Scope of the project

[tristanlatr]

I believe that we should adopt clear guidelines about what kind of stuff we want to have support for. Clearly we need first class support for literals, especially list of strings, since IMO it's the whole point of this project: being able to statically resolve __all__ assignments.

But __all__ is a list, so it can be mutated with append, extend, etc, so I believe we need support for builtins as well.

So I'd propose the following goals and no-goals.

Goals:

• Inferring complex assignments + mutations of built-in types at the module level.
• Inferring complex assignments of built-in types at the scoped nodes level.
• Inferring basic type infos of expressions, potentially by using type annotation as extra informations.
• Keep 100% compatibility on the node usage, unfortunately we must introduce breaking changes in the signature of the main parse() method, but the nodes should stay compatible with the standard library one.

No-goals:

• Inferring calls or objects protocols in a generic manner, this is the job of type checkers.

[tristanlatr]

Here are things that astroid has great support for which we could add to astuce in the future, but not for the first version in my opinion.

For loops and generator expressions:

        for a in [1, 2, 3]:
            print (a)
        for b,c in [(1,2), (3,4)]:
            print (b)
            print (c)

        print ([(d,e) for e,d in ([1,2], [3,4])])

Dict unpack:

        base = {'data': 0}
        new = {**base, 'data': 1}
        new3 = {**base, **new}