Refactor package, separating byte calculations out

[mkitti]

I have done a major refactoring of this package on the total_refactor branch: https://github.com/mkitti/ArrayAllocators.jl/tree/total_refactor

[mkitti]

@bjarthur and @carstenbauer , I would appreciate if you could review the refactored design.

[mkitti]

A documentation preview can be found here:
https://mkitti.github.io/ArrayAllocators.jl/previews/PR3/

[carstenbauer]

A few comments:

• On https://mkitti.github.io/ArrayAllocators.jl/previews/PR3/bytecalculators/, the index list mentions NUMA stuff but the actual functions aren't on that page (but in the NUMA section). Same issue here
• numa and NumaAllocator are not available. I get UndefVarError when I try to run the examples in the docs. And using NumaAllocators, as written in the README.md, also doesn't work. How does that work? (And it should be documented.)
• MemAlign isn't mentioned in the docs? BTW, for testing that the alignment is actually correct, you can get inspiration here.
• IMO, you should cleanly separate examples from API references / interfaces, i.e. avoid a "everything on one page" layout like here. See https://documentation.divio.com/

[mkitti]

Thanks for the review @carstenbauer. Yes, I am primarily trying to improve the documentation now. I think part of the issue is that I need to move some of the docstrings outside of the operating system specific if statements.

• numa and NumaAllocator are not available. I get UndefVarError when I try to run the examples in the docs. And using NumaAllocators, as written in the README.md, also doesn't work. How does that work? (And it should be documented.)

Eventually, using NumaAllocators should work when I register the packages. For now, the following should work:

] activate --temp
] add https://github.com/mkitti/ArrayAllocators.jl#total_refactor
] add https://github.com/mkitti/ArrayAllocators.jl#total_refactor:NumaAllocators
] add https://github.com/mkitti/ArrayAllocators.jl#total_refactor:SafeByteCalculators

[mkitti]

@carstenbauer, what happens when T, the element type, is not a bitstype?

[carstenbauer]

@carstenbauer, what happens when T, the element type, is not a bitstype?

For MemAlign? I would think that it just doesn't work / errors.

[mkitti]

We currently throw an error for PosixMemAlign. I'm wondering if we should throw an error more generally for other allocators other than PosixMemAlign.

[mkitti]

Thank you for the reviews, @carstenbauer and @bjarthur. I'm going to merge and release this now.