Feature request: Expose tag from dollar quoted strings

[dbaynard]

I use this grammar with neovim and I would like to take advantage of the language injection functionality. One elegant way to do this is to parse the language indicated in a dollar quoted string tag.

select $${"some":"json"}$$::jsonb;
select $json${"some":"json"}$json$::jsonb;

(It seems linguist, via https://github.com/textmate/sql.tmbundle, gets the highlighting here correct — I'm not sure how, though.)

At the moment, the whole text (including the tags) gets returned as a (literal). I would like to return an expression like (quoted_string tag: (literal) value: (literal)).

It looks like this requires a change to scanner.c for this, though, and it would also have knock on effects downstream (although these would be small and manageable).

Would such a change be welcome?

[DerekStride]

This seems like a good change to me, feel free to take this as guidance not the exact behaviour but I would expect a test like this (tag should probably be (identifier) and value can be literal):

================================================================================
dollar quoted string with tag
================================================================================

SELECT select $json${"some":"json"}$json$::jsonb;

--------------------------------------------------------------------------------

(program
  (statement
    (select
      (keyword_select)
      (select_expression
        (term
          value: (cast
            (literal
              (quoted_string tag: (identifier) value: (literal))
            (keyword_jsonb))))))

[dbaynard]

Ah, that preserves the currrent behaviour with any matching on literal so it is backwards compatible?

This is the current parse for that example.

(program
  (statement
    (select
      (keyword_select)
      (select_expression
        (term
          value: (cast
            (literal)
            (keyword_jsonb)))))))

Thanks for the quick response. I can't commit to an implementation timeline. That said, I would like to do it this week, while looking at my monochrome embedded json acts as a reminder.

The use of quoted_string (rather than dollar_quoted_string) would apply to unwrapping quotes on other quote forms — I don't propose doing that now. Postgresql refers to 'string constants' and 'dollar-quoted string constants so it could also just be string.

[DerekStride]

The use of quoted_string (rather than dollar_quoted_string) would apply to unwrapping quotes on other quote forms — I don't propose doing that now. Postgresql refers to 'string constants' and 'dollar-quoted string constants so it could also just be string.

Either of those make sense to me 👍