Search code examples
pythonpython-2.xcpython

How to map func_closure entries to variable names?


I have a lambda object that is created in this function:

def add_url_rule(self, rule, endpoint=None, view_func=None, **options):
    self.record(lambda s:
        s.add_url_rule(rule, endpoint, view_func, **options))

Using func_closure of the lambda function object I can access the closure scope of the lambda function:

(<cell at 0x3eb89f0: str object at 0x2fb4378>,
 <cell at 0x3eb8a28: function object at 0x3cb3a28>,
 <cell at 0x3eb8a60: str object at 0x3ebd090>,
 <cell at 0x3eb8b08: dict object at 0x3016ec0>)

A closer look (at the cell_contents attribute of each cell object) shows me this:

>>> [c.cell_contents for c in func.func_closure]
['categoryDisplay',
 <function indico.web.flask.util.RHCategoryDisplay>,
 '/<categId>/',
 {}]

That lambda function was created by this call:

add_url_rule('/<categId>/', 'categoryDisplay', rh_as_view(RHCategoryDisplay))

As you can see, the order does not match the argument order of the function or the order in which the arguments are used inside the lambda. While I could easily find out which element is what based on its type/content, I'd like to do it in a cleaner way.

So my question is: How can I associate it with their original variable names (or at least positions in case of function arguments)?


Solution

  • The closures are created by the LOAD_CLOSURE bytecode, in the same order as their bytecodes are ordered:

    >>> dis.dis(add_url_rule)
      2           0 LOAD_FAST                0 (self)
                  3 LOAD_ATTR                0 (record)
                  6 LOAD_CLOSURE             0 (endpoint)
                  9 LOAD_CLOSURE             1 (options)
                 12 LOAD_CLOSURE             2 (rule)
                 15 LOAD_CLOSURE             3 (view_func)
                 18 BUILD_TUPLE              4
                 21 LOAD_CONST               1 (<code object <lambda> at 0x10faec530, file "<stdin>", line 2>)
                 24 MAKE_CLOSURE             0
                 27 CALL_FUNCTION            1
                 30 POP_TOP             
                 31 LOAD_CONST               0 (None)
                 34 RETURN_VALUE        
    

    so the order is determined at compile time, by compiler_make_closure(); this function uses the func.func_code.co_freevars tuple as a guide, which lists the closures in the same order.

    func.func_code.co_freevars is set when creating a code object in makecode, and the tuple is generated from the keys of a python dictionary, so the order is otherwise arbitrary, as common for dictionaries. If you are curious, the dict is built in compiler_enter_scope(), using the dictbytype() utility function from all the free variables named in the compiler symbol table, itself a python dictionary.

    So, the order of the closures is indeed arbitrary (hash table ordered), and you'd use the func.func_code.co_freevars tuple (which can be seen as a record of the order the compiler processed the dictionary keys) to attach names to the closures:

    dict(zip(func.func_code.co_freevars, (c.cell_contents for c in func.func_closure)))