Re: new release candidate (gsl-1.5.90)

[Giulio Bottazzi <giulio dot bottazzi at libero dot it>]

On Fri, 31 Dec 2004 13:56:58 +0000
Brian Gough <bjg@network-theory.co.uk> wrote:

> You need e.g. GSL_IEEE_MODE="double-precision,mask-all" ...  to avoid
> floating-point exceptions raising an unhandled SIGFPE signal.

Indeed I noticed a behavior that appeared strange to me: on my Pentium
Linux machines the default floating point environment is

extended precision, rounding to the nearest and all exceptions masked

if I define the environment variable with

GSL_IEEE_MODE="double-precision"

the function gsl_ieee_env_setup() set the environment to

double precision, rounding to the nearest and ALL EXCEPTIONS rise an
interrupt (apart loss of precision)

So asking for one thing, a reduction of precision, actually generates a
second thing, a de-masking of almost all exceptions.

I would had expected the sole variation of the precision, without
modification of the other parameters.

Is this a design choice? If it is so, probably it would be useful
to add some warning about it in the documentation.

	G.


-- 
Giulio Bottazzi                 PGP Key ID:BAB0A33F

Attachment: pgp00000.pgp
Description: PGP signature