[clang] [Clang] Mark this pointer in destructors dead_on_return (PR #166276)

[Aiden Grossman via cfe-commits]

boomanaiden154 wrote:

> Just sloppy wording, sorry. Virtual bases are always live out of the base subobject destructor.

Ack.

> I know there's been feedback before from Google folks about having a flag whenever we expand an optimization so that they can more easily check whether the expansion specifically caused a particular regression.

I'm the person at Google currently driving this. I think we should be fine with a single flag and would probably push back against the introduction of a second flag if we agree to postpone the enablement of adding the attribute for classes with a non-zero number of bases.

https://github.com/llvm/llvm-project/pull/166276