On 8/25/08, Andrey G. Grozin <A.G.Grozin at inp.nsk.su> wrote:
> I've built 5.16.3 on Gentoo Linux with gcl-2.6.8, clisp-2.46, sbcl-1.0.19,
> cmucl-19d. With cmucl, I get 1 unexpected test failure:
> I've inserted print(maxerror) into the test and found that all of them
> except the last one are <= 2.664535259100376e-14; the very last maxerror
> is 469616.0 . This corresponds to
>
> float(cot(%pi) - cot(%pi*24/24.0))
>
> When this expression is evaluated at the maxima toplevel, it gives 0.0 -
> no error message. float(cot(%pi)) gives -8.165619676597685e+15. I don't
> know why inside the loop in rtest_trig.mac this difference gives 469616.0,
> and what should be the correct behaviour.
Andrey, thanks for the report. For the record, with cmucl 19c + Fedora,
I see no unexpected errors.
Maybe the problem originates from the difference between single and
double precision floating point values? Just a guess.
best
Robert Dodier