Re: [boost] [XInt] generic algorithm interface [was: Re: [XInt] review]

Subject: Re: [boost] [XInt] generic algorithm interface [was: Re: [XInt] review]
From: Chad Nelson (chad.thecomfychair_at_[hidden])
Date: 2011-03-09 22:37:15

On Wed, 09 Mar 2011 11:25:57 -0800
"Jeffrey Lee Hellrung, Jr." <jhellrung_at_[hidden]> wrote:

>> The design I see would require all such user-supplied magnitude types
>> to have a reallocation function that the library can call. If that
>> function doesn't provide the memory requested, the user gets only the
>> lower bits of the correct result. The code for handling that is
>> already in the library, for the fixed-length integers.
>
> I'd like to hear more specifics about this idea...if you have more 
> specifics at the moment, that is.

Only the design in my head. XInt would require a few things of the
user-specified type: it must have functions for setting and retrieving
the sign and the used memory, accessing the memory, and getting the
current reserved size of the memory. It must have a function for
allocating more memory, which can be a no-op if you don't want the type
to get any larger (though it won't guarantee the right answer in that
case). It must have a typedef for the digits of the magnitude, and that
type must have properties similar to xint::digit_t. Probably a few
other things, which I won't know until I actually try writing code to
use it.

>> I think that can be handled by putting the onus of dealing with it
>> onto the person providing the new type.
>
> E.g., by supplying a policy class of fundamental operations...? [...]

However the implementor desires, so long as it provides sign and
magnitude in the way that the functions can use.

> Perhaps such discussions on the design of a generic interface are
> best left for after the review.  I couldn't help but throw some ideas
> out there already, though ;)

Yes, there's plenty to do before I even seriously think about that.
-- 
Chad Nelson
Oak Circle Software, Inc.
*
*
*