Dieter Kaiser wrote:
> Hello Ray,
>
> perhaps we have the same problem with complex log functions in the tests which
> still fail. I search the code which causes the failures.
I think this is the problem. With clisp, expintegral_ei(0.5) is
.4542199048631735, but with cmucl, we get .4542199048631735 -
6.283185307179586 %i. And this happens because
(expintegral-e 1 #c(-0.5 -0.0)) -> #c(-0.4542199 3.14159)
And this happens because we take (log z) -> #c(-0.693 -3.14159).
A quick fix would be to add #c(0 0) to z in expintegral-e (or higher?)
This will force any negative zeros to become positive zeros.
Ray