develooper Front page | perl.perl5.porters | Postings from February 2003

Re: [perl #21380] 5.8.0 test failure on Cobalt RaQ2

Thread Previous | Thread Next
Nicholas Clark
February 27, 2003 13:17
Re: [perl #21380] 5.8.0 test failure on Cobalt RaQ2
Message ID:
On Wed, Feb 26, 2003 at 08:40:01PM -0000, J. Lasser wrote:

> Are these known bugs?  I'm happy to follow instructions to try and debug
> this problem further...

I wasn't aware of either. I'm glad that the pack test is sorted, as I've
been near that a while back, and feared that it would turn out to be a
bug that I'd get "assigned"

>     osname=linux, osvers=2.0.34c52_sk, archname=mips-linux

On Thu, Feb 27, 2003 at 02:00:34PM -0500, J. Lasser wrote:
> In the wise words of Andy Dougherty:

> > I'm hopeful that'll clear up the op/pack.t tests.  I don't know about the
> > socketpair tests -- that may well be a kernel/glibc issue of some sort.
> > You might be able to learn a bit more by running the test by hand:
> > 	cd t/harness
> > 	./perl -I../lib ../ext/Socket/socketpair.t
> The failed part of that output is below:
> ok 35 - shutdown left for writing
> # Approximate forever as 3 seconds. Wait 'forever'...
> not ok 36 - read on right should be interrupted
> #     Failed test (../ext/Socket/socketpair.t at line 221)
> #          got: '0'
> #     expected: undef
> not ok 37 - alarm should have fired
> #     Failed test (../ext/Socket/socketpair.t at line 223)
> #          got: '0'
> #     expected: '1'
> ok 38 - syswrite to right
> Any suggestions from this point?

Not really. I suspect it's because a Linux 2.0 kernel isn't doing what the
tests expect. I'm not sure how to make it skip based on linux version, but
that seems to be the best thing to do.

The socketpair seems to be non-portable at OS level, with just about every
OS (even FreeBSD) doing something that is inconsistent with the documented
behaviour. "Even FreeBSD" because as I understand it socketpair() was
created in the 4.something BSD code tree that FreeBSD directly evolved from.

loopback sockets (both IP and Unix domain) in general seem to be vile,
particularly in combination with shutdown() This makes it hard to test
whether perl's interface to them is working.

Nicholas Clark

Thread Previous | Thread Next Perl Programming lists via nntp and http.
Comments to Ask Bjørn Hansen at | Group listing | About