[Boost-users] Floating Point Utilities Review Result

From: John Maddock (john_at_[hidden])
Date: 2008-03-08 11:06:04

Floating Point Utilities Review Result.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

There were 6 positive reviews and no negative ones, so Johan Rade's Floating 
Point Utility library has been accepted into Boost.

The main issues and topics that came up for discussion were:

1)  A couple of minor bugs were uncovered relating to signed NaN testing 
when using the SSE2 registers, and deserialising NaN's: both were fixed 
during the review period.

2)  We had a discussion about the usefulness or otherwise of signed NaN's 
and the changesign function: Johan agreed to drop support for these as the 
signbit of NaN's is not considered part of the NaN payload and need not be 
preserved by the compiler, this could usefully be summarised in a rationale: 
especially now that negative NaN's are not formatted with a "-" sign?

3)  JM made a few quite fruitless suggestions about optimising away calls to 
memcpy: this should be documented in a rationale to prevent JM from making 
the same mistake again :-)

4) John Phillips raised the issue of finding these tools - making sure that 
they don't get lost amongst all the other special functions and "stuff" in 
Boost.Math.  JM and Johan have agreed to work together to ensure this 
doesn't happen.

5)  There were a couple of suggestions for a convenience function for 
installing the facets in a locale/iostream: Johan agreed to add this.

6)  There was +1 vote for manipulators rather than flags to the facet 
constructors - further investigation suggested this was problematic in a 
header only library - so the suggestion was dropped.

7)  There was one comment about the library not using quickbook for 
documentation: ideally it would be nice to see this fixed so that the docs 
integrate seamlessly with the rest of Boost.Math - but see #4 as well.

Many thanks to Johan for making these utilities available.

John Maddock.
Review Manager for the Floating Point Utilities.