Tcl Source Code

View Ticket
Login
Ticket UUID: 2966493
Title: chan-io-53.9 test is timing sensitive
Type: Bug Created on: 2010-03-09 14:36:05
Submitter: dgp Assigned to: andreas_kupries
Subsystem: 25. Channel System Severity:
Priority: 5 Medium Last modified: 2010-03-11 00:08:03
Status: Open Closed by:
Resolution: None Closed on:
Version: obsolete: 8.6b1.1
Description:
Low prio, but just saw chan-io-53.9 test fail
when running `make test` on a busy machine.
Suspect this is because of a hard-coded timeout
value.  A test not sensitive to timing variations would
be better.
User Comments:
ferrieux added on 2010-03-10 05:30:31:
Unless I'm misreading, io-chan-53.9 arranges for some IO to happen within 2 secs of the beginning, and timeouts after 5 secs. This is everything but a sensitive timing setup. I suspect a more worrying race condition in the [fcopy] machinery, though I cannot repro on Linux (neither threaded nor unthreaded) on a dual core with 4 processes busylooping...

Any details concerning your setup to help reproducing here ?

Upping prio, race conditions are bad.

dkf added on 2010-03-10 06:08:32:
I sometimes get weird behaviour (overall; not just Tcl) when the system is doing a lot of IO (e.g., paging out my browser...) so maybe that sort of thing is needed as well?

dgp added on 2010-03-11 00:08:03:
Can't repro this reliably and will not block for it.

In fact, I can't even repeat it.

One time this failed.  At the time I
had several builds in progress, so
I blamed the busy machine for the
failure.

I just tried to repro, and there was no
failure of this test, though other timing
sensitive test did fail (clock-33.[58]*).