heredoc as hash value raises Style/RedundantParentheses

[SaltwaterC]

I'll let the command line do the talking:

$ rubocop -V # supplied by ChefDK 0.11.0
0.37.2 (using Parser 2.3.0.5, running on ruby 2.1.6 x86_64-darwin12.0)

$ ruby code.rb
{:foo=>"heredoc\nmore heredoc\n", :bar=>"baz"}

$ cat code.rb
foo = {
  foo: (
  <<-EOF.gsub(/^\s*\|/, '')
    |heredoc
    |more heredoc
  EOF
  ),
  bar: 'baz'
}

p foo

$ rubocop
Inspecting 1 file
C

Offenses:

code.rb:2:8: C: Don't use parentheses around a method call.
  foo: (
       ^

1 file inspected, 1 offense detected

$ rubocop -a
Inspecting 1 file
E

Offenses:

code.rb:2:8: C: [Corrected] Don't use parentheses around a method call.
  foo: (
       ^
code.rb:7:3: E: unexpected token tCOMMA
(Using Ruby 2.0 parser; configure using TargetRubyVersion parameter, under AllCops)
  ,
  ^

1 file inspected, 2 offenses detected, 1 offense corrected

$ ruby code.rb
code.rb:7: syntax error, unexpected ',', expecting '}'
code.rb:9: syntax error, unexpected '}', expecting end-of-input

$ cat code.rb
foo = {
  foo:
  <<-EOF.gsub(/^\s*\|/, '')
    |heredoc
    |more heredoc
  EOF
  ,
  bar: 'baz'
}

p foo

Don't know whether this issue is related to #2654 and/or #2845. This is the closest I got before raising a new issue.

[alexdowad]

This doesn't specifically have anything to do with heredocs.

$ cat test.rb
hash = {
  foo: 1
  ,
  bar: 2
}
$ ruby-parse test.rb
test4.rb:3:3: error: unexpected token tCOMMA
test4.rb:3:   ,
test4.rb:3:   ^

[alexdowad]

From parser's Ruby grammar:

          assocs: assoc
                    {
                      result = [ val[0] ]
                    }
                | assocs tCOMMA assoc
                    {
                      result = val[0] << val[2]
                    }

[alexdowad]

I consider this to be a bug in Ruby's parser. Wonder if they would accept a patch...

[alexdowad]

I guess I can't say that it has nothing to do with heredocs: the issue is that a heredoc must always be terminated with a newline (you can't put anything else on the terminating line), so a bare heredoc can't be used anywhere where the parser can't swallow a newline. If you surround it with parens, it's OK.

[alexdowad]

Maybe we shouldn't consider parens redundant if the right paren comes directly after the terminating line of a heredoc?

That would probably require us to walk the entire subtree looking for a heredoc, which is a bit of a pain.

[alexdowad]

OR, rather than looking at the AST, we look at the tokens. Find the token for the rparen, see if it is preceded by tNL and tSTRING_END. If so, back up looking for the corresponding tSTRING_BEG and see if it begins a heredoc...

[SaltwaterC]

Maybe we shouldn't consider parens redundant if the right paren comes directly after the terminating line of a heredoc?

^ this

Also, this issue appeared between:

$ rubocop -V
0.34.2 (using Parser 2.2.3.0, running on ruby 2.1.6 x86_64-linux)

and the specified version in OP.

[alexdowad]

That's probably when the RedundantParentheses cop was added.