serialize: Document integer width assumptions we are making when calculating compact sizes

[practicalswift]

• Document integer width assumptions we are making when calculating compact sizes
• Document floating-point width assumptions we are making in the serialization code

Rationale:

• Explicit is better than implicit
• Compile-time failure is better than run time failure :-)

[hebasto]

utACK 25ae1a1

[practicalswift]

@hebasto Updated. Please re-review :-)

[hebasto]

Let's assume a platform with sizeof(short) > 2 or sizeof(int) > 4.
Without static_assert statement the logic will remain correct if
nSize <= std::numeric_limits<unsigned short>::max() replace with nSize <= 0xffff
and
nSize <= std::numeric_limits<unsigned int>::max() replace with nSize <= 0xffffffff

[practicalswift]

@hebasto As I read the code more adjustments would be needed to the serialisation code to make it work properly under such a platform. I don't think changing GetSizeOfCompactSize and WriteCompactSize would be enough?

FWIW I think the platform requirements are reasonable. My goal with this PR is just making the platform requirements explicit and immediately obvious to compilers, humans and static analysers :-)

I think a compile time check is the most elegant and non-intrusive way to achieve that goal.

[hebasto]

FWIW I think the platform requirements are reasonable. My goal with this PR is just making the platform requirements explicit and immediately obvious to compilers, humans and static analysers :-)

I think a compile time check is the most elegant and non-intrusive way to achieve that goal.

Agree. ACK 4486f2a