CheckUnused checks span.exists before testing its parts

[som-snytt]

Must check span.exists before using start or point.

Can use pointDelta safely.

It's not obvious that certain checks for isSynthetic should require exists, since those tests will not pass for unpositioned trees. (todo) (synthetic is orthogonal to existing)

Needs a test for producing unpositioned Ident. (todo) It's not unpositioned but spanless. The underlying bug was fixed in 3.3.5 (a macro produced an inlined annotated type tree with the symptom).

Also restore alias "unsafe-warn-patvars". Fix typo in other description.

Fixes #22499

[som-snytt]

The hosted runner: GitHub Actions 13 lost communication with the server. Anything in your workflow that terminates the runner process, starves it for CPU/Memory, or blocks its network access can cause this error.

That's called blaming the victim.

[Linyxus]

The only thing seems to be missing is an accompanying test case. I see that it is left as a todo in the description, what is the status on that? Otherwise LGTM! 👍

[som-snytt]

The (any: @unchecked).isInstanceOf that is inlined from magnolia results in a tree.tpe with the annotation:

  override def transformTypeTree(tree: TypeTree)(using Context): tree.type =
    tree.tpe match
    case AnnotatedType(_, annot) => transformAllDeep(annot.tree)

The annot.tree has the spanless Ident.

Edit: that's what I did not yet reproduce in a dotty test. I assume (without evidence) that it's a bug to see a spanless tree. I suspected tasty, special handling of isInstanceOf, or tree copying for derived. Or "annotations are special".

By cosmic coincidence, it is the unchecked restored the other day in softwaremill/magnolia@5a030d8.

[som-snytt]

I confirmed that tgbot-utils compiles (emits warnings without crashing) under 3.6.3 after ingesting magnolia.

[som-snytt]

Presumably the symptom requires more than separate compilation, but I haven't looked at what's required in a test rig.

[som-snytt]

I will probably give up on constructing a test. I confirmed that 3.3.5 fixes the bug (Ident tree with no span in annotated type).

I submitted a PR to Magnolia, which is used by the failing projects. Magnolia built by 3.3.3 or 3.3.4 results in the spanless tree in derived code after inlining.

For a test, I tried a macro that attempts to reset the tree span to NoSpan, but I couldn't get that to fail.

[som-snytt]

Probably not fixed by #20445
but maybe by #19957

[Linyxus]

Thanks for investigating! LGTM