std.math: №2, fix abs, fabs

[9il]

intrinsics for fabs should be added to DMD.

[kuettler]

Sorry to be picky, but afaik there should be no FIXME in a pull request.

[9il]

No. I have already pushed scores of FIXMEs into phobos. This is normal practise if DMD should be changed first.

[WalterBright]

Fix what?

[9il]

Fix code like

double fabs(double x) @safe pure nothrow @nogc { return fabs(cast(real)x); }

to declaration

double fabs(double x) @safe pure nothrow @nogc;

when DMD implement this intrinsic.

[kuettler]

There is fabsl, fabs and fabsf in core.stdc.math whenever available. Maybe I just do not understand what you are trying to do.

[9il]

This PRs have few goals:

1. fix dmd tests where code like fabs(-2) is used
2. fix unittest in phobos (same code style)
3. allow fabs and other intrinsics to be replaced in other compilers (like LDC)
4. allow to write other functions based on correct API: float log(float), double exp(double), double gamma(double).

This isn't change for fabs itself, but for other functions and compilers.

[9il]

PS: core.stdc.math is not pure.

[burner]

Could you add a unittest that calls all three version, simular to what is done in std.string

foreach (S; TypeTuple!(float, double, real))

[9il]

Unittest was added.

[burner]

Thank, LGTM

[burner]

Auto-merge toggled on

[9rnsr]

This pull request introduced a regression:
https://issues.dlang.org/show_bug.cgi?id=14842

[9il]

... and this is good because implementation bugs like in approxEqual can be fixed faster both in Phobos and user code.