[PATCH] D61479: Finish "Adapt -fsanitize=function to SANITIZER_NON_UNIQUE_TYPEINFO"

[Vedant Kumar via Phabricator via llvm-commits]

vsk added a comment.

This looks reasonable to me. I'd prefer to disallow -fsanitize=function in combination with -fsanitize-minimal-runtime, as I'm not sure it's appropriate for the minimal runtime to handle SANITIZER_NON_UNIQUE_TYPEINFO incorrectly. If this has already been discussed elsewhere I'd appreciate a pointer.

I have a more minor concern about the complexity of the ubsan test's RUN lines and \#include magic. I'm worried that these factors could make the test brittle (w.r.t unexpected directory structures or platforms that don't support $(...) expressions). But I don't have a clearly superior alternative to suggest. One idea is to a) redefine `sanitizer_non_unique_typeinfo` in the lit config and b) define %clangxx to include a path to the sanitizer_common headers. That might make the test easier to read, but not less brittle. Do you see this as a problem and if so are there any alternatives you might suggest?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61479/new/

https://reviews.llvm.org/D61479