[PATCH] D100158: [SantizerCoverage] test case demonstrating PR39531
Nick Desaulniers via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Fri Apr 9 09:52:39 PDT 2021
nickdesaulniers added a comment.
In D100158#2678482 <https://reviews.llvm.org/D100158#2678482>, @dblaikie wrote:
> My best guess would be that any call synthesized from an instruction without a location should probably just get a zero line scoped directly to the current function.
Sure. Should we do that unconditionally for the inserted `CallInst`, or predicate it on debug info being requested for the compilation, or something else?
> It's not ideal (not only the lack of location, but it also punches holes in scopes which can bloat debug info/make stepping more difficult as you hit discontiguous hunks of lines potentially, etc)
For the purpose of ModuleSanitizerCoveragePass, I think that's ok. Developers accept a large increase in binary size anyways using the most aggressive level of tracing every edge, and stepping in a debugger probably isn't the use case folks are intending when using ModuleSanitizerCoveragePass. I would think the major use case for debug info with coverage is to symbolicate traces.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D100158/new/
https://reviews.llvm.org/D100158
More information about the llvm-commits
mailing list