Actually decouple special function (openspecfun) from base

[musm]

SpecialFunctions relies on openspecfun from Base. Like FFTW would the next logical step be to have SpecialFunctions actually carry out the build process instead of in base

[ararslan]

This is on my to-do list, we just need a pure Julia implementation of rem2pi. That's the only thing in Base that still uses openspecfun.

[ararslan]

Regarding a Julia implementation, ref discussion in #22004

[musm]

BTW there is https://github.com/JuliaLang/openlibm/blob/1581174c85f7b645b15ba1ac1c3a98fb601f0fe7/src/e_rem_pio2.c

in openlibm which exactly the same as the openspecfun code .... but perhaps to call it some minor modification would need to be made to the C lib ? I'm not sure why we are calling the openspecfun version anyways

[ararslan]

If I'm not mistaken, it's not standard to have that in system libms. Since we allow building with a libm other than openlibm but we don't allow any special function library other than openspecfun, we're likely calling it from openspecfun instead to ensure the function is always available.

[StefanKarpinski]

This doesn't actually affect any Base APIs, does it? If so, it's a non-breaking change and can be removed from the 1.0 milestone.

[tkelman]

Removing a library that people have been calling into and assuming gets shipped with julia binaries sounds breaking to me.

[ararslan]

Yeah, it's definitely breaking for anyone who could be ccalling. This is a pretty easy removal anyway so I should have no trouble getting this out in the 1.0 timeframe.