http://gcc.gnu.org/bugzilla/show_bug.cgi?id=59231
--- Comment #7 from Jason Merrill <jason at gcc dot gnu.org> --- (In reply to Paolo Carlini from comment #6) > However, what Jason suggested at the time was "ANOTHER job for > c_inhibit_evaluation_warning" (emphasis mine). In my mind that was important > because I saw a clear consistency rationale in the fix. Before changing / > reverting anything here, we should therefore audit all those uses and figure > out which ones we want to remove and finally decide which replacement. > Frankly, I'm not sure this is 4.9 material (whoever does the job ;) Yes, I saw it as a consistency issue. But now I think that was forcing consistency between two importantly different cases: the purpose of c_inhibit_evaluation_warning is to avoid warning about evaluation issues for expressions that are not actually evaluated. But a template instantiation is not an unevaluated context, so it's a different case. Furthermore, the -Wtype-limits warning still makes sense even in unevaluated context, so it shouldn't be conditional on that. I agree that an audit is needed here; we might want to introduce a new counter to indicate dependent context and suppress some warnings within templates. I think this could still go into 4.9 if there's time to take care of it soon.
