Re: [boost] [Review Request: Floating Point Utilities]

From: John Maddock (john_at_[hidden])
Date: 2008-01-02 11:59:13

Johan Råde wrote:
> There is also an overlap with the Math Tool Kit library;
> both libraries provide implementations of fpclassify, isnan etc.
> (The two implementations use completely different approaches.)
>
> My plan has been to have the library included in Boost as an addition
> to Boost.Math
> rather than as an independent library.
>
> However, I'm very busy; the next year will probably decide whether my
> company
> becomes a success or a failure.

The very best of luck with that :-)

Johan, I must apologise, I've been meaning to volunteer as review manager 
for this: if accepted the current Boost.Math implementation could be merged 
with this one (especially as the current Boost.Math implementation already 
has the pp-logic to use the platform's native implementation where 
available).  IMO the only things stopping this being integrated into 
Boost.Math as is (given that it doesn't change the interface, only the back 
end implementation) are:

1) A good eyeball once-over by interested parties, to spot potential issues: 
since the library uses non-portable casts etc this is quite important IMO.
2) The knowledge to know when to use this implementation, and when/where it 
might fail.

So.... I'd also ask you to keep up the review request - if no showstoppers 
turn up, then it can just be integrated into Boost.Math more or less as is - 
hopefully without too much work for yourself or Paul and me.  If there are 
any changes required then we'll have to figure out what to do about them: 
possibly just restrict the platforms/compilers where your implementation is 
used.

So hopefully, your involvement could be restricted to answering queries 
about the implementation during the review: is that possible given your 
current schedule?

Best regards, John Maddock.