Re: MT floating point arithmetic problems on i386 linux

[Bruno H. <br...@cl...>]

Sam asked:
> BTW, Bruno, why are you saving mant on the STACK around L_to_I(-d) there?

Because L_to_I is marked as "can trigger GC" in lispbibl.d.

> d=(float-digits n) must be a fixnum, right?

Not necessarily. In the TYPECODES objects representation, if n is a long-float
that takes up - say - 1 GB of memory, (float-digits n) will be
intDsize * Lfloat_length(n) = 2^5 * 2^28 = 2^33, which is much larger than
the largest positive fixnum (around 2^24-1).

Bruno