NoQ added inline comments.
================ Comment at: clang/include/clang/StaticAnalyzer/Core/PathSensitive/DynamicTypeMap.h:32 +/// A set of dynamic cast informations. +REGISTER_SET_WITH_PROGRAMSTATE(DynamicCastSet, clang::ento::DynamicCastInfo) ---------------- Charusso wrote: > NoQ wrote: > > Emm, so you're saving successes and failures of all casts //regardless of > > which object is getting casted//? That's definitely not sufficient. If `X` > > is a `Stmt` that isn't an `Expr`, you can't automatically infer that `Y` is > > a `Stmt` that isn't an `Expr` for any object `Y` other than `X` . This > > information needs to be per-object. Then you'll need to clean it up in > > `checkDeadSymbols`. > I have two implementations, the other is set factory based on memory regions. > Is it would be a good idea to go through all the regions every time and all > their casts to know every possible cast-assumption? I have made it, and I > felt like it is overcomplicated as the DynamicTypeMap could be used for > asking what is the type now and whether we have better precision and update > that information therefore we could update the cast-set as well. It's about correctness, we don't have much choice here. The current data structure simply cannot work correctly because there's a "one vs many" problem in it: for every pair of types (T₁, T₂) there are *many* possible outcomes of a cast from T₁ to T₂ (depending on the object that's being casted) but the current data structure only has room for *one* such outcome. Your data structure is basically saying "Oh, this shape turned out to be a circle, let's from now on forever believe that triangles don't exist" (?) CHANGES SINCE LAST ACTION https://reviews.llvm.org/D66325/new/ https://reviews.llvm.org/D66325 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
