[PATCH] D40625: Harmonizing attribute GNU/C++ spellings

[Aaron Ballman via Phabricator via cfe-commits]

aaron.ballman added inline comments.


================
Comment at: include/clang/Basic/Attr.td:602
 def AnalyzerNoReturn : InheritableAttr {
-  let Spellings = [GNU<"analyzer_noreturn">];
+  let Spellings = [Clang<"analyzer_noreturn">];
   let Documentation = [Undocumented];
----------------
dcoughlin wrote:
> aaron.ballman wrote:
> > rsmith wrote:
> > > Hmm, should the clang static analyzer reuse the `clang::` namespace, or should it get its own?
> > Good question, I don't have strong opinions on the answer here, but perhaps @dcoughlin does?
> > 
> > If we want to use a separate namespace for the analyzer, would we want to use that same namespace for any clang-tidy specific attributes? Or should clang-tidy get its own namespace? (Do we ever plan to execute clang-tidy through the clang driver? That might change our answer.)
> How would this look if we added a special namespace for the clang static analyzer? Would this lead to duplication (say, [[clang_analyzer::analyzer_noreturn]]) so that we keep the "analyzer_" prefix for __attribute__((analyzer_noreturn))? Or could we have the "analyzer_" prefix only for GNU-style attributes but not for C++ (for example, [[clang_analyzer::noreturn]])?
> 
> As for clang-tidy, I think it probably makes sense for it to have its own namespace, but we should ask @alexfh.
> How would this look if we added a special namespace for the clang static analyzer? Would this lead to duplication (say, [[clang_analyzer::analyzer_noreturn]]) so that we keep the "analyzer_" prefix for attribute((analyzer_noreturn))? Or could we have the "analyzer_" prefix only for GNU-style attributes but not for C++ (for example, [[clang_analyzer::noreturn]])?

We have the ability to do whatever we'd like there. Given that the semantics are so similar to `[[noreturn]]`, I think it would be reasonable to use `[[clang_analyzer::noreturn]]` and `__attribute__((analyzer_noreturn))` if that's the direction you think is best.

> As for clang-tidy, I think it probably makes sense for it to have its own namespace, but we should ask @alexfh.

I'm less enthusiastic about giving clang-tidy a vendor namespace that's separate from the static analyzer, should the need arise. My biggest concern there is that I would *really* like to see clang-tidy be more tightly integrated with the clang driver (so users don't have to manually execute a secondary tool). If that were to happen, then the user experience would be that there are two vendor namespaces both related to analyzer attributes.

That said, I would also not be opposed to putting all of these attributes under the `clang` vendor namespace and not having a separate vendor for the analyzer or clang-tidy.


https://reviews.llvm.org/D40625