Backport:

Fix for SF bug 532646.  This is a little simpler than what Neal
suggested there, based upon a better analysis (__getattr__ is a red
herring).
[This might be a 2.1 bugfix candidate we well, if people care]

Lib/test/test_class.py

@@ -274,3 +274,17 @@ class C2:
 try: hash(C2())
 except TypeError: pass
 else: raise TestFailed, "hash(C2()) should raise an exception"
+
+
+# Test for SF bug 532646
+
+class A:
+    pass
+A.__call__ = A()
+a = A()
+try:
+    a() # This should not segfault
+except RuntimeError:
+    pass
+else:
+    raise TestFailed, "how could this not have overflowed the stack?"

Objects/classobject.c

@@ -1879,6 +1879,7 @@ instance_iternext(PyInstanceObject *self)
 static PyObject *
 instance_call(PyObject *func, PyObject *arg, PyObject *kw)
 {
+	PyThreadState *tstate = PyThreadState_GET();
 	PyObject *res, *call = PyObject_GetAttrString(func, "__call__");
 	if (call == NULL) {
 		PyInstanceObject *inst = (PyInstanceObject*) func;
@@ -1888,7 +1889,22 @@ instance_call(PyObject *func, PyObject *arg, PyObject *kw)
 			     PyString_AsString(inst->in_class->cl_name));
 		return NULL;
 	}
-	res = PyObject_Call(call, arg, kw);
+	/* We must check and increment the recursion depth here. Scenario:
+	       class A:
+	           pass
+	       A.__call__ = A() # that's right
+	       a = A() # ok
+	       a() # infinite recursion
+	   This bounces between instance_call() and PyObject_Call() without
+	   ever hitting eval_frame() (which has the main recursion check). */
+	if (tstate->recursion_depth++ > Py_GetRecursionLimit()) {
+		PyErr_SetString(PyExc_RuntimeError,
+				"maximum __call__ recursion depth exceeded");
+		res = NULL;
+	}
+	else
+		res = PyObject_Call(call, arg, kw);
+	tstate->recursion_depth--;
 	Py_DECREF(call);
 	return res;
 }