[rfc] Improve infinity and NaN support

[Daniel Jacobowitz]

On Mon, Mar 19, 2007 at 11:57:41AM -0400, Daniel Jacobowitz wrote:
> This patch is along the same lines as a patch posted by Dwayne Grant
> McConnell <dgm69@us.ibm.com> in August 2005.  It improves the handling
> of floating point numbers in two ways:
> 
>   - Infinity is more reliably printed as inf.
>   - Infinity and NaN are converted to host DOUBLEST infinity and NaN.
>     While we're still being rather sloppy in implementing target
>     floating point arithmetic, this fixes the most obvious problem.
> 
> Any comments?  I'm sure my new test will fail on some platforms,
> e.g. vax; I do not see a better way than listing insufficiently IEEE
> targets in the .exp file.

I checked this in.

-- 
Daniel Jacobowitz
CodeSourcery