[PATCH] D75677: [Analyzer] Only add iterator note tags to the operations of the affected iterators

[Artem Dergachev via Phabricator via cfe-commits]

NoQ added inline comments.
Herald added a subscriber: ASDenysPetrov.


================
Comment at: clang/lib/StaticAnalyzer/Checkers/IteratorModeling.cpp:541-542
+        BR.markInteresting(It1);
+        if (const auto &LCV1 = It1.getAs<nonloc::LazyCompoundVal>()) {
+          BR.markInteresting(LCV1->getRegion());
+        }
----------------
baloghadamsoftware wrote:
> NoQ wrote:
> > I'm opposed to this code for the same reason that i'm opposed to it in the debug checker. Parent region is an undocumented implementation detail of `RegionStore`. It is supposed to be immaterial to the user. You should not rely on its exact value.
> > 
> > @baloghadamsoftware Can we eliminate all such code from iterator checkers and instead identify all iterators by regions in which they're stored? Does my improved C++ support help with this anyhow whenever it kicks in?
> How to find the region where it is stored? I am open to find better solutions, but it was the only one I could find so far. If we ignore `LazyCompoundVal` then everything falls apart, we can remove all the iterator-related checkers.
It's the region from which you loaded it. If you obtained it as `Call.getArgSVal()` then it's the parameter region for the call. If you obtained it as `Call.getReturnValue()` then it's the target region can be obtained by looking at the //construction context// for the call.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D75677/new/

https://reviews.llvm.org/D75677