This should go through L(E)WG to form a policy opinion; maybe they want to deprecate make_pair.

This doesn't seem editorial.

I don't see that as encouragement to use the function, just a statement of fact, andstd::make_pair doesn't behave exactly the same as CTAD anyway.

andstd::make_pair doesn't behave exactly the same as CTAD anyway.

Well, that's unfortunate. It's true that pair's deduction guide doesn't use unwrap_ref_decay_t. It's still not too late to make them consistent to ease transition in case of deprecating make_pair. Actually not, CTAD is a C++17 feature, so transition seems hard without tools.

You assume making them behave the same is desirable. The different behaviour is intentional, not an oversight.

This is one of the many clean-up papers I have half written that never finishes in time for any mailing deadline. If anyone else thinks this is worth pursuing, I will try to focus and finish for the pre-Kona 2019 mailing. This affects use of both make_pair and make_tuple, and we should choose between CTAD and make_* precisely to indicate whether or not unwrap_ref_decay behavior is expected.