dexonsmith added a comment. Thanks for working on this, I agree it's a really important problem.
I'm as optimistic as Vedant that this is the right approach though. - On compile time, I do think there's reason to be concerned, since dumping IR was fairly expensive last I checked due to numbering metadata. Have you tried measuring a the time for a bootstrap of Clang? - On stability, I'm skeptical this will be an improvement in the general case. Metadata numbering is unstable, and having that as an input isn't great. It could even change between compiler versions. - I'm also worried that the hash will change any time someone touches the body of the block. That could even cause a change if other code changes a `typedef` which is used inside the block. Is the current numbering per-module? Can we change that somehow to per-function? Then blocks would be numbered by order within a function, which seems fairly stable. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D74813/new/ https://reviews.llvm.org/D74813 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
