https://github.com/steakhal commented:
> > I think this would still trip on the same memory issue. The problem to me > > is that we read uninitialized memory. So even if you limit the bounds of > > the result of the read of such uninitialized memory, it would only mask and > > limit the effect of such read. Have you tried running the failing case via > > valgrind? That's usually great for catching uninitialized reads. msan is > > also great, but that needs an instrumented standard library which may be > > tough to set up. > > I don't think so. I have an asan build locally and it worked. > > The methods `getParent()` and `getIndexInBlock()` in CFGElementRef does not > access memory, since they only return the value from the ref. The actual > dereference happens in `->` operator, which is gated after the newly added > `if` condition. My understanding is that asan will not check if a memory is initialized or not. It checks if the access is valid - but since you guard the access we would brush this under the carpet now. https://github.com/llvm/llvm-project/pull/139980 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
