On 5/26/10 2:11 PM, Barton Willis wrote:
> Tests rtest_simplify_sum 52 & 54 are marked as known failures, but they pass with my
> not quite CVS Maxima (a modified orthopoly for one). Is it just my Maxima or do these
> two tests need to be unmarked as known failures?
>
> Also, I've been seeing a failure (Clozure CL) with rtest16 number 383. I'd guess
> this is just a problem with tolerance that is just a bit too small:
>
> Running tests in rtest16:
> ********************** Problem 383 ***************
> Input:
> block([numer : true], is(abs(zeta(%i + 3)
> - (1.10721440843141 - 0.148290867178175 %i)) < 1.0E-15))
>
>
I'll have to check, but I think CCL (and ecl) fails this test because
the result only has single-float precision. The basic issue is that ccl
and ecl computes (expt 2 #c(2d0 0d0)) with only single-float accuracy,
despite returning a double-float result.
I've filed a bug for ccl and I understand they are going to fix this
issue. I haven't filed a bug for ecl. I think sbcl has this problem
too, but I think sbcl will fix this too.
Ray