tree-sitter-vimdoc

This grammar implements the vimdoc spec, :help help-spec. This README describes how the grammar represents the spec (nodes), and its limitations.

Predictable results are the primary goal, so that output formats (e.g. HTML) are well-formed; the input (vimdoc) is secondary. The first step should always be to try to fix the input rather than insist on a grammar that handles vimdoc's endless quirks.

Overview

• block is the main top-level node, delimited by blank line(s) or any line starting with < (codeblock terminator).
  • Contains line and line_li nodes.
• line:
  • Contains atoms (words, tags, taglinks, …)
  • Contains headings (h1, h2, h3, column_heading).
• line_li ("listitem", :help help-list):
  • Lines starting with -/•/[0-9]. (not +/*) are listitems.
  • Use the prefix node to detect if the listitem is ordered (numbered) or unordered.
  • Consumes lines (incl. codeblocks) until blank line, < line, or next listitem.
    • Except: < followed by indented text (on the same or next line) is skipped, so the listitem continues after its codeblock:

      - foo: >
          code
      <
        still foo: >
          code
      <  still foo
      - bar
      

  • Contains a codeblock not preceded by a blank line.
  • Nesting is ignored: indented listitems are parsed as siblings. Consumers can check leading whitespace to decide nesting.
• codeblock (:help help-codeblock):
  • Contained by line or line_li, because ">" can start a codeblock at the end of any line.
  • Contains line nodes without word nodes: it's just the full raw text line including whitespace. This is somewhat dictated by its "preformatted" nature; parsing the contents would require loading a "child" language (injection). See #2.
  • The terminating < (and any following whitespace) is discarded (anonymous).
• h1, h2, h3, column_heading (:help help-heading)

Tables

A table is a block whose first child is a line containing column_heading with ≥2 cells, followed by line nodes (rows, separators). A cell boundary is a <Tab> between two nodes.

Example (Lua):

--- Gets the rows (header first) of `block` if it is a table, else nil.
--- Each row is a list of cells, each cell a list of nodes.
local function get_table(block, src)
  local hdr = block:named_child(0) and block:named_child(0):named_child(0)
  if not hdr or hdr:type() ~= 'column_heading' then
    return nil
  end
  --- Splits the children of `node` into cells, where the text between two children has a tab.
  local function cells(node)
    local row, prev_end = { {} }, nil
    for child in node:iter_children() do
      if child:named() then
        local _, _, start, _, _, end_ = child:range(true)
        if prev_end and src:sub(prev_end + 1, start):find('\t') then
          row[#row + 1] = {}
        end
        table.insert(row[#row], child)
        prev_end = end_
      end
    end
    return row
  end
  local rows = { cells(hdr:named_child(0)) } -- (column_heading (heading …) (delimiter))
  for i = 1, block:named_child_count() - 1 do
    local line = block:named_child(i)
    local text = vim.treesitter.get_node_text(line, src)
    local row = not text:match('^[%s%-=~]+$') and cells(line) -- Skip separator.
    if row and #row == 1 and #rows > 1 then
      vim.list_extend(rows[#rows][#rows[#rows]], row[1]) -- Wrapped line continues last cell.
    elseif row then
      rows[#rows + 1] = row
    end
  end
  return #rows[1] > 1 and #rows > 1 and rows or nil
end

-- This query finds potential "table" blocks; get_table() splits them into cells.
local query = vim.treesitter.query.parse('vimdoc', [[
  (block . (line (column_heading) @header (#lua-match? @header "%S\t"))) @table
]])
local root = vim.treesitter.get_string_parser(src, 'vimdoc'):parse()[1]:root()
for id, block in query:iter_captures(root, src) do
  if query.captures[id] == 'table' then
    local rows = get_table(block, src) -- nil if not a table.
  end
end

Known issues

• Input must end with newline/EOL (\n). Grammar does not support files without EOL.
• Input must end with a blank line. Though this doesn't seem to matter in practice.
• Any line starting with 1. (or other number) is treated as a listitem, even if the first line of its block is not a listitem. Example:

  Foo was 0, not
  1. Uh oh.
  

• The spec requires a space before the codeblock >, but the grammar also accepts a tab. Example: :help lcs-tab.
• url cannot contain a closing bracket ] anywhere in the URL. (Workaround: URL-encode the bracket.) This is a tradeoff so that markdown hyperlinks work:

  [https://example.com](https://example.com)
  

• column_heading children should be plaintext, but currently are parsed as $._atom.

TODO

• h4 ("tag heading") : a line containing only tags, or ending with a tag, is a "h4" heading.

Release

Steps to perform a release: https://tree-sitter.github.io/tree-sitter/creating-parsers/6-publishing.html

1. Update tree-sitter CLI.

   npm install tree-sitter-cli
   

2. Bump the version.

   tree-sitter version x.y.z
   

   • Choose patch/minor/major to indicate query compatibility:
     • patch for bugfixes (no changes to queries needed)
     • minor for added nodes (queries may need changes to use new nodes but will not error)
     • major for removed or renamed nodes (queries will error if not adapted), other breaking changes
3. Regenerate and test.

   tree-sitter generate && tree-sitter test
   

4. Commit the generated files

   git add .
   git commit -m 'release'
   

5. Push

   git push
   

6. Tag and release: https://github.com/neovim/tree-sitter-vimdoc/releases/new

Publis to crates.io

Crate: https://crates.io/crates/tree-sitter-vimdoc

cargo publish --dry-run
cargo login
cargo publish