On Wed, Jul 8, 2009 at 2:19 PM, Christophe Rhodes<csr21@...> wrote:
> Nikodemus Siivola <nikodemus@...> writes:
>> 2009/7/8 Gabriel Dos Reis <gdr@...>:
>>>> I haven't been able to reproduce this on x86 or x86-64 yet, however,
>>>> as both return #.SB-EXT:SINGLE-FLOAT-POSITIVE-INFINITY for me.
>>> I cannot get that. Do you have some preliminary setting that default
>>> all computations to extended IEEE-756?
>> Nope, default setup.
> I would expect this to vary by operating system as well as
> architecture. I remember OS X being particularly difficult to set the
> floating point control word for, on all architectures. The following
> snippet from src/code/float-traps.lisp suggests that I was not the
> only one:
> ;; FIXME: This apparently doesn't work on Darwin
> #!-darwin (setf (floating-point-modes) modes))
The comments in code/float-trap.lisp seem to suggest that on my machine
(x86-64) the actual setter is implemented in C.
Anyway, SBCL repl reports:
I have no idea what the actual value should be.
Anyway, Nikodemus can't you reproduce the behaviour when
trapping is enabled?