SemanticDB diagnostics missing

[bjaglin]

Compiler version

3.3.0-RC6 (but I believe this is not a regression, and is reproducible for any 3.x)

Minimized code

//> using scala 3.3.0-RC6
//> using options -Wunused:all -Ysemanticdb

import java.util.List

object Test

Output

➜  metap Test.scala.semanticdb
Test.scala
--------------

Summary:
Schema => SemanticDB v4
Uri => tmp/Test.scala
Text => empty
Language => Scala
Symbols => 1 entries
Occurrences => 4 entries

Symbols:
_empty_/Test. => final object Test extends Object { self: Test.type => +1 decls }

Occurrences:
[3:7..3:11) => java/
[3:12..3:16) => java/util/
[3:17..3:21) => java/util/List#
[5:7..5:11) <= _empty_/Test.

Expected

 ➜  metap Test.scala.semanticdb
 Test.scala
 --------------

 Summary:
 Schema => SemanticDB v4
 Uri => tmp/Test.scala
 Text => empty
 Language => Scala
 Symbols => 1 entries
 Occurrences => 4 entries

 Symbols:
 _empty_/Test. => final object Test extends Object { self: Test.type => +1 decls }

 Occurrences:
 [3:7..3:11) => java/
 [3:12..3:16) => java/util/
 [3:17..3:21) => java/util/List#
 [5:7..5:11) <= _empty_/Test.

+Diagnostics:
+[3:17..3:21) [warning] Unused import

[tgodzik]

I created very quick try, but it seems we are missing the warnings. Not sure how to extract the diagnostics from context.

Tried blindly to use ctx.reproter.pendingMessages but trying it out in a local project didn't yield any success.

https://github.com/lampepfl/dotty/compare/main...tgodzik:dotty:diags-semanticdb?expand=1

Anyone can help out here?

[tgodzik]

Now this seems to work if a StoreReporter is used, but any reporter that just prints the information right away will lose the info. What is the least hacky way to get all warnings even if they were already reported?

[WojciechMazur]

@tanishiking Can you look at that in your spare time? I think you had some experience with SemanticDB in the past

[tgodzik]

@tanishiking Can you look at that in your spare time? I think you had some experience with SemanticDB in the past

I think we might need to some additional output from someone closer to the core compiler. If we have a ConsoleReporter there is no way to get the warnings from pending messages, so would be great to know if we can work around it. @szymon-rd had worked recently with unused warnings, maybe he knows something?

[tanishiking]

I can take a look on SemanticDB generation side, but it would be great if someone knows how can we retrieve the diagnostics via ctx

[szymon-rd]

We can change how we aggregate diagnostics in the CheckUnused and linting; that change could be introduced in a patch (3.3.1?). What I find interesting is that, if I understand correctly, the semanticdb file is extracted after the typer. We don't have the lining diagnostic information at that point yet. Maybe that's even the cause of it missing?

[tgodzik]

What I find interesting is that, if I understand correctly, the semanticdb file is extracted after the typer. We don't have the lining diagnostic information at that point yet. Maybe that's even the cause of it missing?

Och, when is the linter run then? In which phase?

[szymon-rd]

Most of the linting is run after the typer, but some linting requires information from inlining. Look where the CheckUnused.PostInlining is in Compiler.scala. Warnings are emitted at the end of this phase, not during the CheckUnused.PostTyper.

[tanishiking]

Thanks @szymon-rd !
Ok, dotty runs CheckUnused at PostTyper and PostInlining then report the intersection of both phases as unused warnings #16876. Therefore we have to extract the diagnostics into SemanticDB after the CheckUnused.PostInlining phase.

We may want to add an extra phase that extract the diagnostics and write the SemanticDB file.