Date: Tue, 9 Oct 2001 16:41:50 -0700 (PDT)
From: ###@###.###
By the way, I noticed an annoying oddity about Selectors again
that has to do with cancelling and registering keys. Since
when you cancel a key it doesn't really happen until the
next select call is made, this can mess up a Selector:
If I modify this test to use one Selector for the whole 2000
tests instead of making a new one every time, then the Selector
fails to announce the fd being ready for connect unless I
put in a select call in between closing the old fd and registering
the new one. This is because the old one does not get cancelled
until a select call has been made. If the next call is registering
the same fd as is being cancelled then it gets cancelled and nothing
happens. (Note that you often get the same fd again if you open
one fd and then close it and then ask for another). Without that
extra silly select call, the Selector never announces the any connections
to be ready past the first.
From: ###@###.###
By the way, I noticed an annoying oddity about Selectors again
that has to do with cancelling and registering keys. Since
when you cancel a key it doesn't really happen until the
next select call is made, this can mess up a Selector:
If I modify this test to use one Selector for the whole 2000
tests instead of making a new one every time, then the Selector
fails to announce the fd being ready for connect unless I
put in a select call in between closing the old fd and registering
the new one. This is because the old one does not get cancelled
until a select call has been made. If the next call is registering
the same fd as is being cancelled then it gets cancelled and nothing
happens. (Note that you often get the same fd again if you open
one fd and then close it and then ask for another). Without that
extra silly select call, the Selector never announces the any connections
to be ready past the first.
- duplicates
-
JDK-4519107 RAS: reproducible sig 11 crash with SelectorStressTest -errorID 4F530E43505002BA
-
- Closed
-
-
JDK-4512632 (so) RAS: connect test hangs on win98; fails on Solaris with "too many open file
-
- Closed
-