Fix Issue 18247 - core.stdc.math functions that never set errno should be pure

[n8sh]

These include:

• cos, sin, tan, atan;
• tanh, asinh;
• frexp, modf;
• cbrt;
• erf;
• ceil, floor, trunc, round, rint, nearbyint;
• copysign, nan;
• fdim, fmax, fmin;
• fabs, abs, fma;
• fpclassify, isfinite, isinf, isnan, isnormal, signbit;
• isgreater, isgreaterequal, isless, islessequal, islessgreater, isunordered.

[dlang-bot]

Thanks for your pull request, @n8sh! We are looking forward to reviewing it, and you should be hearing from a maintainer soon.

Some tips to help speed things up:

• smaller, focused PRs are easier to review than big ones

• try not to mix up refactoring or style changes with bug fixes or feature enhancements

• provide helpful commit messages explaining the rationale behind each change

Bear in mind that large or tricky changes may require multiple rounds of review and revision.

Please see CONTRIBUTING.md for more information.

Bugzilla references

Auto-close	Bugzilla	Description
✓	18247	core.stdc.math functions that never set errno should be pure

[ibuclaw]

Are you sure about that? The latter ones seem fine, as they are bit getters/setters.

I don't believe you when you say that cos, sin and tan are pure.

[n8sh]

You were right to disbelieve me. On macOS 10.12.6 man cos says:

cos(+-infinity) returns a NaN and raises the "invalid" floating-point exception

But on Linux:

Domain error: x is an infinity
errno is set to EDOM (but see BUGS). An invalid floating-point exception (FE_INVALID) is raised.
...
Before version 2.10, the glibc implementation did not set errno to EDOM when a domain error occurred.

Will fix. I'll also check the other functions.

[ibuclaw]

I still think there are a couple others (tanh, asinh). I'll check against both glibc and gcc how they attribute these functions.

[n8sh]

Looking at the glibc sources, sysdeps/ieee754/dbl-64/s_sin.c contains #include <errno.h> and __set_errno. In contrast neither of those appear in:

• sysdeps/ieee754/dbl-64/s_tanh.c
• sysdeps/ieee754/dbl-64/s_asinh.c
• sysdeps/ieee754/dbl-64/s_atan.c

Neither do tanh/asinh/atan have branches in sysdeps/ieee754/k_standard.c which implements errno setting for a variety of functions.

However if tanh, asinh, and/or atan are sticking points I can remove the pure annotation from them.

[ibuclaw]

@n8sh - I'll use gcc as a reference, math functions that set errno are marked as MATHFN_ERRNO

Now, having gone through the list, I can say that of all functions you originally gave, only one is listed as setting errno:

• fdim - ✖️

Looks like cos, sin, and tan are fine afterall.

[ibuclaw]

Looking at the glibc sources, sysdeps/ieee754/dbl-64/s_sin.c contains #include <errno.h> and __set_errno. In contrast neither of those appear in:

If I was to be pedantic, tanh() calls expm1(), which may set errno.

[n8sh]

OK, added pure back to sin/cos/tan and removed it from fdim.

[ibuclaw]

Also newlib documentation is a good point of reference too.

[n8sh]

Rebased for musl.