Type feedback tool?

[Scott David Daniels]

Martin Vilcans wrote:
> Hi list,
> 
> I'm wondering if there's a tool that can analyze a Python program
> while it runs, and generate a database with the types of arguments and
> return values for each function. In a way it is like a profiler, that
> instead of measuring how often functions are called and how long time
> it takes, it records the type information. So afterwards, when I'm
> reading the code, I can go to the database to see what data type
> parameter "foo" of function "bar" typically has. It would help a lot
> with deciphering old code.
> 
> When I googled this, I learned that this is called "type feedback",
> and is used (?) to give type information to a compiler to help it
> generate fast code. My needs are much more humble. I just want a
> faster way to understand undocumented code with bad naming.
> 
It should be pretty easy to do this as a decorator, then (perhaps
automatically) sprinkle it around your source.

     def watch(function):
         def wrapped(*args, **kwargs):
             argv = map(type, args)
             argd = dict((key, type(value)) for key, value
                          in kwargs.iteritems())
             try:
                 result = function(*args, **kwargs)
             except Exception, exc:
                 record_exception(function, type(exc), argv, argd)
                 raise
             else:
                 record_result(function, type(result), argv, argd)
         return wrapped

then fill your normal code with:

    @watch
    def somefun(arg, defarg=1):
        ...
    ...


Finally record_* could write to a data structure (using bits like
function.__name__, and f.__module__, and possibly goodies from inspect).
Then you wrap your actual code start with something like:

try:
     main(...)
finally:
     <write data structure to DB>


--Scott David Daniels
Scott.Daniels at Acm.Org