GMP 4.2.1 release candidate available

[Pedro Gimeno]

Sorry for the late reply.

On 08/16/2006 04:50:29 PM, Camm Maguire wrote:

> We use gmp for random numbers.  Latest gmp is giving us some problems
> with random states, returned by gmp_randinit_default:
> 
> 1) the 625 word seed area is allocated with gmp_alloc, but the
>    mp_alloc field is not set.  We need this to know how to gc this
>    object.  Right now we are using a temporary hack to trap the
>    allocated size.

I've submitted a patch to set the allocation size field properly.

> 2) A pointer to a static area is returned in _mp_algdata->_mp_lc
>    (e.g. &Mersenne_Twister_Generator).  Please make this unstatic,
>    i.e. export the struct to the user.  It is very common in these
>    programs to dump the running memory image to a file, and restart it
>    later.  If gmp is dynamically linked, the load address can be
>    different, and aforementioned pointer, stored in the heap, rendered
>    invalid.  If the structure was exported, I could refer to it in a
>    gcl C file, and rely on ldd to do the relocation.  Alternatively,
>    if randseed_mt was non-static, we could do the same for the
>    elements of gmp_randfnptr_t as opposed to the body.

In my opinion, exporting private structures is not the proper way to  
deal with dumping. Instead there are plans for implementing random  
state export/import functions which, I believe, will fit your needs  
much better.

In the meantime you can mirror the private struct to do what you intend.

-- Pedro Gimeno