rjmccall added a comment. My point is that IR pointer type information can't be "incorrect", because there is no semantic meaning for pointer types in LLVM IR. That has never been a part of the semantics of LLVM IR. Even if it were possible to change that — and I don't think it is, not without major changes, although I can very much understand why someone looking at compiler output might think "but we're so close right now" — we're not going to make that effort just for the benefit of one specific analysis, especially given that the consensus position is still that we should go further in the opposite direction and make pointer types stop carrying this meaningless information.
https://reviews.llvm.org/D49403 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
