Neal H. Walfield, le Sat 09 Jun 2007 00:29:38 +0200, a écrit : > The theory is that we don't trust the server to honor the timeout: it > may be malicious and trick the client into waiting forever.
Or it may be buggued and hung. > However, there are enough ways in which we rely on the server for > correct operation that using the Mach timeout mechanism to preempt > the server doesn't bring any additional safety. Mmm, maybe, but is it really the way we want to go? (i.e. trusting the server for such operations). Server hangs do happen, and having applications hung just because of this is not pleasant. > Instead, I think, we should pass the timeout directly to the server > and rely on it to honor the timeout according to the specification. Well, that's a big change, since it requires all servers to be modified. Maybe in the meanwhile we could use my patch? Samuel