We can't unconditionally assign revision 1.1 to
newly added files. In case the file did exist in the
past and was deleted we need to honor the old
revision number.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* fl/cvsserver:
config.txt: Add gitcvs.db* variables
cvsserver: Document the GIT branches -> CVS modules mapping more prominently
cvsserver: Reword documentation on necessity of write access
cvsserver: Allow to "add" a removed file
cvsserver: Add asciidoc documentation for new database backend configuration
cvsserver: Corrections to the database backend configuration
cvsserver: Use DBI->table_info instead of DBI->tables
cvsserver: Abort if connect to database fails
cvsserver: Make the database backend configurable
cvsserver: Allow to override the configuration per access method
cvsserver: Handle three part keys in git config correctly
cvsserver: Introduce new state variable 'method'
Conflicts:
Documentation/config.txt
CVS allows you to add a removed file (where the
removal is not yet committed) which will
cause the server to send the latest revision of the
file and to delete the "removed" status.
Copy this behaviour.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Only send a modified response if the client sent a
"Modified" entry. This fixes the case where the
file was locally deleted on the client without
being removed from CVS. In this case the client
will only have sent the Entry for the file but nothing
else.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Acked-by: Martin Langhoff <martin@catalyst.net.nz>
Acked-by: Daniel Barkalow <barkalow@iabervon.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Don't include the scheme name in gitcvs.dbdriver, it is
always 'dbi' anyway.
Don't allow ':' in driver names nor ';' in database names for
sanity reasons.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
DBI->table_info is portable across different DBD backends,
DBI->tables is not.
Limit the output to objects of type TABLE.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Currently all calls to the database backend make no
error checking or handling at all. At least abort
if the connection to the database failed since
there is really no way we could do anything useful
after that.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Make all the different parts of the database backend connection
configurable. This adds the following string configuration variables:
- gitcvs.dbdriver
- gitcvs.dbname
- gitcvs.dbuser
- gitcvs.dbpass
The default values emulate the current behavior exactly for
backwards compatibility.
All configuration variables can also be specified for a specific
access method (i.e. in the form gitcvs.<method>.<var>)
The dbdriver/dbuser/dbpass variables are added for completness.
No other backend than SQLite is tested yet.
The dbname variable on the other hand is useful with this backend
already (to not discriminate against other possible backends
it was not splitted in dbdir and dbfile).
Both dbname and dbuser support dynamic variable substitution where
the available variables are:
%m -- the CVS 'module' (i.e. GIT 'head') worked on
%a -- CVS access method used (i.e. 'ext' or 'pserver')
%u -- User name of the user invoking git-cvsserver
%G -- .git directory name
%g -- .git directory name, mangled to be used in a filename,
currently this substitutes all chars except for [\w.-]
with '_'
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Allow to override the gitcvs.enabled and gitcvs.logfile configuration
variables for each access method (i.e. "ext" or "pserver") in the
form gitcvs.<method>.<var>
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This is intended to be used in the form gitcvs.<method>.<var>
but this patch doesn't introduce any users yet.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
$state->{method} contains the CVS access method used,
either 'ext' or 'pserver'
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
commit: Also print the old revision similar to how cvs does it and
prepend a line stating the filename so that one can actually
understand what happened when commiting more than one file.
status: Fix the RCS filename displayed. The directory was
printed twice.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Submit some additional messages to the client on commit and update.
Inspired by the standard CVS server though a little more terse.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Using Update-existing leads to the client forgetting about the "locally
modified" status of the file which can lead to loss of local changes on
later updates.
Signed-off-by: Frank Lichtenheld <frank@lichtenheld.de>
Acked-by: Martin Langhoff <martin.langhoff@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The config option gitcvs.allbinary may be set to force all entries to
get the -kb flag.
In the future the gitattributes system will probably be a more
appropriate way of doing this, but that will easily slot in as the
entries lines sent to the CVS client now have their kopts set via the
function kopts_from_path().
In the interim it might be better to not just have a all-or-nothing
approach, but rather detect based on file extension (or file contents?).
That would slot in easily here as well. However, I personally prefer
everything to be binary-safe, so I just switch the switch.
