Fix 868

[mshafer-NI]

The current string based search for E731 (assignment of lambda to variable) relies on a text search. This breaks down if the lambda has a comment stored on it forcing a multi-line assignment in the form of:

f = (  # some long comment that forces this to be on a new line
    lambda x: 731 * x
)

However, existing tests indicate that

f = (lambda o: o.lower()) if isinstance('a', str) else (lambda o: o)
# and
f = (
    lambda o: o.lower(),
    lambda o: o.upper(),
)

Should be considered acceptable, so merely stripping off paranthesis before evaluating would not work.

Implementation

By loading the "line" (in this example, logical_line comes in as f = (lambda x: 731 * x)) using the AST parser, we are able to check if there is an assignment of a Lambda to a Name.

This allows all existing tests to pass while also resolving the new tests for E731.

Fixes #868

[asottile]

unfortunately one of the design principles of pycodestyle is that it does not use the ast to do its linting (this allows it to be ~somewhat cross-version compatible via the tokenizer)

so while this would work -- it can't be merged :(

[mshafer-NI]

May I suggest documenting that in CONTRIBUTING.rst?

[mshafer-NI]

Would you consider a PR that attempts ast and if there's any issue, falls back to the current method?

[asottile]

no I would not, that's why I closed this

[asottile]

and it's already in the docs:

Using the ast module defeats that purpose