Is it possible to @append_space in the middle of a node

[TomMelt]

I am writing a formatter for Fortran. Some of the nodes e.g., end subroutine have a space in the middle. When I format them with topiary, it removes the space.

Is it possible to match part of a node and append a space after that. Something like the following:

(end_subroutine_statement "end" @append_space)

When I try the above it crashes with error:

thread 'main' panicked at fortitude/src/format.rs:31:62:
building topiary query: Query("Error parsing query file", Some(QueryError { row: 29, column: 26, offset: 573, message: "(end_subroutine_statement \"end\" @append_space)\n                          ^", kind: Structure }))
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[yannham]

It's hard to tell without the grammar at hand, but I suspect the answer is "yes", in that end_routine_statement most likely decomposes as keyword subnodes "end" and "subroutine" or something like that. Would you have a link to the Fortran tree-sitter grammar you're using?

[Xophmeister]

Definitely 💯 If Topiary is removing the space, then those nodes must be distinct in the grammar. As you're getting an error when you search for the anonymous node "end", it suggests that some other named node(s) must be involved.

Without seeing the grammar, you could try (end_subroutine_statement (_) @append_space) (presuming there's no nesting).

[TomMelt]

Thanks @yannham and @Xophmeister , the grammar that I am using can be found here: https://github.com/stadelmanma/tree-sitter-fortran/blob/master/grammar.js

@Xophmeister I tried using your suggestion but it didn't work. I am brand new to tree-sitter so I don't understand why. Is it due to the nesting you mentioned?

I think these lines are relevant from grammar.js:

   428     end_subroutine_statement: $ => blockStructureEnding($, 'subroutine'),

and

  2281 // This can be merged with whiteSpacedKeyword, keeping for now.
  2282 function blockStructureEnding ($, structType) {
  2283   const obj = prec.right(seq(
  2284     alias(choice(
  2285       seq(
  2286         caseInsensitive('end', false),
  2287         optional(caseInsensitive(structType, false))),
  2288       caseInsensitive('end' + structType, false)),
  2289     'end' + structType),
  2290     optional($._name),
  2291     $._end_of_statement
  2292   ))
  2293   return obj
  2294 }

A lot of the end statements i.e., program, subroutine, module etc. all use the same format. Is there a general way in tree-sitter query to make a general query for all of these types? Or do I have to do each one separately?

[TomMelt]

FWIW I also tried this, which doesn't work.

(end_subroutine_statement
  (_) @append_space
  .
  (_)
)

The tree looks like this for a simple subroutine:

subroutine parens_but_no_args()
  print *, "hello world"
end subroutine

(translation_unit [0, 0] - [3, 0]
  (subroutine [0, 0] - [3, 0]
    (subroutine_statement [0, 0] - [1, 0]
      name: (name [0, 11] - [0, 29]))
    (print_statement [1, 2] - [1, 24]
      (format_identifier [1, 8] - [1, 9])
      (output_item_list [1, 11] - [1, 24]
        (string_literal [1, 11] - [1, 24])))
    (end_subroutine_statement [2, 0] - [3, 0])))

[TomMelt]

I managed to solve it. I turned on "show anonymous nodes" in playground and it gives this:

So I wrote the following query, which does exactly what I want.

(end_subroutine_statement "endsubroutine" @append_space . "endsubroutine")

If you have any general suggestions for improvements to my approach, please let me know. Otherwise thanks for all your help so far.

[Xophmeister]

You could do (end_subroutine_statement _ @append_space . _); that is, using wildcard patterns instead of explicitly mentioning those anonymous nodes.