[Gcl-devel] Re: GCL and GMP

[Kevin Ryde]

Camm Maguire <address@hidden> writes:
>
> 1) all arguments distinct -- lonjmp back to the beginning of the call,
>    gc, and try the call again.

Sounds good.  Better than patching gmp to re-fetch pointers from mpz_t
after each memory operation.  (Pretty hairy that, since obviously gmp
is typical C code and plays fast and loose with pointer arithmetic.)

Depending how the evaluator is structured you might even be able to
have a single setjmp point to go back to, rather than a new setjmp on
every mpz call.

> 2) destination arg(s) match some source arg(s) -- satisfy allocation
>    by adding new pages to the heap and schedule a gc call after the
>    gmp call finishes.

I guess if getting close to full you could gc beforehand as a
preventative, ie. to avoid unnecessarily growing the heap.

It shouldn't be too hard to have a stab at what space a gmp call will
want, though it may be unwise to rely on getting that right every time
(since obviously new algorithms or whatever in gmp could change it
very easily).