Faster conversion between Rational and Double/Float
The proposal for the new implementation of
toRational was discussed (very briefly) here. Supported by Simon Peyton Jones and (off list) Jason Dagit. No dissent.
The old implementation of
Float was extremely slow. Exploiting the representation of the involved types allows an order of magnitude speedup. In particular a fast base-2 logarithm is needed, this has to be implemented in the integer-gmp resp. integer-simple packages, therefore three patches are needed, for base and each of the integer packages.