Re: [PATCH] "svn commit" performance

On Fri, Jul 11, 2003 at 12:46:06AM +0100, Philip Martin wrote:
> > So is it safe for an access baton just to ignore the already locked
> > descendant items for the current tree locking?
>
> Don't ignore them, but use them if they are already locked. In most
> places the wc code is written so that it "knows" whether a directory
> is locked or not, and errors occur if it attempts to lock something
> twice or if it attempts use a lock that does not exist. I think this
> is the correct behaviour in general, but for this particular function
> I propose to relax it.

in this case i think the do_open function needs to have another flag
for accumulation. also need to have new access type saying this
directory is tree-locked.

so if /foo/bar/baz is locked (access type write_lock), accumulative tree
locking of /foo/bar will go thru baz so it also locks /foo/bar/baz/deep/dir.

and if /foo/bar/baz is tree locked, accumulative tree locking of /foo/bar
will save the traversal time for locking /foo/bar/baz and locks other items.

If this is the desired behaviour, I will file an issue and start working
on this.

Cheers,
CLK

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Jul 11 16:03:02 2003