neovim/scripts/gen_help_html.lua

--- Converts Nvim :help files to HTML.  Validates |tag| links and document syntax (parser errors).
--
-- USAGE (For CI/local testing purposes): Simply `make lintdoc` or `scripts/lintdoc.lua`, which
-- basically does the following:
--   1. :helptags ALL
--   2. nvim -V1 -es +"lua require('scripts.gen_help_html').run_validate()" +q
--   3. nvim -V1 -es +"lua require('scripts.gen_help_html').test_gen()" +q
--
-- USAGE (GENERATE HTML):
--   1. `:helptags ALL` first; this script depends on vim.fn.taglist().
--   2. nvim -V1 -es --clean +"lua require('scripts.gen_help_html').gen('./runtime/doc', 'target/dir/')" +q
--      - Read the docstring at gen().
--   3. cd target/dir/ && jekyll serve --host 0.0.0.0
--   4. Visit http://localhost:4000/…/help.txt.html
--
-- USAGE (VALIDATE):
--   1. nvim -V1 -es +"lua require('scripts.gen_help_html').validate('./runtime/doc')" +q
--      - validate() is 10x faster than gen(), so it is used in CI.
--
-- SELF-TEST MODE:
--   1. nvim -V1 -es +"lua require('scripts.gen_help_html')._test()" +q
--
-- NOTES:
--   * This script is used by the automation repo: https://github.com/neovim/doc
--   * :helptags checks for duplicate tags, whereas this script checks _links_ (to tags).
--   * gen() and validate() are the primary (programmatic) entrypoints. validate() only exists
--     because gen() is too slow (~1 min) to run in per-commit CI.
--   * visit_node() is the core function used by gen() to traverse the document tree and produce HTML.
--   * visit_validate() is the core function used by validate().
--   * Files in `new_layout` will be generated with a "flow" layout instead of preformatted/fixed-width layout.

local tagmap = nil ---@type table<string, string>
local helpfiles = nil ---@type string[]
local invalid_links = {} ---@type table<string, any>
local invalid_urls = {} ---@type table<string, any>
local invalid_spelling = {} ---@type table<string, table<string, string>>
local spell_dict = {
  Neovim = 'Nvim',
  NeoVim = 'Nvim',
  neovim = 'Nvim',
  lua = 'Lua',
  VimL = 'Vimscript',
  vimL = 'Vimscript',
  viml = 'Vimscript',
  ['tree-sitter'] = 'treesitter',
  ['Tree-sitter'] = 'Treesitter',
}
--- specify the list of keywords to ignore (i.e. allow), or true to disable spell check completely.
--- @type table<string, true|string[]>
local spell_ignore_files = {
  ['backers.txt'] = true,
  ['news.txt'] = { 'tree-sitter' }, -- in news, may refer to the upstream "tree-sitter" library
  ['news-0.10.txt'] = { 'tree-sitter' },
}
local language = nil

local M = {}

-- These files are generated with "flow" layout (non fixed-width, wrapped text paragraphs).
-- All other files are "legacy" files which require fixed-width layout.
local new_layout = {
  ['api.txt'] = true,
  ['lsp.txt'] = true,
  ['channel.txt'] = true,
  ['deprecated.txt'] = true,
  ['develop.txt'] = true,
  ['dev_style.txt'] = true,
  ['dev_theme.txt'] = true,
  ['dev_tools.txt'] = true,
  ['dev_vimpatch.txt'] = true,
  ['editorconfig.txt'] = true,
  ['faq.txt'] = true,
  ['lua.txt'] = true,
  ['luaref.txt'] = true,
  ['news.txt'] = true,
  ['news-0.9.txt'] = true,
  ['news-0.10.txt'] = true,
  ['nvim.txt'] = true,
  ['provider.txt'] = true,
  ['tui.txt'] = true,
  ['ui.txt'] = true,
  ['vim_diff.txt'] = true,
}

-- Map of new:old pages, to redirect renamed pages.
local redirects = {
  ['tui'] = 'term',
  ['terminal'] = 'nvim_terminal_emulator',
}

