On Fri, Jan 11, 2002 at 04:14:45PM -0400, David Maxwell wrote:
> Ahh. Well, when I originally wrote that PR, I was only trying to solve
> the issue of multiple builds on a single host.

Would be cool to be able to do bulk build on a pool of hosts though :)

> 
> One option would be to consider shlock to be overkill - it probably
> isn't neccessary to eliminate the race condition between simultaneous
> grabs of a pkgsrc dir. Eliminating the clobber might be enough - use a
> simple lockfile with a single name to indicate that the directory is in
> use, so a new process doesn't start using a dir that's being worked in.
> 
> Any NFS use would leave the problem of stale locks though - there's no
> simple cross-platform way I know of to test for another NFS client
> holding a filehandle.

If one would use is to parallelise bulk builds across machines, I would
use a cron job to check for stale locks. To ease this I would use a single
directory for all locks, instead of spreading them across pkgsrc tree.
The name of the lock file could be the name of the package (or
category_package).

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--