[ 2024-05-26; Daniel comments ]

I think that all violations of the bullets [rand.req.genl] (1.4), (1.5), and (1.6) are missed opportunities of Mandates (That is: Make the program ill-formed), because they can be all checked (easily) at compile-time, regardless whether we agree on the question to support int8_t/uint8_t (Violations of (1.1), (1.2), and (1.3) still have to be remain undefined because of additional runtime requirements imposed).

Given that I also think that we should consider to either normatively extend all of (1.4), (1.5), and (1.6) to corresponding extended floating point types and extended (unsigned) integer types, or to the minimum make these extended types conditionally-supported with implementation-defined semantics (A word of power that is used at several places).