-- TODO: These known invalid |links| require an update to the relevant docs.
local exclude_invalid = {
  ["'string'"] = 'eval.txt',
  Query = 'treesitter.txt',
  matchit = 'vim_diff.txt',
  ['set!'] = 'treesitter.txt',
}

-- False-positive "invalid URLs".
local exclude_invalid_urls = {
  ['http://'] = 'usr_23.txt',
  ['http://.'] = 'usr_23.txt',
  ['http://aspell.net/man-html/Affix-Compression.html'] = 'spell.txt',
  ['http://aspell.net/man-html/Phonetic-Code.html'] = 'spell.txt',
  ['http://canna.sourceforge.jp/'] = 'mbyte.txt',
  ['http://gnuada.sourceforge.net'] = 'ft_ada.txt',
  ['http://lua-users.org/wiki/StringLibraryTutorial'] = 'lua.txt',
  ['http://michael.toren.net/code/'] = 'pi_tar.txt',
  ['http://papp.plan9.de'] = 'syntax.txt',
  ['http://wiki.services.openoffice.org/wiki/Dictionaries'] = 'spell.txt',
  ['http://www.adapower.com'] = 'ft_ada.txt',
  ['http://www.jclark.com/'] = 'quickfix.txt',
  ['http://oldblog.antirez.com/post/redis-and-scripting.html'] = 'faq.txt',
}

-- Deprecated, brain-damaged files that I don't care about.
local ignore_errors = {
  ['pi_netrw.txt'] = true,
  ['backers.txt'] = true,
}

local function tofile(fname, text)
  local f = io.open(fname, 'w')
  if not f then
    error(('failed to write: %s'):format(f))
  else
    f:write(text)
    f:close()
  end
end

---@type fun(s: string): string
local function html_esc(s)
  return (s:gsub('&', '&amp;'):gsub('<', '&lt;'):gsub('>', '&gt;'))
end

local function url_encode(s)
  -- Credit: tpope / vim-unimpaired
  -- NOTE: these chars intentionally *not* escaped: ' ( )
  return vim.fn.substitute(
    vim.fn.iconv(s, 'latin1', 'utf-8'),
    [=[[^A-Za-z0-9()'_.~-]]=],
    [=[\="%".printf("%02X",char2nr(submatch(0)))]=],
    'g'
  )
end

local function expandtabs(s)
  return s:gsub('\t', (' '):rep(8)) --[[ @as string ]]
end

local function to_titlecase(s)
  local text = ''
  for w in vim.gsplit(s, '[ \t]+') do
    text = ('%s %s%s'):format(text, vim.fn.toupper(w:sub(1, 1)), w:sub(2))
  end
  return text
end

local function to_heading_tag(text)
  -- Prepend "_" to avoid conflicts with actual :help tags.
  return text and string.format('_%s', vim.fn.tolower((text:gsub('%s+', '-')))) or 'unknown'
end

local function basename_noext(f)
  return vim.fs.basename(f:gsub('%.txt', ''))
end

local function is_blank(s)
  return not not s:find([[^[\t ]*$]])
end

---@type fun(s: string, dir?:0|1|2): string
local function trim(s, dir)
  return vim.fn.trim(s, '\r\t\n ', dir or 0)
end

--- Removes common punctuation from URLs.
---
--- TODO: fix this in the parser instead... https://github.com/neovim/tree-sitter-vimdoc
---
--- @param url string
--- @return string, string (fixed_url, removed_chars) where `removed_chars` is in the order found in the input.
local function fix_url(url)
  local removed_chars = ''
  local fixed_url = url
  -- Remove up to one of each char from end of the URL, in this order.
  for _, c in ipairs({ '.', ')' }) do
    if fixed_url:sub(-1) == c then
      removed_chars = c .. removed_chars
      fixed_url = fixed_url:sub(1, -2)
    end
  end
  return fixed_url, removed_chars
end

