Regression on LineEndConcatenation

[agrimm]

Recent changes in LineEndConcatenation means that

def foo
  'Hello ' +
  # OPTIMIZE: Turn this into a constant.
  'world'
end

is regarded as incorrect.

[bbatsov]

Ops, seems we didn't have a test case for that scenario. I'll have that fixed.

[bbatsov]

/cc @barunio

[bbatsov]

@jonas054 @yujinakayama I played a bit with this, but didn't manage to solve it in an easy manner (without gutting the cop). I'd appreciate it if you take a look - perhaps you'll find a solution I'm missing.

[jonas054]

How about something like

          comment_in_between = processed_source.comments.find do |c|
            c.loc.line > receiver.loc.line && c.loc.line < first_arg.loc.line
          end
          return if comment_in_between

towards the end of offending_node?. It's a linear search through all comments of a file, but placed late in the method that might be ok.

Seems obvious. Maybe too obvious. I might be missing something. 😄

[bbatsov]

@jonas054 Think I did something similar initially, but it messed up the reporting of the multiple offenses. As I played with this a few weeks back my memory's a bit fuzzy on the subject.

[yujinakayama]

Just a random thought, maybe this can be handled by inspecting tokens rather than AST? Since we are interested only in string literals, there's enough information in tokens. Though this requires gutting the cop.

[6] pry(main)> RuboCop::ProcessedSource.new(%('foo' +\n'bar')).tokens
=> [#<RuboCop::Token:0x007f9087818980
  @pos=#<Parser::Source::Range (string) 0...5>,
  @text="foo",
  @type=:tSTRING>,
 #<RuboCop::Token:0x007f90878188e0
  @pos=#<Parser::Source::Range (string) 6...7>,
  @text="+",
  @type=:tPLUS>,
 #<RuboCop::Token:0x007f9087818840
  @pos=#<Parser::Source::Range (string) 8...13>,
  @text="bar",
  @type=:tSTRING>]
[7] pry(main)> RuboCop::ProcessedSource.new(%('foo' \\\n'bar')).tokens
=> [#<RuboCop::Token:0x007f90824e6e60
  @pos=#<Parser::Source::Range (string) 0...5>,
  @text="foo",
  @type=:tSTRING>,
 #<RuboCop::Token:0x007f90824e6dc0
  @pos=#<Parser::Source::Range (string) 8...13>,
  @text="bar",
  @type=:tSTRING>]
[8] pry(main)> RuboCop::ProcessedSource.new(%('foo' +\n# some comment\n'bar')).tokens
=> [#<RuboCop::Token:0x007f9087663a18
  @pos=#<Parser::Source::Range (string) 0...5>,
  @text="foo",
  @type=:tSTRING>,
 #<RuboCop::Token:0x007f9087663978
  @pos=#<Parser::Source::Range (string) 6...7>,
  @text="+",
  @type=:tPLUS>,
 #<RuboCop::Token:0x007f90876638d8
  @pos=#<Parser::Source::Range (string) 8...22>,
  @text="# some comment",
  @type=:tCOMMENT>,
 #<RuboCop::Token:0x007f9087663838
  @pos=#<Parser::Source::Range (string) 23...28>,
  @text="bar",
  @type=:tSTRING>]

[bbatsov]

I like @yujinakayama's idea (it's something I considered as well) and I think it'd yield performance improvements over the existing solution.

[yujinakayama]

I tried to rework the cop with token information and it looked perfect, but it turned out there's a case that cannot be handled with tokens.

[bbatsov]

Can't we just do a check if the string is followed by * and % - I think they are the only problematic operators.