This is the mail archive of the
java@gcc.gnu.org
mailing list for the Java project.
Re: fix for socket timeouts (and a question about _Jv_select)
- To: tromey at redhat dot com
- Subject: Re: fix for socket timeouts (and a question about _Jv_select)
- From: "Nic Ferrier" <nferrier at tapsellferrier dot co dot uk>
- Date: Sat, 30 Jun 2001 15:36:44 +0100
- Cc: java at gcc dot gnu dot org, cedric at wireless-networks dot com
>>> Tom Tromey <tromey@redhat.com> 30-Jun-01 2:53:50 PM >>>
Yeah. What this means is that select() examines
the slots in the vector numbered 0, 1, ..., nfds-1.
If you pass `1' as nfds, then select() will only look
at the settings for fd 0. This is a goofy interface.
Maybe we ought to use poll() instead. I don't know. I
use select() since I have a bsd-ish background.
Ah! I see. Sorry, I think somewhere in the mists of time I did know
that...
I've been dealing with java non-blocking in a variety of different
ways this week (nbio, jdk1.4 non-blocking io and this GCJ change) and
I got all confused.
poll seems to be accepted as being quicker... /dev/poll is used by
the NBIO project to make things really quick... I'm not sure to what
extent it matters doing things the way we are though.
It's probably more important to get the threading layer better.
I will probably investigate the use of poll by hammering
GNU-Paperclips running under GIJ.
Nic