We currently try to fix all failing phobos unittests for gdc and we have failures in the  std/internal/math/gammafunction.d(367) tests. It seems the reason is that gdc doesn't use the custom inline versions of exp. We use the C expl function instead and this function provides slightly different results:

Testcase: http://dpaste.dzfl.pl/bb319763

Can anyone tell if
* This difference is expected?
* Which version is 'correct', the inline asm implementation or the C implementation?
* What should we do about this in the tests? should we just relax the tests?

(We could of course adapt the inline asm version to gdc inline asm, but using the C function is better regarding portability)

-- 
Johannes Pfau

_______________________________________________
phobos mailing list
phobos@puremagic.com
http://lists.puremagic.com/mailman/listinfo/phobos