Refactor FLINT randstate

[albinahlback]

• Rename flint_rand_s to flint_rand_struct,

• Replace fields gmp_state and gmp_init in flint_rand_s by void * __gmp_state,

• Rename the following functions:

  flint_randinit     -> flint_rand_init      (different content)
  flint_randclear    -> flint_rand_clear     (different content)
  flint_randseed     -> flint_rand_set_seed
  flint_get_randseed -> flint_rand_get_seed
  

• Remove _flint_rand_init_gmp and replace with the two new functions _flint_rand_init_gmp_state and _flint_rand_clear_gmp_state.

  NOTE: The new init-function for the GMP-state assumes that __gmp_state is not initialized, and the new clear-function assumes that __gmp_state is initialized. To help with this, a new macro FLINT_RAND_GMP_STATE_IS_INITIALISED(state) can be utilized to check whether the GMP-state is initialized in state, assuming that state has been initialized via flint_rand_init.

• Typedef ulong and slong to proper primitive C type instead of relying on GMP's mp_limb_t and mp_limb_signed_t.

[fredrik-johansson]

There should be macros for flint_randinit and flint_randclear for backward compatibility.

[albinahlback]

There should be macros for flint_randinit and flint_randclear for backward compatibility.

Yes, agreed!

[albinahlback]

I should also mention that I added clears of the random state in two embed.c files for finite fields. Looks like they haven't shown up on Valgrind yet, but they cause leaks.

[fredrik-johansson]

This looks fine to me. The missing underscore in flint_rand_init has been annoying for a long time.

[albinahlback]

Nemo still needs a fix, but I will push that soon.

[fredrik-johansson]

CI fails on Cygwin.

[albinahlback]

Yes, I saw that. Will fix.

[user202729]

Note that the struct rename breaks backwards compatibility sagemath/sage#39413 .

If backwards compatibility is supported for flint_rand_init maybe it should also be supported for flint_rand_s? Or is the name of that struct not supported to be in the public API?

p/s this might have to do with cython/cython#1984

[albinahlback]

Note that the struct rename breaks backwards compatibility sagemath/sage#39413 .

If backwards compatibility is supported for flint_rand_init maybe it should also be supported for flint_rand_s? Or is the name of that struct not supported to be in the public API?

p/s this might have to do with cython/cython#1984

This is a breaking change. To mitigate this, you can use preprocessor stuff to maintain compatibility with older versions.

[tornaria]

This is a breaking change. To mitigate this, you can use preprocessor stuff to maintain compatibility with older versions.

Sure. However, since the only public use of flint_rand_s afaict was in the prototypes of flint_rand_alloc() and flint_rand_free(), which are deprecated, maybe it's reasonable that you keep those prototypes using flint_rand_s for the time being?

If you eliminate flint_rand_s you may as well just remove those functions instead of deprecating, it seems to be equivalent.

[albinahlback]

Sure. However, since the only public use of flint_rand_s afaict was in the prototypes of flint_rand_alloc() and flint_rand_free(), which are deprecated, maybe it's reasonable that you keep those prototypes using flint_rand_s for the time being?

Sorry, but I will not bring back flint_rand_s.

If you eliminate flint_rand_s you may as well just remove those functions instead of deprecating, it seems to be equivalent.

Seems reasonable.