Fix binary operation inference crash by using cloned contextsbinop-inference-fix

Idea is that we shouldn't use the same context to infer left-hand side and
right-hand side of the binary operation because inferring lhs may leave
some metadata in context.path that is irrelevant (and can even break inference)
to inference of rhs.
So we make two clones of the original context and use them to infer left-hand
side and right-hand side independently. And then we use the original context to
infer the result of the operation applied to inferred values.

This patch fixes pylint's issue #646 and may be also related to other
similar astroid issues (#198, #199).

1 files changed, 7 insertions, 7 deletions

diff --git a/astroid/inference.py b/astroid/inference.py
index 010607c..16d5311 100644
--- a/astroid/inference.py
+++ b/astroid/inference.py
@@ -575,18 +575,18 @@ def _infer_binop(self, context):
     right = self.right
     op = self.op
 
-    for lhs in left.infer(context=context):
+    # we use two separate contexts for evaluating lhs and rhs because
+    # 1. evaluating lhs may leave some undesired entries in context.path
+    #    which may not let us infer right value of rhs
+    lhs_context = context.clone()
+    rhs_context = context.clone()
+
+    for lhs in left.infer(context=lhs_context):
         if lhs is util.YES:
             # Don't know how to process this.
             yield util.YES
             return
 
-        # TODO(cpopa): if we have A() * A(), trying to infer
-        # the rhs with the same context will result in an
-        # inferrence error, so we create another context for it.
-        # This is a bug which should be fixed in InferenceContext at some point.
-        rhs_context = context.clone()
-        rhs_context.path = set()
         for rhs in right.infer(context=rhs_context):
             if rhs is util.YES:
                 # Don't know how to process this.