Re: [boost] GGL Review

Subject: Re: [boost] GGL Review
From: Jonathan Franklin (franklin.jonathan_at_[hidden])
Date: 2009-11-16 11:53:31

On Mon, Nov 16, 2009 at 9:33 AM, Brandon Kohn <blkohn_at_[hidden]> wrote:
> Jonathan Franklin wrote:
>>
>> In my application domain, I'm really not that interested in numerical
>> instability due to floating point imprecision.  For my use-cases, a
>> few things work fine w/ single precision, and double precision works
>> well for everything else.
>
> Yes, but for something like boolean operations, this means it works
> sometimes and completely fails at others.

Well, for me, it means that it *almost always* works, and that *one*
time it failed, I didn't notice.
;-)

> I can see the logic of accepting
> the uncertainty and hoping for the best in practice, but it does beg the
> question whether Boost should contain such libraries.

As long as the known or suspected numerical issues are documented,
then the user can make her own judgment as to whether the library is
suitable for her needs.

Clearly, there are people on this list that need complete and total
numerical stability.

> There are many buggy
> floating point boolean op libraries out there. Why do we need another?

Do we have a *standard* buggy FP boolean op library?  Because I need
one to build some other buggy algorithms on top of.
;-)

And are we talking about numerical floating point issues, or
algorithmic bugs?  I don't consider those to be the same thing.

Jon