Re: Pushing the `gnus-range-*' functions down into the C layer

[Ted Zlatanov <tzz@lifelogs.com>]

On Fri, 10 Sep 2010 16:01:41 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> Ted Zlatanov <tzz@lifelogs.com> writes:
>> 64-bit int support in Emacs should not require a 64-bit system.  If they
>> are natively available it's faster, that's all.  It's not like you'd be
>> packing and unpacking the bits directly.

LMI> Yeah.  Bignum support in Emacs would totally rock.  No more manual
LMI> conversions to floating point numbers to do safe-ish computations on
LMI> numbers that might be too big would be needed.

LMI> But I suspect that it's a kinda big project.  :-)

If everything was inside a num64-* namespace and a num64.el package,
it's a pretty easy implementation.  I'm not an expert on numeric
algorithms[1] but I'd guess the internals can be pretty easily
implemented in ELisp as a list of up to 3 native 28-bit ints.  In fact
I'm surprised that doesn't exist yet; maybe I've missed it.

Of course everyone will want 64-bit numbers native from the very start.
I say that would delay the availability of the feature; just make people
do (featurep 'num64) and get it done ASAP.

Specifically for Gnus purposes, if range-* knew about the num64 API, it
would be sufficient in order to use it.

Ted

[1] but I know enough not to assume anything at that level is trivial