Re: crash building in groff

[Matt Thomas <matt%3am-software.com@localhost>]

On Jan 9, 2009, at 2:05 PM, Michael L. Hitch wrote:

> On Fri, 9 Jan 2009, Brad Parker wrote:
>
> > Turns out the crash in groff is due to some problem with hypot()
> > in the arc object creation code.  (what a pain to find!)
> >
> > If I make the c++ hypot() return 1.0 all is good.
> >
> > I'll try and figure out why.
>
>  That sounds like it may be a problem in hypot() attempting to return
> a "quiet" NaN, but the vax doesn't have a real NaN of any type, and
> someone decided to use a "reserved operand" for that, which would  
> result in an illegal instruction trap.
>
>  One possible fix or work-around would be to have the trap handler
> recognize the reserved operand and do something else (other than
> signal SIGILL to the process).

I fixed my local gas to deal with QNaN/SNah/Inf properly for
vax to deal with gcc-trunk,  Maybe those need are needed gcc-4.1
as well.