[cfe-dev] Notes in Clang diagnostics and LLVM DiagnosticInfo

[Tobias Grosser]

Hi,

I just saw the LLVM DiagnosticInfo being integrated into clang in
r200931 by Quentin and noticed interesting behaviour when using Notes.

Specifically, I was using the interface in Polly to emit informative 
messages to the user (similar to icc's -vec-report). To emit this 
information I used the Severity 'Note' as these are neither warnings nor 
errors, just pieces of information that help the user understand what 
happens.

However, when running clang I realized my notes are not emitted as 
expected. I tracked it down to this piece of code:

// If the client doesn't care about this message, don't issue it.
// If this is  a note and the last real diagnostic was ignored, ignore
// it too.
   if (DiagLevel == DiagnosticIDs::Ignored ||
       (DiagLevel == DiagnosticIDs::Note &&
        Diag.LastDiagLevel == DiagnosticIDs::Ignored))
     return false;

 From my reading it seems the only way in which clang supports notes is 
as a piece of additional information for 'real diagnostic'. Now I wonder 
what is the right way to emit independent informative messages. I see 
two approaches:

1) Make it possible to create first-class notes, that do not depend on 
previous diagnostics and that are not ignored.

2) Introduce an additional severity level (e.g. Info) that is a first 
class Diagnostic.

2) allows us to also provide 'notes' for our info-diagnostics.

E.g. something like this:

/tmp/test.c:1:6: info: Polly detected an optimizable code region
/tmp/test.c:3:9: note: End of optimizable code region

Does anybody have some ideas/opinions regarding this issue?

Cheers,
Tobias