[PATCH] [analyzer] [proposal] Fix for PR20653

[Jordan Rose]

I really don't want states tracking liveness information—for one, you never //remove// this information even after the symbol is marked dead. Moreover, this //still// wouldn't handle removing a binding, nor would it handle binding to an entire aggregate region (i.e. what happens for struct assignment). I think the most-semantically-correct thing to do would be to ask the Store to refetch the value of the region, but that's not too cheap. Maybe we just need a new API to ask "are there any bindings to this region or any of its super-regions?".

http://reviews.llvm.org/D5104