I also still think that whoever came up with that specification needs a whap upside the head, but then I've never encountered a reason to print a floating-point value in hex, so I assume there's some justification for it.

Gavin, Hexfloats are useful for comparison of numerical accuracy. We use them in the Boost.Math unit tests when outputting ULP distances. https://github.com/boostorg/math/blob/develop/test/math_unit_test.hpp#L136 Matt