Symbol literals with interpolation are not allowed in hash patterns

[kddnewton]

Found this while importing your tests into YARP. On this line:

https://github.com/mbj/unparser/blob/8260882beae39cd3d88a29625d07dd31bbc66ece/test/corpus/literal/pattern.rb#L20

It's actually a syntax error.

case foo
in {"#{"a"}": 1}
end

Running ruby -c gives:

test.rb: test.rb:2: symbol literal with interpolation is not allowed (SyntaxError)
in {"#{"a"}": 1}
    ^~~~~~~~~

[mbj]

Interesting, whitequark/parser seems to accept it right now, producing this AST, its accepted in 2.7, 3.0, 3.1 and 3.2 here.

(case-match
  (send nil :foo)
  (in-pattern
    (hash-pattern
      (pair
        (dsym
          (begin
            (str "a")))
        (int 1))) nil nil) nil)

On MRI: Ruby 2.7 accepts it. But 3.0, 3.1 and 3.2 not.

Unparsers goal was to be able to unparser whitequark/parser so it does the job "right". This pattern tests are extracted from whitequark parser corpus. So you should have found it also via your test extraction.

I think we have to file this issue in whitequark/parser. Once they remove support unparser will also.

[mbj]

For full visibility, I've reported this in: https://github.com/whitequark/parser/issues/919. I do not want to filter it out unparser side for the moment. I'll do so with a fixing parser release.

[mbj]

@kddnewton if you are importing unparsers tests, you can just ignore this pattern, or would it be better for you I pushed a removal?

[kddnewton]

That's okay! I'm just ignored that one line for now. I appreciate the explanation!

[mbj]

Well I'd be happy to use YARP in the future as source AST for unparser. So thanks to you.

I hope you made dstrings less ambigious in your native AST, I'll check it out soon. Dynamic strings are the only construct that unparser fails on in edge cases right now.

[kddnewton]

YARP has two kinds of strings: StringNode and InterpolatedStringNode. InterpolatedStringNode has a list of child nodes which are interpolated expressions, string nodes, or interpolated variables. Does that answer your question? I'm not entirely sure.

[mbj]

@kddnewton yes helps me already, the story goes a bit deeper. I'll reply here later explaining it.