Problem

Call Graph building has the problem of call site resolution. For example:

def f1():
  f2()

def f2():
  pass

main():
  f1()

Here the call graph is simple to construct and looks like main -> f1 -> f2. But real world cases are more complex. Consider this:

import base64

class Encoding:
    def __init__(self):
        pass

    def apply(self, msg, func):
        return func(msg)

encoder = Encoding()

encoded = encoder.apply("Hello, World!".encode('utf-8') , base64.b64encode)
print(encoded)

decoded = encoder.apply(encoded, base64.b64decode)
print(decoded)

Ability to build a call graph here requires tracking the type of each variable during any function call invocation so that we can resolve the function target (call site). For example func in apply method above. Also OO programs has the added complexity of class and class hierarchy that need to be tracked. A good description of the problem is given in the following paper:

https://arxiv.org/abs/2103.00587

Requirements

We need the ability to build a call graph for Python language within a single source file including handling cases that require inter procedural analysis such as the example given above.

Limitations

• Limit the call graph to a single source file only accepting the limitations of inter procedural analysis across source files