--- Checks if a given line is a "noise" line that doesn't look good in HTML form.
local function is_noise(line, noise_lines)
  if
    -- First line is always noise.
    (noise_lines ~= nil and vim.tbl_count(noise_lines) == 0)
    or line:find('Type .*gO.* to see the table of contents')
    -- Title line of traditional :help pages.
    -- Example: "NVIM REFERENCE MANUAL    by ..."
    or line:find([[^%s*N?VIM[ \t]*REFERENCE[ \t]*MANUAL]])
    -- First line of traditional :help pages.
    -- Example: "*api.txt*    Nvim"
    or line:find('%s*%*?[a-zA-Z]+%.txt%*?%s+N?[vV]im%s*$')
    -- modeline
    -- Example: "vim:tw=78:ts=8:sw=4:sts=4:et:ft=help:norl:"
    or line:find('^%s*vim?%:.*ft=help')
    or line:find('^%s*vim?%:.*filetype=help')
    or line:find('[*>]local%-additions[*<]')
  then
    -- table.insert(stats.noise_lines, getbuflinestr(root, opt.buf, 0))
    table.insert(noise_lines or {}, line)
    return true
  end
  return false
end

--- Creates a github issue URL at neovim/tree-sitter-vimdoc with prefilled content.
--- @return string
local function get_bug_url_vimdoc(fname, to_fname, sample_text)
  local this_url = string.format('https://neovim.io/doc/user/%s', vim.fs.basename(to_fname))
  local bug_url = (
    'https://github.com/neovim/tree-sitter-vimdoc/issues/new?labels=bug&title=parse+error%3A+'
    .. vim.fs.basename(fname)
    .. '+&body=Found+%60tree-sitter-vimdoc%60+parse+error+at%3A+'
    .. this_url
    .. '%0D%0DContext%3A%0D%0D%60%60%60%0D'
    .. url_encode(sample_text)
    .. '%0D%60%60%60'
  )
  return bug_url
end

--- Creates a github issue URL at neovim/neovim with prefilled content.
--- @return string
local function get_bug_url_nvim(fname, to_fname, sample_text, token_name)
  local this_url = string.format('https://neovim.io/doc/user/%s', vim.fs.basename(to_fname))
  local bug_url = (
    'https://github.com/neovim/neovim/issues/new?labels=bug&title=user+docs+HTML%3A+'
    .. vim.fs.basename(fname)
    .. '+&body=%60gen_help_html.lua%60+problem+at%3A+'
    .. this_url
    .. '%0D'
    .. (token_name and '+unhandled+token%3A+%60' .. token_name .. '%60' or '')
    .. '%0DContext%3A%0D%0D%60%60%60%0D'
    .. url_encode(sample_text)
    .. '%0D%60%60%60'
  )
  return bug_url
end

--- Gets a "foo.html" name from a "foo.txt" helpfile name.
local function get_helppage(f)
  if not f then
    return nil
  end
  -- Special case: help.txt is the "main landing page" of :help files, not index.txt.
  if f == 'index.txt' then
    return 'vimindex.html'
  elseif f == 'help.txt' then
    return 'index.html'
  end

  return (f:gsub('%.txt$', '')) .. '.html'
end

--- Counts leading spaces (tab=8) to decide the indent size of multiline text.
---
--- Blank lines (empty or whitespace-only) are ignored.
local function get_indent(s)
  local min_indent = nil
  for line in vim.gsplit(s, '\n') do
    if line and not is_blank(line) then
      local ws = expandtabs(line:match('^%s+') or '')
      min_indent = (not min_indent or ws:len() < min_indent) and ws:len() or min_indent
    end
  end
  return min_indent or 0
end

--- Removes the common indent level, after expanding tabs to 8 spaces.
local function trim_indent(s)
  local indent_size = get_indent(s)
  local trimmed = ''
  for line in vim.gsplit(s, '\n') do
    line = expandtabs(line)
    trimmed = ('%s%s\n'):format(trimmed, line:sub(indent_size + 1))
  end
  return trimmed:sub(1, -2)
end

--- Gets raw buffer text in the node's range (+/- an offset), as a newline-delimited string.
---@param node TSNode
---@param bufnr integer
---@param offset integer
local function getbuflinestr(node, bufnr, offset)
  local line1, _, line2, _ = node:range()
  line1 = line1 - offset
  line2 = line2 + offset
  local lines = vim.fn.getbufline(bufnr, line1 + 1, line2 + 1)
  return table.concat(lines, '\n')
end

