(): do not accept malformatted refs.
We used to use lock_any_ref_for_update() because the command
needs to also update HEAD (which is not under refs/, so
lock_ref_sha1() cannot be used). The function however did not
check for refs with illegal characters in them.
Use check_ref_format() to catch malformed refs. For this check,
we specifically do not want to say having less than two levels
in the name is illegal to allow HEAD (and perhaps other special
refs in the future).
I still really want a section on interoperability with CVS, subversion,
etc., but I'm not getting around to it very fast, so just add this to
the TODO section for now. And a few other minor todo updates.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Add a brief discussion of reflogs. Also recovery of dangling commits
seems to fit in here, so move some of the discussion out of Linus's
email to here.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
user-manual: add references to git-config man page
Direct editing of config files may be more natural for users than using
the git-config commandline; but we should still reference the
git-config man page when we describe such editing, so people know where
to go for details on the config file syntax and meanings of the
variables.
Signed-off-by: "J. Bruce Fields" <bfields@citi.umich.edu>
Heavily expanded update hook to send more useful emails than the old hook
I know it's only an example, but having this might save someone else the
trouble of writing an enhanced version for themselves.
It basically does the same job as the old update hook, but with these
differences:
* The recipients list is read from the repository config file from
hooks.mailinglist
* Updating unannotated tags can be allowed by setting
hooks.allowunannotated
* Announcement emails (via annotated tag creation) can be sent to a
different mailing list by setting hooks.announcelist
* Output email is more verbose and generates specific content depending
on whether the ref is a tag, an annotated tag, a branch, or a
tracking branch
* The email is easier to filter; the subject line is prefixed with
[SCM] and a project description pulled from the "description" file
* It catches (and displays differently) branch updates that are
performed with a --force
Obviously, it's nothing that clever - it's the update hook I use on my
repositories but I've tried to keep it general, and tried to make the
output always relevant to the type of update.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
UNIX reference time of 1970-01-01 00:00 is UTC timezone, not local time zone
I got bitten because in the UK (where one would expect 1970-01-01 00:00
to be UTC 0) some politicians decided to mess around with daylight
savings time from 1968 to 1971; it was permanently BST (+0100). That
means that on my computer the following is true:
It's actually saying (in my timezone) "1970-01-01 01:00:00 UTC" + $ts.
Clearly this is wrong. The UNIX epoch started at midnight UTC not 1am
UTC.
This leads to the tagged time in hooks--update being shown as one hour
earlier than the true tagged time (in my timezone). The problem would
be worse for other timezones. For a +1300 timezone on 1970-01-01, the
tagged time would be 13 hours earlier. Oops.
The solution is to force the reference time to UTC, which is what this
patch does. In my timezone:
Teach for-each-ref about a little language called Tcl.
Love it or hate it, some people actually still program in Tcl. Some
of those programs are meant for interfacing with Git. Programs such as
gitk and git-gui. It may be useful to have Tcl-safe output available
from for-each-ref, just like shell, Perl and Python already enjoy.
Thanks to Sergey Vlasov for pointing out the horrible flaws in the
first and second version of this patch, and steering me in the right
direction for Tcl value quoting.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
This allows pushing over the git:// protocol, and while it's not
authenticated, it could make sense from within a firewalled
setup where nobody but trusted internal people can reach the git
port. git-daemon is possibly easier and faster to set up in the
kind of situation where you set up git instead of CVS inside a
company.
"git-receive-pack" is disabled by default, so you need to enable it
explicitly by starting git-daemon with the "--enable=receive-pack"
command line argument, or by having your config enable it automatically.
Signed-off-by: Linus Torvalds <torvalds@osdl.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Documentation/config.txt: Fix documentation of colour config tweaks.
* The description of valid colour specifications was rather
incomplete, so fix it so that it actually describes colour specs as
accepted by color_parse().
* The list of colour items allowed in color.diff.BLAH was missing the
`commit' and `whitespace' entries.
Signed-off-by: Mark Wooding <mdw@distorted.org.uk> Signed-off-by: Junio C Hamano <junkio@cox.net>
Compute accurate distances in git-describe before output.
My prior change to git-describe attempts to print the distance
between the input commit and the best matching tag, but this distance
was usually only an estimate as we always aborted revision walking
as soon as we overflowed the configured limit on the number of
possible tags (as set by --candidates).
Displaying an estimated distance is not very useful and can just be
downright confusing. Most users (heck, most Git developers) don't
immediately understand why this distance differs from the output
of common tools such as `git rev-list | wc -l`. Even worse, the
estimated distance could change in the future (including decreasing
despite no rebase occuring) if we find more possible tags earlier
on during traversal. (This could happen if more tags are merged
into the branch between queries.) These factors basically make an
estimated distance useless.
