[PATCH] Add DiagnosticSuppressionScope RAII class

[Richard Smith]

On Mon, Nov 17, 2014 at 8:29 AM, Brad King <brad.king at kitware.com> wrote:

> On 11/12/2014 08:43 PM, Richard Smith wrote:
> > On Fri, Oct 31, 2014 at 6:26 AM, Brad King wrote:
> >> once one error has been emitted the failed
> >> declaration is marked as invalid and any further error resulting from
> >> encountering the invalid declaration should not be emitted.
> >
> > Yes, that's the goal and the intent.
>
> Thanks.  Ideally any declaration that fails to compile as a result
> of encountering another invalid declaration should also be marked
> as invalid.  That way tools can identify and work with the valid
> subset of the translation unit with no false positives.  This is
> useful for tools that work with code that is in progress in an
> editor, for example.  Is that also a goal?
>

Yes, in a restricted sense: if the declaration of an entity is ill-formed,
we want to be aware of that fact when analyzing a use of that entity, to
avoid any follow-on diagnostics. But this doesn't extend arbitrarily far:
if foo is invalid, and the declaration of bar is well-formed but its
definition uses foo, we shouldn't mark the declaration of bar as invalid --
we can and should still use information about bar's declaration in later
semantic checks.

I feel that DiagnosticSuppressionScope is still a valid primitive.
> FrontendAction::EndSourceFile deletes the output file if the
> Diags.hasErrorOccurred() method returns true.  Tools that want to
> produce output even in the face of errors need a way to prevent that
> along with suppressing error diagnostics.  DiagnosticSuppressionScope
> does both cleanly so long as it surrounds the proper scope in the
> tool's source.  Perhaps we can add it with a suitable disclaimer in
> the documentation comments about non-local errors?


I don't think it's reasonable for this to be an RAII primitive, because
once you leave the suppression scope you are left with a
potentially-invalid AST and no guarantee you'll ever see a diagnostic for
future ill-formed code; once you enter this mode, you can't reasonably
leave it again. For that, you can just set the SuppressAllDiagnostics flag
on the DiagnosticsEngine.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20141119/8b7feccf/attachment.html>