Help adding support for Nix lang needed

[pogopaule]

I try to add support for Nix.

I am not sure if treesj in its current form can handle some "special" Nix formattings and tree-sitter-nix specifics or if I just don't understand treesj good enough to express these things.

The easiest way to get what I mean is to check out my nix branch and run the tests. (See also this diff between main and feat/nix)

Problem 1

What I struggle with the most is that tree-sitter-nix parses an attribute set

{
  a = 1;
  b = 2;
  c = "c";
}

into

element: attrset_expression [20, 4] - [24, 5]
  binding_set [21, 6] - [23, 14]
    binding: binding [21, 6] - [21, 12]
      attrpath: attrpath [21, 6] - [21, 7]
        attr: identifier [21, 6] - [21, 7]
      expression: integer_expression [21, 10] - [21, 11]
    binding: binding [22, 6] - [22, 12]
      attrpath: attrpath [22, 6] - [22, 7]
        attr: identifier [22, 6] - [22, 7]
      expression: integer_expression [22, 10] - [22, 11]
    binding: binding [23, 6] - [23, 14]
      attrpath: attrpath [23, 6] - [23, 7]
        attr: identifier [23, 6] - [23, 7]
      expression: string_expression [23, 10] - [23, 13]
        string_fragment [23, 11] - [23, 12]

With my current knowledge of treesj I am only able to split an attribute set { a = 1; b = 2; c = "c"; } into either

{  a = 1;
  b = 2;
  c = "c"; }

with the preset binding_set = ...

or

{
  a = 1;  b = 2; c = "c";
}

with the preset attrset_expression.

I tried to work with recursive = true in various forms but did not succeed.

Problem 2

The other problem I struggle with is that Nix has a very unusual formatting when splitting "function arguments". Take a look at the placement of the commas in https://github.com/pogopaule/treesj/blob/446df8a2833bdda43817dbb78302ce480946225e/tests/sample/index.nix#L29-L41

Potential workarounds

In both cases I can make things work by using treesj and the Nix formatter in combination. So for example I split an attribute set into

{  a = 1;
  b = 2;
  c = "c"; }

and let the formatter do the rest of the work. Same applies for the strange comma placement for function arguments.

This would kind of work but I am not satisfied with it.

[Wansmer]

Hello!

Problem 1:

{
  attrset_expression = {
    target_nodes = { 'binding_set' }, -- because target node for format is not attrset_expression, but binding_set
  },
  binding_set = u.set_preset_for_dict({
    both = {
      non_bracket_node = true, -- It will created empty framing nodes for get new line after '{' and before '}'
      separator = ';',
      last_separator = true,
    },
  }),
}

Problem 2:

Hmm, I see what all children of formals should be with same indents, but at the moment all inner indents calculating with first indent + shiftwidth.

Maximum what can be settings with TreeSJ it:

# from
  func1 =
    { foo, bar, blub, ... }:
    { };
# to
  func1 =
    { foo
      , bar
      , blub
      , ...
    }:
    { };

via

{
  formals = u.set_preset_for_args({
    both = {
      omit = { 'formal', 'ellipses' }
    },
    split = {
      separator = '',
    },
    join = {
      separator = ',',
      space_in_brackets = true,
    },
  }),
}

But this form adding extra space before separator, when joining. Looks like bug, will fix it.

About let_expression:

Via

{
  let_expression = { target_nodes = { 'binding_set' } },
}

It can be transformed

  bla =
    let x=1; y=2; in x + y;

  blu =
    let
      x = 1;
      y = 2;
    in x + y;

At the moment, no way to move x + y at new line, unfortunately.

Thanks for sharing these questions. I'll take them into account in further development and improving TreeSJ.

[pogopaule]

Thank you very much for your quick and detailed answer! I think those solutions are good enough for the moment. They do 99% of what is important. The nix formatter will do the rest.

[Wansmer]

Btw, I fixed a problem with extra spaces (094a0df) and added option for inner indent (120ec44) – you can use it for formals.

[Wansmer]

Add preset for let_expression with #74