Disable method unpacking inside of module init function by default as it...

Disable method unpacking inside of module init function by default as it really just increases the code size to uselessly speed up one-time code.
See #2102.

CHANGES.rst

@@ -20,7 +20,8 @@ Features added
 
 * When compiling with gcc, the module init function is now tuned for small
   code size instead of whatever compile flags were provided externally.
-  (Github issue #2102)
+  Cython now also disables some code intensive optimisations in that function
+  to further reduce the code size.  (Github issue #2102)
 
 * C file includes are moved behind the module declarations if possible, to allow
   them to depend on module declarations themselves.

Cython/Compiler/Optimize.py

@@ -4584,7 +4584,7 @@ class ConstantFolding(Visitor.VisitorTransform, SkipDeclarations):
     visit_Node = Visitor.VisitorTransform.recurse_to_children
 
 
-class FinalOptimizePhase(Visitor.CythonTransform, Visitor.NodeRefCleanupMixin):
+class FinalOptimizePhase(Visitor.EnvTransform, Visitor.NodeRefCleanupMixin):
     """
     This visitor handles several commuting optimizations, and is run
     just before the C code generation phase.
@@ -4622,8 +4622,8 @@ class FinalOptimizePhase(Visitor.CythonTransform, Visitor.NodeRefCleanupMixin):
                     function.type = function.entry.type
                     PyTypeObjectPtr = PyrexTypes.CPtrType(cython_scope.lookup('PyTypeObject').type)
                     node.args[1] = ExprNodes.CastNode(node.args[1], PyTypeObjectPtr)
-        elif (self.current_directives.get("optimize.unpack_method_calls")
-                and node.is_temp and function.type.is_pyobject):
+        elif (node.is_temp and function.type.is_pyobject and self.current_directives.get(
+                "optimize.unpack_method_calls_in_pyinit" if self.current_env().is_module_scope else "optimize.unpack_method_calls")):
             # optimise simple Python methods calls
             if isinstance(node.arg_tuple, ExprNodes.TupleNode) and not (
                     node.arg_tuple.mult_factor or (node.arg_tuple.is_literal and node.arg_tuple.args)):

Cython/Compiler/Options.py

@@ -194,6 +194,7 @@ _directive_defaults = {
 # optimizations
     'optimize.inline_defnode_calls': True,
     'optimize.unpack_method_calls': True,   # increases code size when True
+    'optimize.unpack_method_calls_in_pyinit': False,   # uselessly increases code size when True
     'optimize.use_switch': True,
 
 # remove unreachable code