--- Gets the whitespace just before `node` from the raw buffer text.
--- Needed for preformatted `old` lines.
---@param node TSNode
---@param bufnr integer
---@return string
local function getws(node, bufnr)
  local line1, c1, line2, _ = node:range()
  ---@type string
  local raw = vim.fn.getbufline(bufnr, line1 + 1, line2 + 1)[1]
  local text_before = raw:sub(1, c1)
  local leading_ws = text_before:match('%s+$') or ''
  return leading_ws
end

local function get_tagname(node, bufnr)
  local text = vim.treesitter.get_node_text(node, bufnr)
  local tag = (node:type() == 'optionlink' or node:parent():type() == 'optionlink')
      and ("'%s'"):format(text)
    or text
  local helpfile = vim.fs.basename(tagmap[tag]) or nil -- "api.txt"
  local helppage = get_helppage(helpfile) -- "api.html"
  return helppage, tag
end

--- Returns true if the given invalid tagname is a false positive.
local function ignore_invalid(s)
  return not not (
    exclude_invalid[s]
    -- Strings like |~/====| appear in various places and the parser thinks they are links, but they
    -- are just table borders.
    or s:find('===')
    or s:find('%-%-%-')
  )
end

local function ignore_parse_error(fname, s)
  if ignore_errors[vim.fs.basename(fname)] then
    return true
  end
  -- Ignore parse errors for unclosed tag.
  -- This is common in vimdocs and is treated as plaintext by :help.
  return s:find("^[`'|*]")
end

---@param node TSNode
local function has_ancestor(node, ancestor_name)
  local p = node ---@type TSNode?
  while p do
    p = p:parent()
    if not p or p:type() == 'help_file' then
      break
    elseif p:type() == ancestor_name then
      return true
    end
  end
  return false
end

--- Gets the first matching child node matching `name`.
---@param node TSNode
local function first(node, name)
  for c, _ in node:iter_children() do
    if c:named() and c:type() == name then
      return c
    end
  end
  return nil
end

local function validate_link(node, bufnr, fname)
  local helppage, tagname = get_tagname(node:child(1), bufnr)
  local ignored = false
  if not tagmap[tagname] then
    ignored = has_ancestor(node, 'column_heading') or node:has_error() or ignore_invalid(tagname)
    if not ignored then
      invalid_links[tagname] = vim.fs.basename(fname)
    end
  end
  return helppage, tagname, ignored
end

--- TODO: port the logic from scripts/check_urls.vim
local function validate_url(text, fname)
  local ignored = false
  if ignore_errors[vim.fs.basename(fname)] then
    ignored = true
  elseif text:find('http%:') and not exclude_invalid_urls[text] then
    invalid_urls[text] = vim.fs.basename(fname)
  end
  return ignored
end

--- Traverses the tree at `root` and checks that |tag| links point to valid helptags.
---@param root TSNode
---@param level integer
---@param lang_tree TSTree
---@param opt table
---@param stats table
local function visit_validate(root, level, lang_tree, opt, stats)
  level = level or 0
  local node_name = (root.named and root:named()) and root:type() or nil
  -- Parent kind (string).
  local parent = root:parent() and root:parent():type() or nil
  local toplevel = level < 1
  local function node_text(node)
    return vim.treesitter.get_node_text(node or root, opt.buf)
  end
  local text = trim(node_text())

  if root:child_count() > 0 then
    for node, _ in root:iter_children() do
      if node:named() then
        visit_validate(node, level + 1, lang_tree, opt, stats)
      end
    end
  end

