Possible to cache astroid AST between runs?

[rueberger]

I'm not sure if this at all technically feasible, but pylint is currently the slowest part of our test suite, takes almost a minute for a modest size project with parallelization enabled.

I presume that the most expensive part of running pylint is computing the AST, and am wondering if it is possible to reuse it.

I apologize if this is obviously infeasible, I have no idea how astroid works.

[PCManticore]

Hey, thanks for creating an issue!

Computing an AST is actually one of the most trivial part in pylint's run, it's not that significant. What instead is slow is the inference, that is the actual process of solving the types of values in your program when pylint is looking for its patterns.

In pylint 2.0 though we should be a little bit faster, although were not where we'd like to be just yet:

• This change should improve the performance when dealing with large functions Limit inference to a maximum of 100 results at a time astroid#586
• These two changes (Making Pylint faster astroid#497 and Making Pylint faster 2 astroid#519) should bring some improvements for tree walking

What would be useful in your case is to run a profiler against pylint, yappi or something else, and with those results we might check what exactly is affecting the performance for your code base.

[Pierre-Sassoulas]

Closing as duplicate of #1416