If a diagnostic would be emitted but it has an unknown range, we sometimes drop it on the floor. This seems non-ideal.
Repro steps
See the F# compiler built from https://github.com/Smaug123/fsharp-1/commit/76ea258b45dbea90e2ea28069d76065f1de6c354 . The test "Three types refined to obj are all warned" in ObjInference.fs incorrectly fails because it reports that no diagnostics were emitted. In fact diagnostics were requested, but were suppressed just before reporting: place a breakpoint at FSharpDiagnostic.fs line 203 in ReportDiagnostic to observe our decision not to emit the errors that arose on line 185 of TypeRelations.fs because their associated ranges were all the unknown range.
Expected behavior
The test "Three types refined to obj are all warned" should pass, because it notes that three diagnostics were emitted.
Actual behavior
No diagnostics were emitted, so the test fails.
Known workarounds
We could possibly just make sure that every diagnostic always has a meaningful range (which seems invasive and difficult), or we could run in "all errors" mode (but I don't understand the implications of that).
Smaug123
commented
1 year ago
If a diagnostic would be emitted but it has an unknown range, we sometimes drop it on the floor. This seems non-ideal.
Repro steps
See the F# compiler built from https://github.com/Smaug123/fsharp-1/commit/76ea258b45dbea90e2ea28069d76065f1de6c354 . The test "Three types refined to obj are all warned" in ObjInference.fs incorrectly fails because it reports that no diagnostics were emitted. In fact diagnostics were requested, but were suppressed just before reporting: place a breakpoint at FSharpDiagnostic.fs line 203 in
ReportDiagnosticto observe our decision not to emit the errors that arose on line 185 of TypeRelations.fs because their associated ranges were all the unknown range.Expected behavior
The test "Three types refined to obj are all warned" should pass, because it notes that three diagnostics were emitted.
Actual behavior
No diagnostics were emitted, so the test fails.
Known workarounds
We could possibly just make sure that every diagnostic always has a meaningful range (which seems invasive and difficult), or we could run in "all errors" mode (but I don't understand the implications of that).