Expand non-transparent macros after Typer

[nicolasstucki]

The technical change

• Inline transparent inline definitions in typer
• Inline other inline after typer.

Sematic changes

• inline given
  • transparent inline given f: T = ...: does the same as all inline definition did before
    • transparent inline given f: T = (...): T can be used to emulate the previous version inline given
  • inline given f: T = ...: if an error is emitted during the inlining then the error is reported to the user.
• scala.compiletime.testing.{typeChecks, typeCheckErrors} must be used directly or inside a transparent inline
• Add missing type parameter bound checks for non-transparent inlines (see Missing bounds check for transparent inline type parameters #10552). Transparent inlines still have this issue.

[dottybot]

performance test scheduled: 9 job(s) in queue, 1 running.

[odersky]

Can we try without overriding typer and test performance again?

[odersky]

Why is this needed here? Don't we get to an Ident or Select eventually, which sets needsInlining. Also, why do we handle Apply here but not TypeApply. If this is in fact needed for some reason it would be good to add a comment.

[nicolasstucki]

It is redundant. I removed it.

[odersky]

Would this be triggered by

inline val x = 2

? Those things are extremely common, so we should avoid running the inliner just for these.

[nicolasstucki]

Yes, then we would need to check that the RHS of an inline val is a constant after the Inlining phase. I try to fit it into another phase.

[odersky]

That would cause many more traversals, so yes, good to do this somewhere else.

[odersky]

In fact, it might be better to do this test only in typedApply and typedTypeApply. Concentrating everything in typed is more elegant but might have performance implications. I am also nervous about making calls to typed more megamorphic.

[nicolasstucki]

I moved the logic back to typedApply, typedTypeApply, typedIdent and typedSelect. I kept the rest of the refactoring which avoids the duplication of the code that is now in needsInlining.

[dottybot]

Performance test finished successfully:

Visit http://dotty-bench.epfl.ch/9984/ to see the changes.

Benchmarks is based on merging with master (a63b1d1)

[nicolasstucki]

test performance with #quotes please

[dottybot]

performance test scheduled: 1 job(s) in queue, 0 running.

[dottybot]

Performance test finished successfully:

Visit http://dotty-bench.epfl.ch/9984/ to see the changes.

Benchmarks is based on merging with master (16bf86f)

[nicolasstucki]

@liufengyun could you have a look at the new commits?

[liufengyun]

LGTM

[deusaquilus]

@nicolasstucki It seems like this is working for me out of the box! I don't need to modify anything.
I do have one question though. If the typing phase is now before blackbox macros, how come the Quill parser is running after a typing mistake.

Say I make a typing mistake on purpose:

    inline def q = quote {
      query[SanePerson] // hello
    }
    inline def name(p: SanePerson) = p.name
    inline def qq = quote {
      q.map(p => name(p.name)) // purposely making a typing mistake
    }

Why does the macro even run if there was an error on a previous phase?