avoid kind type dispatch bugs in tuple_type_head and tuple_type_tail

[jrevels]

This is a much more narrowly-scoped alternative to some of the tuple code changes that were originally in #27736 (i.e. c16a109).

It would still be nice eventually to figure out alternative implementations for c16a109, since switching to fieldtype is more sound in general (see discussion from #27736). However, for now, this much more palatable change works around the problems that are blocking Cassette.

Unfortunately, I don't have a non-Cassette test case for these bugs yet; they're pretty hard to work out MWEs for.

[jrevels]

will merge tomorrow if there are no objections

[jrevels]

(let's make sure nanosoldier doesn't object)

@nanosoldier runbenchmarks(ALL, vs = ":master")

[martinholters]

I didn't follow along what issues this is solving, but ... can't they be tested for?

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[jrevels]

I didn't follow along what issues this is solving, but ... can't they be tested for?

Definitely, it'd be great to have a non-Cassette MWE here, I just haven't figured one out yet. From the OP:

Unfortunately, I don't have a non-Cassette test case for these bugs yet; they're pretty hard to work out MWEs for.

If you're curious, here's a Cassette-based one I posted: #27736 (comment)

Looks like Nanosoldier doesn't like the new implementations...maybe adding a @_pure_meta to tuple_type_head will fix it? Let's see...

[jrevels]

@nanosoldier runbenchmarks(ALL, vs = ":master")

[ararslan]

Had to restart the server. @nanosoldier runbenchmarks(ALL, vs=":master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @ararslan

[jrevels]

Cool, the @_pure_meta seemed to do the trick.