Signed-off-by: Andy Parkins <andyparkins@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
The commithash for updating the ref is obtained from a call to
git-commit-tree. However, it was returned (and stored) with the
trailing newline. This meant that the later call to git-update-ref that
was trying to update to $commithash was including the newline in the
parameter - obviously that hash would never exist, and so git-update-ref
would always fail.
The solution is to chomp() the commithash as soon as it is returned by
git-commit-tree.
Signed-off-by: Andy Parkins <andyparkins@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This makes "ci" codepath lockless by following the usual
"remember the tip, do your thing, then compare and swap at the
end" update pattern using update-ref. Incidentally, by updating
the code that reads where the tip of the head is to use
show-ref, it makes it safe to use in a repository whose refs are
pack-pruned.
I noticed that other parts of the program are not yet pack-refs
safe, but tried to keep the changes to the minimum.
Signed-off-by: Junio C Hamano <junkio@cox.net>
git-cvsserver is analogous to git-receive-pack; a checking from a cvs
client to a central server is like a git-push from a working repository.
Therefore it's nice to use the same access control (and email sending)
that a receive-pack would perform.
This patch tests for an executable update hook; if it is it is run with
the ref being updated and the old and new hashes as normal. If the
update hook returns an error code the update is aborted and the ref is
never updated. The cvsserver returns "error 1" to the client to signal
there was an EPERM error.
Signed-off-by: Andy Parkins <andyparkins@gmail.com>
Signed-off-by: Junio C Hamano <junkio@cox.net>
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>
With this patch, cvs add / cvs commit echoes back to the client
the correct file version (1.1) so that the file in the checkout
is recognised as up-to-date.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
if the SHA1 of our head matches the last SHA1 seen in the DB, avoid further
processing.
[jc: an "Oops, please amend" patch rolled in]
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
In the same vein as 8336afa563,
this fixes the the RCS merge to git-merge-file conversion in
commit e2b70087.
Signed-off-by: Eric Wong <normalperson@yhbt.net>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* master: (42 commits)
git-svn: correctly handle packed-refs in refs/remotes/
add test case for recursive merge
git-svn: correctly display fatal() error messages
git-svn: allow dcommit to take an alternate head
git-svn: enable logging of information not supported by git
Clarify fetch error for missing objects.
Move Fink and Ports check to after config file
shortlog: fix segfault on empty authorname
shortlog: remove "[PATCH]" prefix from shortlog output
Make sure the empty tree exists when needed in merge-recursive.
Don't use memcpy when source and dest. buffers may overlap
no need to install manpages as executable
Documentation: simpler shared repository creation
shortlog: fix segfault on empty authorname
Add branch.*.merge warning and documentation update
Fix perl/ build.
git-svn: use do_switch for --follow-parent if the SVN library supports it
Fix documentation copy&paste typo
git-svn: extra error check to ensure we open a file correctly
Documentation: update git-clone man page with new behavior
...
Now that we have git-merge-file, an RCS merge lookalike, we no longer
need it. So long, merge, and thanks for all the fish!
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
At some point between v5.6 and 5.8 Perl started to assume its input,
output and filehandles are UTF-8. This breaks the counting of bytes
for the CVS protocol, resulting in the client expecting less data
than we actually send, and storing truncated files.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Tonight I found a git-cvsserver instance spending a lot of time in
disk IO while trying to process operations against a Git repository
with >30,000 objects contained in it.
Blowing away my SQLLite database and rebuilding all tables with
indexes on the attributes that git-cvsserver frequently runs queries
against seems to have resolved the issue quite nicely.
Since the indexes shouldn't hurt performance on small repositories
and always helps on larger repositories we should just always create
them when creating the revision storage tables.
Signed-off-by: Shawn O. Pearce <spearce@spearce.org>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Earlier, cvsserver showed always an 'U', sometimes even without a space
between the 'U' and the name. Now, the correct letter is shown, with a
space.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
git-update-ref writes into the lockfile, and renames it afterwards. Like
commit v1.3.0-rc3~22, it is not only cleaner, but also helps with shared
setups: every developer can have a different primary group; what matters
is that $GIT_DIR/refs/heads has to be writable by a group you are in.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Earlier, cvsserver barfed when you tried to check in files with a
multiline commit message.