  if node_name == 'ERROR' then
    if ignore_parse_error(opt.fname, text) then
      return
    end
    -- Store the raw text to give context to the error report.
    local sample_text = not toplevel and getbuflinestr(root, opt.buf, 0) or '[top level!]'
    -- Flatten the sample text to a single, truncated line.
    sample_text = vim.trim(sample_text):gsub('[\t\n]', ' '):sub(1, 80)
    table.insert(stats.parse_errors, sample_text)
  elseif
    (node_name == 'word' or node_name == 'uppercase_name')
    and (not vim.tbl_contains({ 'codespan', 'taglink', 'tag' }, parent))
  then
    local text_nopunct = vim.fn.trim(text, '.,', 0) -- Ignore some punctuation.
    local fname_basename = assert(vim.fs.basename(opt.fname))
    if spell_dict[text_nopunct] then
      local should_ignore = (
        spell_ignore_files[fname_basename] == true
        or vim.tbl_contains(
          (spell_ignore_files[fname_basename] or {}) --[[ @as string[] ]],
          text_nopunct
        )
      )
      if not should_ignore then
        invalid_spelling[text_nopunct] = invalid_spelling[text_nopunct] or {}
        invalid_spelling[text_nopunct][fname_basename] = node_text(root:parent())
      end
    end
  elseif node_name == 'url' then
    local fixed_url, _ = fix_url(trim(text))
    validate_url(fixed_url, opt.fname)
  elseif node_name == 'taglink' or node_name == 'optionlink' then
    local _, _, _ = validate_link(root, opt.buf, opt.fname)
  end
end

-- Fix tab alignment issues caused by concealed characters like |, `, * in tags
-- and code blocks.
---@param text string
---@param next_node_text string
local function fix_tab_after_conceal(text, next_node_text)
  -- Vim tabs take into account the two concealed characters even though they
  -- are invisible, so we need to add back in the two spaces if this is
  -- followed by a tab to make the tab alignment to match Vim's behavior.
  if string.sub(next_node_text, 1, 1) == '\t' then
    text = text .. '  '
  end
  return text
end

---@class (exact) nvim.gen_help_html.heading
---@field name string
---@field subheadings nvim.gen_help_html.heading[]
---@field tag string

-- Generates HTML from node `root` recursively.
---@param root TSNode
---@param level integer
---@param lang_tree TSTree
---@param headings nvim.gen_help_html.heading[]
---@param opt table
---@param stats table
local function visit_node(root, level, lang_tree, headings, opt, stats)
  level = level or 0

  local node_name = (root.named and root:named()) and root:type() or nil
  -- Previous sibling kind (string).
  local prev = root:prev_sibling()
      and (root:prev_sibling().named and root:prev_sibling():named())
      and root:prev_sibling():type()
    or nil
  -- Next sibling kind (string).
  local next_ = root:next_sibling()
      and (root:next_sibling().named and root:next_sibling():named())
      and root:next_sibling():type()
    or nil
  -- Parent kind (string).
  local parent = root:parent() and root:parent():type() or nil
  -- Gets leading whitespace of `node`.
  local function ws(node)
    node = node or root
    local ws_ = getws(node, opt.buf)
    -- XXX: first node of a (line) includes whitespace, even after
    -- https://github.com/neovim/tree-sitter-vimdoc/pull/31 ?
    if ws_ == '' then
      ws_ = vim.treesitter.get_node_text(node, opt.buf):match('^%s+') or ''
    end
    return ws_
  end
  local function node_text(node, ws_)
    node = node or root
    ws_ = (ws_ == nil or ws_ == true) and getws(node, opt.buf) or ''
    return string.format('%s%s', ws_, vim.treesitter.get_node_text(node, opt.buf))
  end

  local text = ''
  local trimmed ---@type string
  if root:named_child_count() == 0 or node_name == 'ERROR' then
    text = node_text()
    trimmed = html_esc(trim(text))
    text = html_esc(text)
  else
    -- Process children and join them with whitespace.
    for node, _ in root:iter_children() do
      if node:named() then
        local r = visit_node(node, level + 1, lang_tree, headings, opt, stats)
        text = string.format('%s%s', text, r)
      end
    end
    trimmed = trim(text)
  end

  if node_name == 'help_file' then -- root node
    return text
  elseif node_name == 'url' then
    local fixed_url, removed_chars = fix_url(trimmed)
    return ('%s<a href="%s">%s</a>%s'):format(ws(), fixed_url, fixed_url, removed_chars)
  elseif node_name == 'word' or node_name == 'uppercase_name' then
    return text
  elseif node_name == 'note' then
    return ('<b>%s</b>'):format(text)
  elseif node_name == 'h1' or node_name == 'h2' or node_name == 'h3' then
    if is_noise(text, stats.noise_lines) then
      return '' -- Discard common "noise" lines.
    end
    -- Remove tags from ToC text.
    local heading_node = first(root, 'heading')
    local hname = trim(node_text(heading_node):gsub('%*.*%*', ''))
    if not heading_node or hname == '' then
      return '' -- Spurious "===" or "---" in the help doc.
    end

