Consider secondary implementation with token injection

[RyanJK5]

P3294 has a motivating example that looks very similar to what protocol is attempting to achieve (see the "Type Erasure" section). As far as I am aware, it is currently the most popular proposal for generating code in C++29 and has some functioning implementations already.

If it helps with standardization, we could pursue a secondary C++29 implementation using token injection.

[jbcoe]

I think that all we really need for standardisation is a proof of concept. Another working (pure C++) implementation cannot hurt but success will hinge on the design and clarity around decisions made rather than quality of implementation. Some proposals are merged without implementations.

That said, in time, I’m sure that clang/GCC will be grateful for a clean implementation.

[RyanJK5]

I think that all we really need for standardisation is a proof of concept. Another working (pure C++) implementation cannot hurt but success will hinge on the design and clarity around decisions made rather than quality of implementation. Some proposals are merged without implementations.

That said, in time, I’m sure that clang/GCC will be grateful for a clean implementation.

Totally makes sense. My only concern is that the duck-style implementation is a bit hacky in some ways and restricts some capabilities that we probably want in the proposal (see constexpr, for example, which we just discussed).

Hearing that the implementation does not need to be perfect is reassuring though. I am sure some of these edge cases that we cannot currently support can be handled on the part of compiler implementers.