@alanz Good catch! Changed to

A rational consists of an optional -, followed by one or more digits 0 - 9, followed by a /, followed by one or more digits 0 - 9.

I also realized the canonical representation of floats is ambiguous, it would allow both 0.1E0 and 0.01E1. (Hopefully) fixed by changing to

Floats must be encoded such that there is a 0 left of the decimal point, a nonzero digit after the decimal point, and the exponent must always be included.

Equivalently, it could mandate to always use the lowest possible exponent (still with a single 0 left of the decimal point), but the other formulations seems easier to understand to me.

I just found a canonical float encoding in the XSD spec, which might be better than rolling my own.

Also, simply printing and reading floats is apparently rather complicated.

I fear I'll have to read the full IEEE 754 standard and at least those papers, if I want to make sdn a real thing.