Re: [coLinux-users] random floating point errors

[miguel m. <jj...@gm...>]

thanks to those who have replied.

I've just upgraded to 0.6.1 from 0.6.0, and I think things became
worst. For proper context, I'd try to explain what I'm doing. I have a
lisp program (in cmucl) that generates tons of circuits (~100k). This
is piped to (a modified) gnucap, a circuit simulator and then my lisp
program gets the simulation result and do some stuff with it. I'm
doing this on my "job" pc, which is in windows and there is no windows
port of cmucl (gnucap too, but that is hack-able) that's why I have to
go through the colinux route. cygwin won't do.

Last time on 0.6.0, cmucl is spitting "Floating point
exception|error|something-to-that-effect" and then I apt-get upgrade-d
and then cmucl stopped doing that. Then in the middle of the program
it would suddenly halt because gnucap generated NaN, which is not
expected. I simulate the offending circuit and, if gnucap won't abort
and return 139, I don't see that NaN.

On Tue, 3 Aug 2004 21:17:17 +0300 (EEST), ti...@fu...
<ti...@fu...> wrote:
> 
> Wasn't there a bug in earlier colinux version that caused fpu state
> sometimes not completely managed between colinux and host os..?

I am not aware of this. So I guess it just can't be done (for now).

> On Tue, 3 Aug 2004, Henry Nestler wrote:
> > You say the errorcode is 136? So thats fine: Search this number in your
> > source or in source of the tools lib header files that you use!
I think this is in the kernel headers, and probably corresponds to
some floating point error =)