Move public headers into include/ and private headers into src/.

[EdSchouten]

While there, also modify the install target. We should make sure to
install all openlibm*.h headers. There is still some work to be done:
openlibm_fenv_*.h still depends on some additional bits. I'd propose
that we eventually create an include/openlibm_cdefs.h that contains all
of the macros we need.

[Keno]

Thanks! @ViralBShah, can we give commit access to @EdSchouten, seems more than deserved! This might require adding a new team, since it's really separate from the regular Julia team.

[ViralBShah]

I think I did that yesterday and also gave access to @talex5. Is that not the case?

[Keno]

Oh, didn't know, I checked at some point yesterday, but you may not have done it then.

[ViralBShah]

BTW, I notice that in the tests, we have to include openlibm.h, openlibm_fenv.h and openlibm_complex.h. In the corresponding C headers, one does have to include all these, but for openlibm, should openlibm.h automatically include all the other openlibm headers?

[tkelman]

we use these headers in julia on windows, these should be put back

[EdSchouten]

Viral,

Right now openlibm.h is close to identical to math.h, so it doesn't automatically pull in openlibm_fenv.h and openlibm_complex.h. Something that came to mind was that we could rename openlibm.h to openlibm_math.h for consistency and add a very simple openlibm.h that does nothing more than include openlibm_{math,complex,fenv}.h. How does that sound to you?

Tony,

That's interesting. These headers are used by the build phase, but do we consider them to be public API? Would you mind explaining in what way Julia on Windows uses these? Thanks!

[tkelman]

No, they're not exactly public API, so it is perhaps incorrect to install them. We have a handful of assembly source files in Julia that only get used by the Windows build, which use these headers for a few assembly-related macro definitions. We could make independent copies of them in the julia repository, or refer to these files via their uninstalled source locations since we have openlibm as a submodule (this is actually what we're doing now), but I don't really like either of these options and would rather be using installed copies of the headers than private source locations.

[EdSchouten]

I see. I would personally consider copying these files into the Julia repository if we eventually want OpenLibm to become an independent library.

Shall I add these install directives back to the Makefile for now, but add a TODO item there that we should reconsider this in the future?

[tkelman]

I think this can remain a Julia TODO for now. We may eventually want to try adapting OpenLibm so it is possible to compile with MSVC, which would require some tweaking of these assembly macros (and translating all of the assembly to Intel syntax), at which point we can revisit this problem. As long as these files don't change relative location I think it should still work, though whenever we next bump the submodule we'll definitely want to test it thoroughly in a PR first.

[ViralBShah]

I think so long as we clearly mark the things we do not want to support long term, @tkelman and I can work towards removing those dependencies from base Julia, and then from openlibm.