develooper Front page | perl.vmsperl | Postings from October 2002

Re: MMK + 7.3-1 = :-( [was Re: Problem building Perl]

Thread Previous | Thread Next
Craig A. Berry
October 4, 2002 11:59
Re: MMK + 7.3-1 = :-( [was Re: Problem building Perl]
Message ID:
At 1:12 PM -0400 10/4/02, wrote:

>I have just succeeded in building an unpatched perl 5.8.0
>using DEC C V6.0-001 on OpenVMS Alpha V7.3-1 with MMK V3.9-3
>on an ODS-2 volume. 

Interesting.  Is XFC enabled?  Have you installed the DEC AXPVMS
VMS731_RMS V1.0 ECO?

>I note that Steve reported a failure to build the File::Glob

I can reproduce it with any extension and it exhibits the problem
when run outside of  For example, I can just go to
[.ext.attrs] and do an MMK and it gets stuck doing lots of CPU and
I/O.  I can't tell what it's doing I/O *with* since none of the
channels it has open is busy.  Oh, and just to add to the surreal
aspects of this, MONITOR PROCESS/TOPDIO tells me that it thinks it's
doing over 50,000 direct I/Os per second!  I'll check for antimatter
emanating from the system disk but otherwise I think it's safe to say
this is not possible.  The system is spending 70-80% of its time in
kernel mode.

If I do SET WATCH FILE/CLASS=ALL and repeat, then MMK does not get
stuck, so there may be a timing issue of some sort involved and the
dump of all the file accesses slows things down sufficiently to make
a difference.

>My suspicions with the
>trouble that Steve is seeing is that perhaps this is a
>different signature to the UAF limit exceeeded problem(???).
>FWIW I had:
>$ search pgflq
>which is in excess of the UAF default value.

Excess is right.  15 million pagelets is roughly 7.5 GB of pagefile.
I increased mine up to 1.5 million pagelets from 200,000 and I still
see the problem.  We've always said 100,000 was enough to build Perl
so I doubt whether anything like what you have could be necessary
(or even possible without a much larger pagefile than I have).  But
of course there are other quotas than pagefile.

But poking around in the SDA while the stuck process is spinning
away I see no evidence of quota problems.  It's still possible that a
system service call hits a quota problem and MMK doesn't handle the
failure properly and then gets stuck later while trying to do something
that it can't.

I think the I/O wackiness is the real smoking gun here, I just don't
know exactly what it means yet.
Craig A. Berry

"... getting out of a sonnet is much more
 difficult than getting in."
                 Brad Leithauser

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