That is what Argumentx is for... Argument: lines can be followed by
several Argumentx: lines, which means they should be appended.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
Turns out that DBD::SQLite does not favour preparing statements which are
never executed. So, turn all 4 statements, which were prepared _always_,
into methods, like the other 12 prepared statements.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Signed-off-by: Junio C Hamano <junkio@cox.net>
This patch defines $state->{prependdir} as the empty string, so that
quite a few warnings are avoided.
Signed-off-by: Johannes Schindelin <Johannes.Schindelin@gmx.de>
Acked-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
* Implemented global -n option
* Implemented "Questionable"
* Fixed Directory method, I _believe_ it's now correct in both cmdline and Eclipse.
* Directory method Now looks for localdir of "." and compares the repo dir, uses THIS as a basis for all directory level calculations.
* Added extra parameter to filenamesplit() to force stripping of "prepended" directory name. This ensures commits/updates etc work from any directory in the source tree.
* Modified argsfromdir() so it is "always" called. This means that when the client specifies a directory, the method can detect this and behave accordingly (this is currently only implemented for the '.' directory)
* Fixed "commit" method to correctly work from in a subdir
On 5/4/06, Linus Torvalds <torvalds@osdl.org> wrote:
> No it wasn't. "git log --parents" was definitely supposed to still work.
>
> That said, I suspect a git-cvsserver kind of usage is better off using
> "git-rev-list --parents HEAD" instead, which didn't break in the first
> place.
This has been an unfortunate sideway in the git API evolution.
We use git-repo-config for all the other .git/config interaction
so let's also use git-repo-config -l for the variable listing.
Signed-off-by: Petr Baudis <pasky@suse.cz>
git-cvsserver now knows how to do the pserver auth chat when the user
is anonymous. To get it to work, add a line to your inetd.conf like
cvspserver stream tcp nowait nobody git-cvsserver pserver
(On some inetd implementations you may have to put the pserver parameter twice.)
Commits are blocked. Naively, git-cvsserver assumes non-malicious users. Please
review the code before setting this up on an internet-accessible server.
NOTE: the <nobody> user above will need write access to the .git directory
to maintain the sqlite database. Updating of the sqlite database should be
put in an update hook to avoid this problem, so that it is maintained by
users with write access.
We now have different error messages when the repo is not found vs repo is
not configured to allow gitcvs. Should help users during initial checkouts.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
To create nested directories without (or before) sending file entries
is rather tricky. Most clients just work. Eclipse, however, expects
a very specific sequence of events. With this patch, cvsserver meets
those expectations.
Note: we may want to reuse prepdir() in req_update -- should move it
outside of req_co. Right now prepdir() is tied to how req_co() works.
Signed-off-by: Junio C Hamano <junkio@cox.net>
A recent Eclipse compat fix broke checkouts with -d. Fix it so that the server
sends the correct module name instead of the destination directory name.
Just by sending the files in an ordered fashion, clients can process them
much faster. And we can optimize our check of whether we created this
directory already -- faster.
Timings for a checkout on a commandline cvs client for a project with
~13K files totalling ~100MB:
Unsorted:
603.12 real 16.89 user 42.88 sys
Sorted:
298.19 real 26.37 user 42.42 sys
The Eclipse client uses cvs update when that menu option is triggered.
And doesn't like the standard cvs update response. Give it *exactly* what
it wants.
And hope the other clients don't lose the plot too badly.
Signed-off-by: Junio C Hamano <junkio@cox.net>
Initial checkouts were failing to create Entries files under Eclipse.
Eclipse was waiting for two non-standard directory-resets to prepare for a new
directory from the server.
This patch is tricky, because the same directory resets tend to confuse other
clients. It's taken a bit of fiddling to get the commandline cvs client and
Eclipse to get a good, clean checkout.
Signed-off-by: Martin Langhoff <martin@catalyst.net.nz>
Signed-off-by: Junio C Hamano <junkio@cox.net>
