[libc][math] should `fmod(-NAN,-NAN) = +NAN`?

[Flandini]

Following up on #61092, the report seems to indicate that fmod(-NAN,-NAN) = +NAN is a problem.

Is this wrong? I can't tell from 7.12.10.1.3 or F.10.7.1 what the behavior for two NaN arguments should be.

We seem to match glibc on this.

This behavior still happens and is already tested for, just a question of if we have the semantics correct?

Tagging @nickdesaulniers and @michaelrj-google.

[nickdesaulniers]

cc @lntue

[lntue]

so I don't think the standard requires any explicit nan, so any nan output would be sufficient. On the other hand, probably we shouldn't test for explicit nan value here, maybe testing whether the output is nan or not is more appropriate.