bug: Incorrect language selector for code blocks

[sergey-tikhonenko]

Neovim version (nvim -v)

NVIM v0.10.0 Build type: Release LuaJIT 2.1.1713484068

Operating system

Linux, Ubuntu 22.04.4 LTS

Terminal emulator / GUI

kitty 0.21.2

Describe the bug

The plugin doesn't support all forms of the language specification for code blocks that are supported by the Markdown Treesitter parser.

The language selector here should be (info_string (language)) instead of (info_string).

The Treesitter markdown parser explicitly identifies the separate language node in the AST, leaving the surroundings unspecified.

See also the snippet below with several options for specifying the code block language:

The ordinary code block with the short form language.

```python
print("Hello")

Code blocks that use braces around the language name (e.g. ```{python}) are executable, and will be run by Quarto during render. Treesitter supports this form.

```{python}
1 + 1

This is code block with the short form language and separate attributes. Treesitter supports this form.

```python  {.numberLines filename="hello.py"}
print("Hello")

This code block with the Pandoc form language and attributes. At the moment Treesitter doesn't recognise the language for this form.

```{.python  .numberLines filename="hello.py"}
print("Hello")

Expected behavior

The plugin for language detection should use the (info_string (language)) selector instead of (info_string).

Additional information

The functionality was added during the "Feature: add language icon for code block" (6eef62c) implementation.

[MeanderingProgrammer]

That makes sense, was not even aware of all of these options for code blocks, neat!

Have pushed a fix here: https://github.com/MeanderingProgrammer/markdown.nvim/commit/90072fdbc28042add4cd08bef282df032bf6ac42.

LMK how it works!

[sergey-tikhonenko]

Works as expected. Thanks.

Ideally it might display the filename next to the language, but this requires the Markdown Treesitter parser support for the attribute map.

[MeanderingProgrammer]

Do have any special conceal logic removing the attribute map?

For me the attribute map stays in the 3rd example, like in this demo: https://github.com/MeanderingProgrammer/markdown.nvim/blob/main/demo/heading_code.gif

[sergey-tikhonenko]

Thanks for the question.

I thought it was your plugin that conceals attributes, but there were my custom Treesitter rules. This was an attempt to conceal the curly braces around the language while there are no other attributes.

; hide curly braces for quatro files
 (fenced_code_block
   (info_string) @label
   (#set! conceal ""))

But this doesn't work if any attributes are specified.

Not sure if conditions are supported in the Treesitter rules. Need to checked, if not, then only a custom handler can help in the case.

In 4th case I was wrong - Treesitter supports this form too. This highlighting doesn't work.

Thus to correctly determine the snippet language, you need to take into account that might be the dot in the beginning.

[sergey-tikhonenko]

It seems the plugin hides the language name if its symbol is not found.

Here is the mermaid code block. Both the Treesitter parser and highlighting work correctly, but the language name is disappeared.

flowchart LR
  A[Hard edge] --> B(Round edge)
  B --> C{Decision}
  C --> D[Result one]
  C --> E[Result two]

[sergey-tikhonenko]

LazyVim replaces nvim-tree/nvim-web-devicons with echasnovski/mini.icons. This give better integrations (more languages). There are also defaults for unknown languages, so the language name does not disappear in the above mentioned cases.

[MeanderingProgrammer]

Thanks for the heads up!

Added support for mini.icons here: https://github.com/MeanderingProgrammer/markdown.nvim/commit/353e4459938dd58873772e27a45c1d92bc83bafc.

Gets used instead of nvim-web-devicons when its available.

[sergey-tikhonenko]

Just for a feedback.

Finally, I've got a configuration that satisfy my needs for now. To hide the curly braces, I have to write a small custom handler:

custom_handlers = {
  markdown = {
    -- hides curly braces around the language line for code blocks
    render = function(namespace, root, buf)
      local query = vim.treesitter.query.parse("markdown", "(info_string) @_info")
      for _, node in query:iter_captures(root, buf) do
        local value = vim.treesitter.get_node_text(node, buf)
        local start_row, start_col, end_row, end_col = node:range()
        local start_match, _, start_meta, end_meta = string.find(value, "^%s*{%s*().*()%s*}%s*$")
        if start_match then
          vim.api.nvim_buf_set_extmark(buf, namespace, start_row, start_col, {
            end_row = start_row,
            end_col = start_col + start_meta - 1,
            conceal = "",
          })
          vim.api.nvim_buf_set_extmark(buf, namespace, end_row, start_col + end_meta - 1, {
            end_row = end_row,
            end_col = end_col,
            conceal = "",
          })
        end
      end
    end,
    extends = true,
  },
},

Might be this will be useful for someone.

[sergey-tikhonenko]

In 90072fd you use virt_text_pos = 'inline' relying on default highlighting NeoVim configuration for markdown. Where the language node is concealed.

I've faced with an environment where user has his own ~/.config/nvim/queries/markdown/highlights.scm due to conceal directives can't be reverted. And, among other things, concealment of the the language node is not enabled.

Enabling the plugin in such environment results in duplication in the language name.

I don't know how it might be fixed without change users highlights.scm. Possible some check for region concealment. Don't have enough experience in Neovim machinery.

[MeanderingProgrammer]

Yeah, that is definitely a growing trend of issues.

I'm going to track that here: https://github.com/MeanderingProgrammer/markdown.nvim/issues/50, see if I can think of a reasonable approach to it.