On Dec 8, 2005, at 8:05 AM, Paul Brook wrote:
1) Make long double == double.
Eventually the day will come when one want something bigger, then,
you have to break abi for this. We did this on darwin, and ick,
whatta pain. I think I prefer a hard error for even mentioning long
double, as the upgrade path is clear.
2) Choose a sensible format for long double. The obvious candidate
is a
128-bit PPC/MIPS stye almost-quad precision type implemented with a
pair of
64-bit doubles.
Yup, makes sense, though I do worry when CPU designers give up a real
128 long double in hardware that then doesn't match the software,
but, maybe they'll never do that...
