Generate rules from Python AST?

[void-main]

Thanks for building such a great lib on Python AST! However, writing asserts in rules is pretty tedious, I wonder if it's possible to parse a target python AST, and generate the rules automatically?

Also, another question is, now that the contract is based on a single ast.Node, is it possible to do a partial replacement with multiple statements (spans multiple ast.Nodes)?

For example, how do I write the rule to replace all matches of the following 2 statements:

    boxes[:, 0] = boxes_xywhr[:, 0] - half_w
    boxes[:, 1] = boxes_xywhr[:, 1] - half_h

Do I have to do it in a larger scope node (say ast.FunctionDef)? Or is it possible to get some UD-Chains based on the current ast.Node?

[MementoRC]

@void-main I am just a beginner using this package! The way I do these is by iterating the body of say a FunctionDef, find the match, clone and make the changes, then yield Replace(..., ...) - It seems to work. Note that I look for the minimal change, for example if I just want to change say 1 is your example, I make sure that I get to that Constant() and only change it, this way I think it minimizes the MaybeOverlapping

[isidentical]

However, writing asserts in rules is pretty tedious, I wonder if it's possible to parse a target python AST, and generate the rules automatically?

One example I can think of is @lmmx's refactory which uses Refactor as a backend and takes the input search form which can be later compiled down to refactor expression. I can't see why this won't be possible as well, as long as some one is interested in writing it down :)

Also, another question is, now that the contract is based on a single ast.Node, is it possible to do a partial replacement with multiple statements (spans multiple ast.Nodes)?

As @MementoRC pointed, indeed it is a use case which is supported but needs a bit more precision to avoid overlaps. This is called Chained or composite actions in our docs, which includes a few more examples. It is still an experimental feature, so would love to hear your feedback as well.