Barton Willis wrote:
> Thanks--using Clozure CL, I also see these failures. Either I'll mark
> these as known bugs, make bfloat_approx_equal less strict, or fix the
> hypergeometric code. For now, let's live with these bugs (once bugs
> are marked as known, it's easy to forget about them :)
>
This reminds me of a few failures in the share testsuite with ecl. I
think the problem comes down to (float <rational> 1d0). I expected this
to round the rational to the nearest float, and it seems most Lisps do
that. But ecl does not and truncates instead. I couldn't find anything
in the CL spec that says FLOAT should round, so I think ecl is allowed
to do that. But that certainly breaks my expectations of what FLOAT
should do.
The only way around this would be to do it ourselves, but I think we
shouldn't. I think it should be fixed in ecl, or we just live with the
issue.
Ray