Re: MATH_EMULATE

["Perry E. Metzger" <perry%piermont.com@localhost>]

Vit Herman <v.herman%sh.cvut.cz@localhost> writes:
> On Tue, 15 Jan 2008, Perry E. Metzger wrote:
>
>> It rotted once, didn't it?
>
> Yes, but it was my understanding that it was a bit different kind of
> rotting :) I mean, it has rotten because nobody realized that inlining
> of FPU ops could break something - but it did.

No, it rotted because no one even realized they were changing the
compile options in that manner.

Anyway, in general, things rot if they're not tested after changes. It
is the way of software. My worry is that, if we set up a special
FPUless x86 compile option set, no one will be testing it regularly
and it will break again soon.

> Other kinds of rotting (non-FPU related) should affect 486 DX as
> well, no?

In fact, Andrew Doran committed a change today to fix the 486DX -- it
had apparently been broken in -current for 9 months... (I don't know
if that breakage was in NetBSD 4 as well, but it may have been, in
which case you might want to test a -current from right now.)

>> -current would be the real target here since that's what would get
>> fixed for the SX. I presume it fails like 4.0 does -- we would want to
>> figure out why.
>
> Just tried, all the same, restart right after loading the kernel.

Did it panic? If so, can you force it into ddb?

> I'm not sure it's 486SX specific - the machine in question is IBM
> PS/ValuePoint which I suspect could be the root of evil here. I
> could try with some more "normal" 486SX tomorow.

Thanks...

-- 
Perry E. Metzger                perry%piermont.com@localhost