Missing label (on missing beginning bracket/parens)

[schneems]

When we have a missing closing bracket i.e. [ but not ] dead_end detects it correct. Same with { and missing } etc. But when the first one is missing, dead_end doesn't know it and so it doesn't give us a label.

Here's an example:

$ cat scratch.rb
class Cat
  def to_json
      name: name,
      fur_color: fur_color,
      type: 'Cat',
      better_than_dog: false }.to_json
  end
end
⛄ 3.0.2 🚀  /private/tmp
$ dead_end ./scratch.rb

file: /private/tmp/scratch.rb
simplified:

      1  class Cat
      2    def to_json
    ❯ 3        name: name,
    ❯ 4        fur_color: fur_color,
    ❯ 5        type: 'Cat',
    ❯ 6        better_than_dog: false }.to_json
      7    end
      8  end

Expected

I expect a label/banner describing the issue as missing a {

Actual

There is no description at all.

[schneems]

Same for

$ cat scratch.rb
class Animals
  def ones_my_three_year_old_likes
    array =
      Dog.new,
      Lion.new,
      Tiger.new,
      Aligator.new,
    ]

    return array
  end
end
⛄ 3.0.2 🚀  /private/tmp
$ dead_end ./scratch.rb



file: /private/tmp/scratch.rb
simplified:

       1  class Animals
       2    def ones_my_three_year_old_likes
    ❯  3      array =
    ❯  4        Dog.new,
    ❯  5        Lion.new,
    ❯  6        Tiger.new,
    ❯  7        Aligator.new,
    ❯  8      ]
      10      return array
      11    end
      12  end

[schneems]

Sometimes ruby doesn't know there's an issue with the bracket and some times it does:

syntax error, unexpected ']', expecting `end'

file: /Users/rschneeman/Documents/projects/cutlass/spec/cutlass_spec.rb
simplified:

       4  module Cutlass
       5    RSpec.describe Cutlass do
       6      it "has a version number" do
    ❯  7        foo = 1,2,3,4]
       9      end
      10    end
      11  end

⛄️ 3.0.2 🚀 /Users/rschneeman/Documents/projects/dead_end (main)
$ ./exe/dead_end /Users/rschneeman/Documents/projects/cutlass/spec/cutlass_spec.rb
syntax error, unexpected ',', expecting `end'

file: /Users/rschneeman/Documents/projects/cutlass/spec/cutlass_spec.rb
simplified:

       4  module Cutlass
       5    RSpec.describe Cutlass do
       6      it "has a version number" do
    ❯  7        1,2,3,4]
       9      end
      10    end
      11  end

Unexpected ]

We are currently matching /unexpected .* expecting '(?<unmatched_symbol>.*)'/ which doesn't match to syntax error, unexpected ']', expecting end'` because the first "quote" in front of end is a backtick instead of a quote.

It's actually good, because we're not expecting an end in this case, we're expecting a [ somewhere else.

We could fix this via regex, pulling out the ], and then using a mapping with inverse i.e. ] => [ and then outputting that.

But that doesn't handle our other case.

Unexpected ,

In the case where it doesn't the error is that it has an unexpected comma, not that it's missing a bracket:

syntax error, unexpected ',', expecting `end'

The thing triggering it is an "unexpected ," which makes sense as the parser doesn't yet know that there's a ] in there that it's supposed to match to later on.

In the short term, it's fine to not have an explicit "here's what you're missing" for these cases. Here's a lex with and without the leading bracket:

irb(main):019:0> Ripper.lex("foo = [1,2,3]")
=>
[[[1, 0], :on_ident, "foo", CMDARG],
 [[1, 3], :on_sp, " ", CMDARG],
 [[1, 4], :on_op, "=", BEG],
 [[1, 5], :on_sp, " ", BEG],
 [[1, 6], :on_lbracket, "[", BEG|LABEL],
 [[1, 7], :on_int, "1", END],
 [[1, 8], :on_comma, ",", BEG|LABEL],
 [[1, 9], :on_int, "2", END],
 [[1, 10], :on_comma, ",", BEG|LABEL],
 [[1, 11], :on_int, "3", END],
 [[1, 12], :on_rbracket, "]", END]]
irb(main):020:0> Ripper.lex("foo = 1,2,3]")
=>
[[[1, 0], :on_ident, "foo", CMDARG],
 [[1, 3], :on_sp, " ", CMDARG],
 [[1, 4], :on_op, "=", BEG],
 [[1, 5], :on_sp, " ", BEG],
 [[1, 6], :on_int, "1", END],
 [[1, 7], :on_comma, ",", BEG|LABEL],
 [[1, 8], :on_int, "2", END],
 [[1, 9], :on_comma, ",", BEG|LABEL],
 [[1, 10], :on_int, "3", END],
 [[1, 11], :on_rbracket, "]", END]]

One outlandish possible method of determining the problem could be to lex the visible count :on_l* and :on_r* and see which one has a mismatch, then output something like:

Maybe missing: `[` ?

Seems like a lot of work just to annotate the error message. Of course, this library is a lot of work just to show people their own source code with some extra highlighting. I'm not sure how robust it would be in practice, but in theory, it seems good.

[schneems]

I think what I want to do to move forward is temporarily move away from the banner approach.

I think the current value in dead_end is highlighting. I think we can follow this pattern (pseudo code)

@invalid_blocks.each do |block|
  display_block(block)
  display_original_ruby_syntax_error(block)
end

That buys us the ability to remove this logic https://github.com/zombocom/dead_end/blob/e770a4b2fb1790174f9c6c522e621b65ba9a9627/lib/dead_end/internals.rb#L24 and not have to worry about displaying bad/confusing banners. While the ruby syntax error messages aren't great, at least they're what people are already used to.

I want to release that change, and then go back and re-write the banner logic to rely on lex input of each block rather than relying on regex-ing the parser errors which does not seem accurate or sustainable.