I looked at its serializing documentation and I was surprized to see that every floating point value is serialized to a double precision value. There are now fo

I looked at its serializing documentation and I was surprized to see that every floating point value is serialized to a double precision value. There are now four types of foating point values and this simplification is downgrading the capabilities of data tranfer. Extended is needed in some scientific computations. Appart from this, the base type encoding is great. The other weird choice is the structure encoding with field numbers and omitting null value fields. My impression is that this is less efficient than coding the null fields. Is this choice based on measurements ?