user stack support for TMP_ALLOC

[Bill Allombert]

On Thu, Oct 07, 2004 at 05:57:39PM +0200, Torbjorn Granlund wrote:
> Bill Allombert <allomber at math.u-bordeaux.fr> writes:
> 
>   The only issue we have with this multiprecision kernel is the alloca stack
>   overflow problem.
>   
> We are changning the use of alloca for GMP 5.  It will never allocate large
> chunks on the stack.  That change should make the stack usage very small.

That is a very good news! Also I see on your website you plan to add
more mpn_ routine, that also will be very useful to us.

>   I have written a patch [snip]
> 
> I am unfortunately rejecting your patch.  Some usages of alloca in GMP are
> too critical for going through a mechanism such as the one you are
> proposing.

Are you implying that --enable-alloca=stack would be slower than
--enable-alloca=noreentrant, or did I seriously misunderstood how
TMP_ALLOC is used ?

Anyway, GMP 5 should make this patch useless.

Cheers,
Bill.