On Fri, 27 Jul 2007 15:02:30 +0200
Goran Mekic <[EMAIL PROTECTED]> wrote:
> > The program isn't how we'd do it for all sorts of reasons (it's not
> > correct, reliable or error-safe). If we were to introduce locking,
> > we'd make fetchers output to stdout instead of a file and move the
> > file handling into the C++ code.
> > 
>       Is there some paper about this? I mean, the reasons why
> programs like flock are not correct?

flock isn't NFS-safe. The lockf system call is NFS-safe, given
sufficiently modern NFS implementations, but only does exclusive locks,
which makes things slightly trickier.

> Second question is if file
> locking is an option, anyway. I mean, it maybe breaks some future
> plans or such?

Locking done on the C++ side, with fetchers changed to output to
stdout, is an option. Locking in individual fetchers isn't, since it
makes them much more complicated and unreliable and requires different
fetchers to share a common locking protocol.

-- 
Ciaran McCreesh

Attachment: signature.asc
Description: PGP signature

_______________________________________________
paludis-user mailing list
[email protected]
http://lists.pioto.org/mailman/listinfo/paludis-user

Reply via email to