Attn: IEEE FP lawyers

[Bob Alverson]

Reading the 754 standard, it appears that conversion from floating
point to integer format is not well-defined.  It says to raise
invalid operation whenever something precludes a faithful representation
*and* this cannot otherwise be signaled.  Gee, if we have overflow,
couldn't we signal it by raising overflow?  The text on overflow
could easily apply to converting to integer, but it is not explicit.

In any case, I note that Sun raises invalid on overflowed conversion
to integer.  I haven't been able to try it on other machines.

Bob