Re: bignum branch

[Stefan Monnier <monnier@IRO.UMontreal.CA>]

>> > Yes, we need such wrappers in those cases.  Another use case is a
>> > 32-bit build --with-wide-int.
>> Tho, eventually, bignums should make --with-wide-int redundant.
> Only if we allow buffer and string text be referenced by a bignum,

I'd have said "when" rather than "if", but otherwise, yes.

> and if the performance is comparable with --with-wide-int.

Not sure what kind of performance you have in mind: a plain old 32bit
build with bignums will almost inevitably be more efficient than
one --with-wide-int when dealing with buffers <512MB, but it will just
as inevitably be less efficient when dealing with buffers between 512MB
and 2GB.

> Is it reasonable to expect a comparable performance from native 32-bit
> code calculating 64-bit values vs function calls?  I think I'd be
> surprised.

Operations on (small) bignums will be significantly slower than
operations of "long long" 64bit integers, yes.  How this will impact the
overall performance when dealing with buffers between 512MB and 2GB
I don't know: these already tend to suffer from various other
performance problems and I don't know if one will dwarf the other or if
they will make each other more painful.


        Stefan