Python part of Ariadne in WALA?

[khatchad]

We may be getting to a point now where the Python parsing part of Ariadne (com.ibm.wala.cast.python) can join Java and Javascript in the cast module on WALA. The ML part of Ariadne should, I believe, remain in the original repo for now until it is more mature to handle modern versions of TensorFlow more fully.

Is there any desire for this? My thinking is that, in doing so, it will be become more accessible to WALA users. Currently, being in its own repository completely may make it a bit obscure. Moving it would also potentially make people aware that it has a lot of new features since it's inception. Specifically, perhaps people tried it before without much luck on some modern Python.

[msridhar]

I would be very open to moving the Python frontend to the main WALA repo. One concern that comes to mind is tests. Does the frontend itself have decent test coverage that is independent of the ML usage in Ariadne? I think we would want to have a good test suite that has assertions about correctness of call graph construction, etc.

[khatchad]

Yes, this was one of my concerns as well. There is quite a bit of testing outside of the ML usage for the original features, but perhaps we leaned on the ML testing to test the new features.

I can run a coverage tool on the frontend tests to get a better sense of this.

[khatchad]

One feature specifically that we learned quite a bit on the ML infrastructure was to test the functors (callable objects) as this is heavily used in ML code. We might have to write some additional tests outside of the ML suite for features like these.