Constant propagation class: instance variables are not covered

[simisimon]

Instance variables, as you can see in the code snippet below, are currently not covered by the constant propagation class. That is, when applying Scalpel, I cannot identify the values of self.x. However, I would expect Scalpel to be able to do this.

class Test:
    def __init__(self):
        self.x = 2

    def change_x(self):
        self.x = 5

There are two parts in the source code that need to be modified. First, Scalpel needs to cover the ast.Attribute object.

Scalpel/scalpel/SSA/const.py

Lines 195 to 197 in 2ba6cab

	elif isinstance(targets[0], ast.Attribute):
	#TODO: resolve attributes
	pass

Here, we just need to add two lines:

left_name = ast.unparse(stmt.targets[0])
const_dict[left_name] = stmt.value  

Second, we need to modify the last code block of the method get_stmt_idents_ctx. As the name and attr attribute of ast.Attribute objects are usually connected by a dot, we need to remove "." in r["name"] in the first if condition (Line 316).

Scalpel/scalpel/SSA/const.py

Lines 314 to 324 in 2ba6cab

	ident_info = get_vars(visit_node)
	for r in ident_info:
	if r['name'] is None or "." in r['name'] or "_hidden_" in r['name']:
	continue
	if r['usage'] == 'store':
	stored_idents.append(r['name'])
	else:
	loaded_idents.append(r['name'])
	if r['usage'] == 'del':
	del_set.append(r['name'])
	return stored_idents, loaded_idents, []

With these two changes, Scalpel is able to identify all the values for self.x.

[Jarvx]

Thanks for raising this, this was the todo comment for ourselves. In fact, the attribute name is now supported in terms of variable extraction . In the initial stage , I removed the dotted name because we will provide a better way to model attribute object. At the moment, we will support this one using its simple string name.