c-rhodes wrote: > LGTM. This means we'll be leaning on state a little more, with the risk of > something getting stale, but I guess that's the trade-off for these > optimisations. > > An entirely optional extra bit of validation would be building stage2-clang > with/without this patch and seeing whether it builds an identical binary -- > that usually sets my mind at rest that there aren't any other sneek paths > that are getting stale state. (Not necessary to land IMO).
stage2-clang build is a good idea. With a few config tweaks for reproducibility (`-DLLVM_APPEND_VC_REV=false`, `-ffile-prefix-map`/`-fdebug-prefix-map`) I'm getting identical binaries. I also added instrumentation to verify the cache is kicking in and there's lot of hits. https://github.com/llvm/llvm-project/pull/193918 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
