Re: more on proff crash building native

[Anders Magnusson <ragge%ludd.ltu.se@localhost>]

Michael L. Hitch wrote:
> On Sat, 14 Feb 2009, Greg Oster wrote:
>
> > >    Native builds (when continued several times after the above problem)
> > > will hang the machine when it tries compiling libm/catan.ln.  I'm 
> > > trying
> > > to collect some more information on that and will submit a PR.
> >
> > Hmm.. I'm not quite up to that point yet (but have restarted the
> > build often enough to be past the pic/grops/groff/whatever issue),
> > but I see I'm using some 69MB of swap right now.. (the machine has 
> > 32MB).
> > I've added an extra 500MB swap -- could your box be running out of
> > memory/swap when it gets to working on libm/catan.ln ?
>
>   Hmmm, I'll guess this has something to do with it:
>
> lib/libm/complex/catan.c:41:#define MAXNUM 1.0e308
> lib/libm/complex/catan.c:77:    w = MAXNUM + MAXNUM * I;
>
>   I got this by doing one single step in DDB after it hung:
>
> /home/mhitch/NetBSD-5/src/lib/libm/complex/catan.c(77): warning: 
> floating-point constant out of range [248]
> /home/mhitch/NetBSD-5/src/lib/libm/complex/catan.c(77): warning: 
> floating-point constant out of range [248]
> lint: /usr/libexec/lint1 got SIGTRAP
>
>   The complex routines were added 17 months ago, and probably without 
> knowing or realizing that vax floating point is different.
It should use DBL_MAX instead of hard-code some max. 

But I assume that there may be other places where the complex routines 
are wrong,
like infinity do not exist on vax so that must be taken into account.  
Should be checked.

>   Doing a show uvmexp prior to running the lint command and a couple of
> times during the run shows that the trap count is increasing rapidly:
>
>   faults=736035, traps=1495911, intrs=9165866, ctxswitch=1024010
>   faults=736760, traps=1653997, intrs=9169457, ctxswitch=1024510
>   faults=736760, traps=2074777, intrs=9172623, ctxswitch=1024530
Lint shouldn't get signals here.  Hm, maybe it is floating point stuff 
that is
causing it?  Since vax can give signals for floating point errors.

-- Ragge


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