ChuanqiXu9 wrote: > I'd feel uncomfortable exposing a `noload_redecls` interface because it can > potentially break the AST invariants. I think allowing non-modules developers > to choose when to complete a redeclaration chain will introduce more bugs > than it can fix ;) > > I'd propose to move forward with `getMostRecentExistingDecl` if it does the > job. We can have a separate PR discussing the ergonomics if there is a strong > feeling about going this way.
I'd like it as: (1) We already have `noload_lookup` and other similar "unpublic" API (e.g., `getOwningModuleID` `invalidateCachedLinkage`) in DeclBase.h. (2) Merging Redecls is a major performance hurting point of using modules. (3) In the past, there are many "fix" which just tried to mitigate the bug by loading all redecls without understanding the real problem. https://github.com/llvm/llvm-project/pull/133057 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
