Fix #4982: Synthesize This nodes with primitive types in hijacked classes.

[sjrd]

No description provided.

[gzm0]

The changes look good, but I'm a bit concerned about not addressing the underlying issue here: Is it really infeasible to run the ClassDef checker on synthesized methods?

I guess this PR is evidence that the method synthesizer is sufficiently complex that class def checking is a good idea :)

[sjrd]

I initially thought this was a somewhat severe issue, but now I don't think it is. The checkers are required to make sure that the input we receive is well-formed and well-typed. If we have linker-internal methods that generate wrong IR, it's not a big deal because we can fix it without care for binary compatibility. It only matters that we produce IR that survives the rest of our pipeline without causing any user-visible bug.

[gzm0]

Ah yes, I agree with this. From a "safety" POV this is not an issue. However, it just feels kind of silly that we have this piece of infrastructure to validate arbitrary IR and we don't use it for code generated in the linker.

A compromise might be to amend the test to use the ClassDef checker instead of a hand crafted test condition.

[sjrd]

A compromise might be to amend the test to use the ClassDef checker instead of a hand crafted test condition.

Done.