[PATCH] D159083: Clang: Don't warn about unused private fields of types declared maybe_unused

[Hans Wennborg via Phabricator via cfe-commits]

hans added inline comments.


================
Comment at: clang/test/SemaCXX/warn-unused-private-field.cpp:291-292
+enum [[maybe_unused]] MaybeUnusedEnum {};
+typedef int MaybeUnusedTypedef [[maybe_unused]];
+class C {
+  MaybeUnusedClass c; // no-warning
----------------
aaron.ballman wrote:
> hans wrote:
> > aaron.ballman wrote:
> > > Let's try one more thing:
> > > ```
> > > template <typename Ty, typename Uy>
> > > class C2 {
> > >   Ty c; // no-warning
> > >   Uy d; // warning
> > > };
> > > 
> > > class [[maybe_unused]] Good {};
> > > class Bad {};
> > > 
> > > C2<Good, Bad> c;
> > > ```
> > The `!FD->getParent()->isDependentContext()` condition prevents the warning from firing on either `C2<>::c` or `d`. (It seems that was there from the beginning: https://github.com/llvm/llvm-project/commit/0baec549a3f49d8c04a0092917f758cc89ef238d#diff-9ced4fa47ee2b9c03b6996ce89a1d131c0f5b71013993bc582209f50d5e934daR1649)
> Hmmm. I think that's intended to ensure we don't diagnose before we've instantiated the template, but once we instantiate with `C2<Good, Bad> c;`, the field decl is no longer in a dependent context and we should still diagnose when instantiating, right?
I tried, and it's not diagnosing: https://godbolt.org/z/osnWbofY5

I'm guessing the motivation was that we shouldn't warn about things that depend on how the template is instantiated, since that could make it hard to write a warning-free template definition.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D159083/new/

https://reviews.llvm.org/D159083