[PATCH] [Clang SA] Fix double delete warning reported at wrong location.

[Jordan Rose]

================
Comment at: include/clang/StaticAnalyzer/Core/PathSensitive/CallEvent.h:127-129
@@ -126,4 +126,5 @@
 protected:
+  const Stmt *TStmt;
   // This is user data for subclasses.
   const void *Data;
 
----------------
Karthik Bhat wrote:
> Jordan Rose wrote:
> > I really don't want to increase the size of CallEvent. How about using `Origin` to either represent a CXXDestructorDecl or a CXXDeleteExpr?
> Hi Jordan,
> Origin already contains CXXDestructorDecl hence CallEvent->getDecl() will return CXXDestructorDecl but we will not be able to get the CXXDeleteExpr which triggered it. 
> CXXDeleteExpr is required while reporting warning in MallocChecker as we are checking if argument to delete is already deleted or not.
> If we try to save CXXDeleteExpr in Origin instead then we are will face problem when we try to inline code inside destructor call as CallEvent->getDecl() will be null in this case.
> 
> I assume Origin can have only 1 info at a time i.e. either Decl* or Expr* but in this case we might need both. Hence added a new Stmt to represent the triggered stmt.
> 
> I'm still trying to get a hold of clang SA code so please let me know if i'm wrong somewere or if there is some way to extract CXXDeleteExpr from CXXDestructorDecl. I tried to have a look but couldnt find one.
Since the CXXDeleteExpr already contains a reference to the CXXDestructorDecl, I think we can get this to work. (That's how CFGDeleteDtor works, right?)

It might make sense to make a completely new kind of CallEvent as well, rather than switching on whether the origin is a CXXDeleteExpr or a CXXDestructorDecl. The point of CallEvent is to abstract over the various ways a call can be written. I haven't thought this through, though—it might be more trouble than it's worth.


http://llvm-reviews.chandlerc.com/D2441