Implement a cop that checks for something.length == 0

[bbatsov]

We should suggest replacing such checks with empty?.

This cop should also handle != and the common alias size.

[jawshooah]

Is coll.any? equivalent to !coll.empty? in terms of performance? If so, the check should probably suggest it as a replacement for coll.length != 0.

[bbatsov]

It's not the same in terms of semantics.

[jawshooah]

Really? What's the difference? (I ask because I've been using #any? as an equivalent to !#empty? and really hope I haven't been introducing subtle bugs in my code)

[jawshooah]

I know #any? takes an optional predicate block, but without the block it just returns true on the first element it sees (I would hope). Found the difference. Excuse my ignorance!

[rrosenblum]

Interesting. I am probably guilty of using any? as a replacement to !#empty?. Glad to have learned something here.

[alexdowad]

Another difference.

5:50 ~ % pry
[1] pry(main)> "".any?
NoMethodError: undefined method `any?' for "":String
from (pry):1:in `__pry__'
[2] pry(main)> "".empty?
=> true

[Drenmi]

The same check should probably also be done for the logically equivalent:

0 == something.length

This might seem weird to some, but the habit is carried over by some from languages where this has benefit.

[bbatsov]

Yeah, we should check this this as well.

[Drenmi]

I implemented this cop as an exercise, and soon realized that something can be whatever the user wants, as long as it implements length or size. But that something may not necessarily also implement empty?, as is the case, for example, with Parser::Source::Range. (Causing two offenses in the Rubocop source.)

What is the official strategy for dealing with such user implemented "duds"? I can see three possible paths, none of which seem ideal to me:

1. Ignore it and let the user mark it as a false positive. (This could make a potential --auto-correct rather dangerous.)
2. Check if the receiver is a known implementor of empty?. (This could get rather verbose.)
3. Do something funky and check if the receiver respond_to?(:empty?). (This isn't really static code analysis any more, is it?)

Please enlighten me. 😃

[bbatsov]

Ignore it and let the user mark it as a false positive. (This could make a potential --auto-correct rather dangerous.)

This is the answer. :-)

[Drenmi]

Added in #2683. I guess this issue can be closed now.