Fortunately we are usually most of the way through an accurate
distance computation by the time we abort (due to reaching the
current --candidates limit). This means we can simply finish
counting out the revisions still in our commit queue to present
the accurate distance at the end. The number of commits remaining
in the commit queue is probably less than the number of commits
already traversed, so finishing out the count is not likely to take
very long. This final distance will then always match the output of
`git rev-list | wc -l`.
We can easily reduce the total number of commits that need to be
walked at the end by stopping as soon as all of the commits in the
commit queue are flagged as being merged into the already selected
best possible tag. If that's true then there are no remaining
unseen commits which can contribute to our best possible tag's
depth counter, so further traversal is useless.
Basic testing on my Mac OS X system shows there is no noticable
performance difference between this accurate distance counting
version of git-describe and the prior version of git-describe,
at least when run on git.git.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Teach git-describe to display distances from tags.
If you get two different describes at different
times from a non-rewinding branch and they both come up with the same
tag name, you can tell which is the 'newer' one by distance. This is
rather common in practice, so its incredibly useful.
[jc: still needs documentation and fixups when traversal gives up
early.]
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
git-blame --porcelain: quote filename in c-style when needed.
Otherwise a pathname that has funny characters such as LF would
screw up the parsing programs of the output.
Strictly speaking, this is not backward compatible, but the
current output for pathnames that have embedded LF and such
cannot be sanely parsed anyway, and pathnames that only use
characters from the portable pathname character set won't be
affected.
This adds --incremental option to help GUI porcelains to show
the result from git-blame incrementally. The output gives the
origin information in the same format as the porcelain format.
The first line has commit object name, the line number of the
first line in the group in the original file, the line number of
that file in the final image, and number of lines in the group.
Then subsequent lines show the metainformation for the commit
when the commit is shown for the first time, except the filename
information is always shown (we cannot even make it conditional
to -C option as blame always follows the renaming of the file
wholesale).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Back when only handful commands that created commit and tag were
the only users of committer identity information, it made sense
to explicitly call setup_ident() to pre-fill the default value
from the gecos information. But it is much simpler for programs
to make the call automatic when get_ident() is called these days,
since many more programs want to use the information when updating
the reflog.
git-log -g --pretty=oneline should display the reflog message
In the context of reflog output the reflog message is more useful than
the commit message's first line. When relevant the reflog message
will contain that line anyway.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Allow the tag signing key to be specified in the config file
I did this:
$ git tag -s test-sign
gpg: skipped "Andy Parkins <andyparkins@gmail.com>": secret key not available
gpg: signing failed: secret key not available
failed to sign the tag with GPG.
The problem is that I have used the comment field in my key's UID
definition.
$ gpg --list-keys andy
pub 1024D/4F712F6D 2003-08-14
uid Andy Parkins (Google) <andyparkins@gmail.com>
So when git-tag looks for "Andy Parkins <andyparkins@gmail.com>";
obviously it's not going to be found.
There shouldn't be a requirement that I use the same form of my name in
my git repository and my gpg key - I might want to be formal (Andrew) in
my gpg key and informal (Andy) in the repository. Further I might have
multiple keys in my keyring, and might want to use one that doesn't
match up with the address I use in commit messages.
This patch adds a configuration entry "user.signingkey" which, if
present, will be passed to the "-u" switch for gpg, allowing the tag
signing key to be overridden. If the entry is not present, the fallback
is the original method, which means existing behaviour will continue
untouched.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
If abbrev is set to zero in git-describe, don't add the unique suffix
When on a non-tag commit, git-describe normally outputs descriptions of
the form v1.0.0-g1234567890
Some scripts (for example the update hook script) might just want to
know the name of the nearest tag, so they then have to do
x=$(git-describe HEAD | sed 's/-g*//')
This is costly, but more importantly is fragile as it is relying on the
output format of git-describe, which we would then have to maintain
forever.
This patch adds support for setting the --abbrev option to zero. In
that case git-describe does as it always has, but outputs only the
nearest found tag instead of a completely unique name. This means that
scripts would not have to parse the output format and won't need
changing if the git-describe suffix is ever changed.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>
fix suggested branch creation command when detaching head
Doing:
$ git checkout HEAD^
Generates the following message:
|warning: you are not on ANY branch anymore.
|If you meant to create a new branch from the commit, you need -b to
|associate a new branch with the wanted checkout. Example:
| git checkout -b <new_branch_name> HEAD^
Of course if the user does as told at this point the created branch
won't be located at the expected commit. Reword this message a bit to
avoid such confusion.
Signed-off-by: Nicolas Pitre <nico@cam.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
vc-git.el: Take into account the destination name in vc-checkout.
This is necessary for vc-version-other-window. Based on a patch by Sam
Vilain <sam.vilain@catalyst.net.nz>.
