Relax compile-time assertion in gammafunction to runtime

[ibuclaw]

Should fix builds on non-x86 targets, where the pretty lame yl2x intrinsic is not available at compile-time.

[9il]

Please add this description to the code.

[ibuclaw]

Shouldn't be needed as I'm going to propose yanking out uses of yl2x.

[yebblies]

This isn't really a fix.

[ibuclaw]

This isn't really a fix.

And breaking phobos builds is not a solution.

[ibuclaw]

Ping, Phobos is still broken.

[9il]

Please, add //FIXME with description.

[ibuclaw]

Please give me a description to fill in, because I don't think I can hold myself back from writing something detrimental about dmd.

[9il]

=)

    // FIXME: should be returned back to enum.
    // Fix requires CTFEable `log` on non-x86 targets (check both LDC and GDC).
    immutable maxY = logmdigamma(real.min_normal);

[ibuclaw]

Though FYI, you should probably go the way of std.math and use pre-computed manifests. One for each kind of real float format. We should either find a way to expose floatTraits to this module, or declare it public (and document it).

[9il]

See #3011.

[ibuclaw]

@9il ping
@MartinNowak - this one is for 2.068 beta2

[9il]

LGTM

[ibuclaw]

I'll merge this in 24 hours unless someone responds in the meantime (as @9il has given his approving 👍 and it's his code I'm changing)

[ibuclaw]

Auto-merge toggled on