tommath.h: do not expose limits.h

[minad]

If we make MP_WARRAY private we do not have to expose limits.h in the tommath.h header. I believe MP_WARRAY is never used by library consumers. We have deprecated it for now, after this change it would be only available in tommath_private.h.

[minad]

Grepping through the debian code base shows that MP_WARRAY appears only in vendored tommath code, but never in library user code.

https://codesearch.debian.net/search?q=MP_WARRAY&perpkg=1

[nijtmans]

I fully agree! However, this can be improved a little bit:

See:
#define MP_WARRAY (int)(1uLL << (((CHAR_BIT * sizeof(private_mp_word)) - (2 * MP_DIGIT_BIT)) + 1))
It's not really usefull to shift a long long, and finally cast it to an (int).
Better:
#define MP_WARRAY (1 << (((CHAR_BIT * (int)sizeof(private_mp_word)) - (2 * MP_DIGIT_BIT)) + 1))

So: cast the result of the sizeof() operator to (int).

(I know, it already was like that, but if I propose to fix this in another PR then this one has to be rebased anyway)

[minad]

I think we also have mp_sizeof_bits. Unfortunately I am on mobile right now. If you have time, @nijtmans feel free to improve and push to this PR. I think you can push directly to this branch instead of opening another PR.

[minad]

@nijtmans I guess it is better now?

[nijtmans]

Yes, that's exactly what I mean!

[sjaeckel]

I'm not against this change itself, but as it would break our contracts with the user we should push it to a later stage and leave it as deprecated in tommath.h

[minad]

@sjaeckel Ok! Any plans on when you want to do the new release (1.2)? After that we could remove the deprecations and continue with 2.0? Would that be ok?

[minad]

close in favor of #378