On Fri, Mar 04, 2005 at 06:23:06AM +1100, matthew green wrote:
> 
> i wonder what the normal restrictions for signal handlers says
> about using anything but sig_atomic_t arith?  doesn't that apply
> here or am i crazy? :-)

I don't know of any such restriction.  AFAIK, sig_atomic_t is required
when doing an access to a datum from a signal handler that is also
accessed from the main program without it first blocking signals, but
there's nothing that I know of that says that non sig_atomic_t types
can't be used at all in a signal handler.

There was some discussion of the floating point issue in the tech-kern
mailing list a few years back.  That was for NetBSD in general, but the
x86 was specifically mentioned as well.  I don't know what decision was
made at that time, but the alternatives were layed out here:

  http://mail-index.netbsd.org/tech-kern/1999/10/16/0004.html

and there was some followup to that article.

--Doug.