Currently, the vc-git-checkout function uses `git checkout' to fetch a
file from the git repository to the working copy. However, it is
completely ignoring the input argument that specifies the destination
file. `git-checkout' does not support specifying this, so we have to
use `git-cat-file', capture the output in a buffer and then save it.
Signed-off-by: Alexandre Julliard <julliard@winehq.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
The file format dictates that entries are LF terminated so
the message cannot have one in it. Chomp the message to make
sure it only has a single line if necessary, while removing the
leading whitespace.
Junio added the found variable to enforce commit date order when two
tags have the same distance from the requested commit. Except it is
unnecessary as match_cnt is already used to record how many possible
tags have been identified thus far.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org> Signed-off-by: Junio C Hamano <junkio@cox.net>
Older Solaris machines lack stdint.h but have inttypes.h.
The standard has inttypes.h including stdint.h, so at worst
this pollutes the namespace a bit.
Signed-off-by: Jason Riedy <ejr@cs.berkeley.edu> Signed-off-by: Junio C Hamano <junkio@cox.net>
diffcore-pickaxe: fix infinite loop on zero-length needle
The "contains" algorithm runs into an infinite loop if the needle string
has zero length. The loop could be modified to handle this, but it makes
more sense to simply have an empty needle return no matches. Thus, a
command like
git log -S
produces no output.
We place the check at the top of the function so that we get the same
results with or without --pickaxe-regex. Note that until now,
git log -S --pickaxe-regex
would match everything, not nothing.
Arguably, an empty pickaxe string should simply produce an error
message; however, this is still a useful assertion to add to the
algorithm at this layer of the code.
Noticed by Bill Lear.
Signed-off-by: Jeff King <peff@peff.net> Signed-off-by: Junio C Hamano <junkio@cox.net>
Allow non-developer to clone, checkout and fetch more easily.
The code that uses committer_info() in reflog can barf and die
whenever it is asked to update a ref. And I do not think
calling ignore_missing_committer_name() upfront like recent
receive-pack did in the aplication is a reasonable workaround.
What the patch does.
- git_committer_info() takes one parameter. It used to be "if
this is true, then die() if the name is not available due to
bad GECOS, otherwise issue a warning once but leave the name
empty". The reason was because we wanted to prevent bad
commits from being made by git-commit-tree (and its
callers). The value 0 is only used by "git var -l".
Now it takes -1, 0 or 1. When set to -1, it does not
complain but uses the pw->pw_name when name is not
available. Existing 0 and 1 values mean the same thing as
they used to mean before. 0 means issue warnings and leave
it empty, 1 means barf and die.
- ignore_missing_committer_name() and its existing caller
(receive-pack, to set the reflog) have been removed.
- git-format-patch, to come up with the phoney message ID when
asked to thread, now passes -1 to git_committer_info(). This
codepath uses only the e-mail part, ignoring the name. It
used to barf and die. The other call in the same program
when asked to add signed-off-by line based on committer
identity still passes 1 to make sure it barfs instead of
adding a bogus s-o-b line.
- log_ref_write in refs.c, to come up with the name to record
who initiated the ref update in the reflog, passes -1. It
used to barf and die.
The last change means that git-update-ref, git-branch, and
commit walker backends can now be used in a repository with
reflog by somebody who does not have the user identity required
to make a commit. They all used to barf and die.
I've run tests and all of them seem to pass, and also tried "git
clone" as a user whose GECOS is empty -- git clone works again
now (it was broken when reflog was enabled by default).
contrib/emacs/vc-git.el: support vc-version-other-window
Currently, the vc-git-checkout function uses `git checkout' to fetch a
file from the git repository to the working copy. However, it is
completely ignoring the input argument that specifies the destination
file. `git-checkout' does not support specifying this, so we have to
use `git-cat-file', capture the output in a buffer and then save it.
New files in git weren't being downloaded during CVS update
If a repository was checked out via git-cvsserver and then later a new
file is added to the git repository via some other method; a CVS update
wasn't fetching the new file.
It would be reported as a new file as
A some/dir/newfile.c
but would never appear in the directory.
The problem seems to be that git-cvsserver was treating these two cases
identically, as "A" type results.
1. New file in repository
2. New file locally
In fact, traditionally, case 1 is treated as a "U" result, and case 2
only is treated as an "A" result. "A", should just report that the file
is added locally and then skip that file during an update as there is
(of course) nothing to send.
In both these cases there is no working revision, so the checking for
"is there no working revision" will return true. The test for case 2
needs refining to say "if there is no working revision and no upstream
revision". This patch does just that, leaving case 1 to be handled by
the normal "U" handler.
I've also updated the log message to more accurately describe the
operation. i.e. that "A" means that content is scheduled for addition;
not that it actually has been added.
Signed-off-by: Andy Parkins <andyparkins@gmail.com> Signed-off-by: Junio C Hamano <junkio@cox.net>