[PATCH] Enhance CFG to model C++ new more precisely

[Jordan Rose]

  This looks like a good start, and I'm okay with it being committed separately. I'll e-mail you about my idea for evaluating the allocator separately.

  I did find one issue; once you get the chance to fix that please add some tests with placement arguments as well.


================
Comment at: lib/Analysis/CFG.cpp:3140-3141
@@ +3139,4 @@
+  appendStmt(Block, NE);
+  CFGBlock *allocatedBlock = VisitChildren(NE);
+  appendNewAllocator(allocatedBlock, NE);
+  return allocatedBlock;
----------------
This isn't quite correct: the placement arguments have to be evaluated before the allocator call, but the constructor has to come afterwards.


http://llvm-reviews.chandlerc.com/D2423