jvoung wrote: Sorry for the delay!
I ran a few more tests and the downstream null analysis over a corpus of code with a small change to the patch. The small tweak (below) seems to pass the tests and your fix helps cover more code, as expected =) (e.g., analyze both branches when there is an `if (enum1 == enum2)` vs only one branch) I think the issue is that the downstream code doesn't expect to `nullptr_t` to have a pointer value / modeled nullability state until there is a cast (e.g., comments around https://github.com/google/crubit/blob/859520eca82d60a169fb85cdbf648c57d0a14a99/nullability/pointer_nullability_analysis.cc#L1130) It would take more to change that assumption thoroughly. So a suggestion is to: - do the original fix to `evaluateBooleanEquality` - to handle 'nullptr == nullptr' could special case that in `evaluateBooleanEquality` (e.g., check if LHS and RHS getType().isNullPtrType()) -- instead of creating PointerValues in VisitCXXNullPtrLiteralExpr WDYT ? https://github.com/llvm/llvm-project/pull/129502 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
