ChuanqiXu9 wrote: > The newest version of this patch still doesn't work correctly. Switching the > new `LoadExternalSpecializationsLazily` to disabled by default (somehow the > argument didn't work on its own) instead crashes, most of the cases involving > `MultiOnDiskHashTable`. I suspect some kind of memory error maybe? Sorry to > not be more helpful at the moment...
Thanks for testing it. It shows it is really complex. Maybe I am missing some points. > How about taking a step at a time with this patch. Perhaps we should > introduce the on-disk hash table infrastructure and always deserialize > everything. Then we can verify that part works on our build infrastructure > and then move on with the deferring the template loading. I believe that > should be relatively easy to achieve with the current version of the patch. > > Essentially I am proposing making `-fno-load-external-specializations-lazily` > default so that we can test it without having to modify build > infrastructure... Good suggestion. It sounds good to me. Then we can test it and improve it step by step. I've made it. Now anything in the patch shouldn't be accessed without `-fload-external-specializations-lazily`. https://github.com/llvm/llvm-project/pull/76774 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
