aaron.ballman added a comment. In D108560#3167621 <https://reviews.llvm.org/D108560#3167621>, @ymandel wrote:
> Aaron, Salman, > > We've seen a serious perfomance regression caused by this patch. The issue is > that the new code scans every file in its entirety for every single > diagnostic encountered in that file. In fact, each lines is copied and > scanned twice (respectively), so it's like O(n * m), n = size of files, m = > number of diagnostics, with high constant factor. Oof, good catch! > Can we roll back this patch until a fix is developed? At the least, we'd > request that this feature be disabled by default, either with a flag or in > the preprocessor. Additionally, we'd ask you to reconsider the design to > ensure that each file is processed at most once (via a cache or whatnot). > We'd be happy to review the patches. I'm fine (if a bit sad) with reverting these changes until we have a better approach. @salman-javed-nz , can you look into that (either the revert or the fix)? Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.org/D108560/new/ https://reviews.llvm.org/D108560 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
