[PATCH] D102273: [analyzer] LoopUnrolling: fix crash when a loop counter is captured in a lambda by reference

[Valeriy Savchenko via Phabricator via cfe-commits]

vsavchenko added inline comments.


================
Comment at: clang/lib/StaticAnalyzer/Core/LoopUnrolling.cpp:185-186
+      return FD->getType()->isReferenceType();
+    } else {
+      assert(false && "Unknown captured variable");
+    }
----------------
AbbasSabra wrote:
> vsavchenko wrote:
> > But actually, it's a bit different with this one. I don't know exact details and rules how clang sema fills in the class for lambda.
> > According to [[ https://en.cppreference.com/w/cpp/language/lambda | cppreference ]]:
> > > For the entities that are captured by reference (with the default capture [&] or when using the character &, e.g. [&a, &b, &c]), it is unspecified if additional data members are declared in the closure type
> > 
> > It can be pretty much specified in clang, that's true, but it looks like in `DeadStoreChecker` we have a very similar situation and we do not assume that captured variable always have a corresponding field.
> Yes, I based this on the fact that DeadStoreChecker considers it possible, but I have never faced a case where it does not have a corresponding field.
It still would be good to have some proof that it is indeed like this or simply fallback into returning true (which you already do when in doubt).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D102273/new/

https://reviews.llvm.org/D102273