Re: Problems with awk/nbawk on m68k

["Michael L. Hitch" <mhitch%lightning.msu.montana.edu@localhost>]

On Sat, 8 Jan 2011, Frank Wille wrote:

> Few people will build a release on a 68k Amiga, so it's good that you tested
> that.

  I try to run a native build once in a great while just to make sure it 
still works.

> > It's not obvious that this was due to nbawk, but that was my suspicion
> > at the time, and I have verfied that it was nbawk.
>
> How did you do that? Did you replace it with a different awk?
> It could also be a bug in the kernel or some driver which shows up in nbawk
> due to heavy disk access?

  Perhaps I should have said that it was nbawk that created the bad file 
list, whether it was something in nbawk or something corrupting the input 
to nbawk.

> That didn't work for me on my build machine until I also defined MKDTRACE=no
> and MKZFS=no.

  I think those are not in 5.1, so I didn't need to use them.  When I ran 
under 4.0 or 3.1, I had to define a couple of other things (X11FLAVOUR-x11 
and MAKESYSPATH=/home/NetBSD-5/src/share/mk).

> Do you need a full release build to verify that? Or is it sufficient to
> check out src and then run the script? I will do a checkout on my Amiga
> now.

  I think just the source tree is sufficient, as long as enough 
appropriate variables get defined (that usually get defined in build.sh).

> So parallel disk activity makes it fail?

  Disk activity appears to be sufficient to cause failures more often.

> Did you also try to run something without disk access in parallel? For
> example a benchmark?

  I've run 'openssl speed' to generate other cpu activity, and it will 
occasionaly hiccup, but not as often.

> Or do you have the possibility to try a different disk driver (e.g. ahsc(4)
> compared to cbiisc(4) or wesc(4))? Or try it with tmpfs.

  I'm running entirely from the IDE drive right now, and it's had one awk 
abort, and one erroneous output file so far.

> Yes, that would be worth a test.

  It also fails on 4.0 and 3.0/3.1 [3.0_RC5 userland and 3.1_RC1 kernel].

> Are there other M68k ports where it can be reproduced?

  That's for someone else to determine.  I've only got amigas, and have 
only tested with 68060 (my only other amigas have 68040).

Mike
--
Michael L. Hitch                        mhitch%montana.edu@localhost
Computer Consultant
Information Technology Center
Montana State University        Bozeman, MT     USA