Issue 341

[cdonovick]

Summary

resolves #341

Adds _PrecedenceNode as a helper class to automatically parenthesize child nodes when necessary

Test Plan

Adds basic unit tests for implicit parens and extends fuzz testing.

Notes

I had originally tried to allow MaybeSentinel in {l,r}par attrs but this proved difficulty to get right. It would require numerous changes to the parser to ensure proper construction of _BaseParenthesizableNode as they are often explicitly constructed with empty lpar and rpar attributes.

[notEvil]

Isn't this pretty important? Afaik any transformer might produce logically different code. Is there a simple "workaround" I don't see?

thank you for libcst!

[cdonovick]

@notEvil I have a quick a dirty work around in the linked issue. #341 (comment)

You could probably do something more clever by incorporating the logic I use here in the transformer.

[notEvil]

@notEvil I have a quick a dirty work around in the linked issue. #341 (comment)

Thanks, I've seen it but don't want to introduce parentheses.

You could probably do something more clever by incorporating the logic I use here in the transformer.

I just did, except for Power which probably takes me another hour to figure out (because transformer act at child level)*. Still think this PR should get more attention.

* eventually, after I recognized the relevance of wrap_ties

[kiri11]

Looks amazing! Prevents tricky issues.

[zsol]

I'm happy to review this if someone's willing to resurrect and rebase

[notEvil]

I've just published my recent work on parentheses and precedence at https://gitlab.com/notEvil/normcst. The tests generate ~7.5k combinations of expressions/statements/match patterns and ensure that normcst.ParenthesisTransformer produces valid and minimal results. I will use it in my own transformations for a while before pushing it to PyPI and would be happy about anyone who helps by giving it a try.