[ccan] Mathematics Module

[Rusty Russell]

On Monday 26 January 2009 02:02:13 Rupinder Singh wrote:
> Hello,

Hi Rupinder,

> You are correct, and I agree with that. But I think the more complex its
> made, the more application dependent it becomes. Not as basic as str, but I
> believe the modules should be such that the routines are frequently used but
> not many have ever cared to keep them typed. I'll redo it if you say so.
> 
> Just a thought.

Well, gcd is optimal, but there are better methods of establishing primes and squares (your gcd example string is wrong, too BTW).  For a start, I'd make them take unsigned long, rather than int.  Then for squares I believe you can make a faster approximation using the most significant bit, but I'd have to do some math to figure out how.  Google might have an even better algo.

For primes, there's surely a better way than this too.  A classic sieve approach seems like it might fit here, but there are definitely papers on how to do this.

A simple benchmark should demonstrate these (*and* serve as a good testcase).

Oh, I also prefer is_prime, is_square and gcd over BumpyCaps, but that's just a personal thing.

> > Hmm, you had to create an err.h?
> >
> 
> Done it : Can we change the Werror somehow ? So that it doesnt treat the
> warnings as errors

Hmm, it's proven useful in the past to have -Werr.  But delete it for now from the Makefile.

We should really use our own on platforms that deprecate it: err/errx and warn/warnx are really sweet interfaces.

ccan/err?

Thanks!
Rusty.