jyknight wrote: I'm trying to update some non-in-tree code after this change, and I don't think I understand the newly-expected lifetime rules after this change. I see all the comments that say this makes lifetime clearer...so I expect I'm just misunderstanding things.
Previously, DiagnosticsEngine kept a DiagnosticOptions member via IntrusiveRefCntPtr, which meant the DiagnosticsOptions lifetime would be at least as long as the DiagnosticsEngine. Now, it stores a reference. Which means, I think, that the creator of a DiagnosticsEngine needs to somehow ensure by themselves that the DiagnosticsOptions value has a sufficiently long lifetime. But this seems very hard to figure out what lifetimes should be now, because DiagnosticsEngine is, itself, refcounted. How does it make sense to have a borrowed reference to DiagnosticsOption in the refcounted DiagnosticsEngine type? ISTM the caller who creates a DiagnosticsEngine cannot now easily know what the lifetime of the DiagnosticsEngine is going to be, and thus, has no way to ensure the DiagnosticsOptions lifetime is correct? https://github.com/llvm/llvm-project/pull/139584 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
