Fix checking ctx to carry correct modes

[noti0na1]

In TreeChecker, the context for checking is not initialized properly. val checkingCtx uses setMode(Mode.ImplicitsEnabled), which removes other modes in the ctx.

This causes Mode.SafeNulls removed from the context. So the checking for explicit nulls tests is always running with unsafe nulls enabled!!

This PR fixs the checking ctx and some other bugs exposed by the correct checking:

• Remove Imports during Erasure, so all other phases before it can be checked correctly
• Cast the last expression of a block to the expected type,
  if unsafe nulls is enabled inside but not enabled outside
  and the type does not conform the expected type without unsafe nulls.
  See: tests/explicit-nulls/pos/unsafe-block.scala
• Use unsafe nulls when forwarding a member before erasure if the target is Java defined
• Cast null to AnyRef when generating testNotNull
• When creating nameRef in SyntheticMembers.scala, cast the result of name to String if isJavaEnumValue

#15096 is blocked by this PR, #14946 will also wait for this

[noti0na1]

test performance please

[dottybot]

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

[dottybot]

Performance test finished successfully:

Visit https://dotty-bench.epfl.ch/15350/ to see the changes.

Benchmarks is based on merging with main (42b5941)

[odersky]

I think it's better to drop imports in Erasure during the regular tree pass. E.g. add:

  override def typedImport(imp: untpd.Import, sym: Symbol)(using Context) = EmptyTree

[noti0na1]

test performance please

[dottybot]

performance test scheduled: 3 job(s) in queue, 1 running.

[noti0na1]

Hi @odersky , I removed the DropImports phase and drop imports in Erasure now. I also fixed two other bugs related to checking since last review.

This PR is ready for another review now. Thanks

[dottybot]

Performance test finished successfully:

Visit https://dotty-bench.epfl.ch/15350/ to see the changes.

Benchmarks is based on merging with main (140693d)

[olhotak]

LGTM

[odersky]

Otherwise LGTM

[odersky]

The comment does not correspond to the code. exprCtx is the context inside the block valid for the final expression. ctx is the context outside the block. The cast is performed if unsafeNulls is disabled inside the block but enabled around it. Is this what's intended? Either the comment or the code has to be fixed.

[noti0na1]

I think the comment is correct: !exprCtx.mode.is(Mode.SafeNulls) indicates SafeNulls is disabled inside the block, which means unsafeNulls is enabled here.

[olhotak]

The confusion is that the mode is SafeNulls which is a negation of unsafeNulls. I think the comment correctly corresponds to the code.

[odersky]

Yes, my bad. I misread the code.