    -- Generate an anchor id from the heading text.
    local tagname = to_heading_tag(hname)
    if node_name == 'h1' or #headings == 0 then
      ---@type nvim.gen_help_html.heading
      local heading = { name = hname, subheadings = {}, tag = tagname }
      headings[#headings + 1] = heading
    else
      table.insert(
        headings[#headings].subheadings,
        { name = hname, subheadings = {}, tag = tagname }
      )
    end
    local el = node_name == 'h1' and 'h2' or 'h3'
    return ('<%s id="%s" class="help-heading">%s</%s>\n'):format(el, tagname, trimmed, el)
  elseif node_name == 'heading' then
    return trimmed
  elseif node_name == 'column_heading' or node_name == 'column_name' then
    if root:has_error() then
      return text
    end
    return ('<div class="help-column_heading">%s</div>'):format(text)
  elseif node_name == 'block' then
    if is_blank(text) then
      return ''
    end
    if opt.old then
      -- XXX: Treat "old" docs as preformatted: they use indentation for layout.
      --      Trim trailing newlines to avoid too much whitespace between divs.
      return ('<div class="old-help-para">%s</div>\n'):format(trim(text, 2))
    end
    return string.format('<div class="help-para">\n%s\n</div>\n', text)
  elseif node_name == 'line' then
    if
      (parent ~= 'codeblock' or parent ~= 'code')
      and (is_blank(text) or is_noise(text, stats.noise_lines))
    then
      return '' -- Discard common "noise" lines.
    end
    -- XXX: Avoid newlines (too much whitespace) after block elements in old (preformatted) layout.
    local div = opt.old
      and root:child(0)
      and vim.list_contains({ 'column_heading', 'h1', 'h2', 'h3' }, root:child(0):type())
    return string.format('%s%s', div and trim(text) or text, div and '' or '\n')
  elseif node_name == 'line_li' then
    local sib = root:prev_sibling()
    local prev_li = sib and sib:type() == 'line_li'

    if not prev_li then
      opt.indent = 1
    else
      -- The previous listitem _sibling_ is _logically_ the _parent_ if it is indented less.
      local parent_indent = get_indent(node_text(sib))
      local this_indent = get_indent(node_text())
      if this_indent > parent_indent then
        opt.indent = opt.indent + 1
      elseif this_indent < parent_indent then
        opt.indent = math.max(1, opt.indent - 1)
      end
    end
    local margin = opt.indent == 1 and '' or ('margin-left: %drem;'):format((1.5 * opt.indent))

    return string.format('<div class="help-li" style="%s">%s</div>', margin, text)
  elseif node_name == 'taglink' or node_name == 'optionlink' then
    local helppage, tagname, ignored = validate_link(root, opt.buf, opt.fname)
    if ignored then
      return text
    end
    local s = ('%s<a href="%s#%s">%s</a>'):format(
      ws(),
      helppage,
      url_encode(tagname),
      html_esc(tagname)
    )
    if opt.old and node_name == 'taglink' then
      s = fix_tab_after_conceal(s, node_text(root:next_sibling()))
    end
    return s
  elseif vim.list_contains({ 'codespan', 'keycode' }, node_name) then
    if root:has_error() then
      return text
    end
    local s = ('%s<code>%s</code>'):format(ws(), trimmed)
    if opt.old and node_name == 'codespan' then
      s = fix_tab_after_conceal(s, node_text(root:next_sibling()))
    end
    return s
  elseif node_name == 'argument' then
    return ('%s<code>{%s}</code>'):format(ws(), text)
  elseif node_name == 'codeblock' then
    return text
  elseif node_name == 'language' then
    language = node_text(root)
    return ''
  elseif node_name == 'code' then -- Highlighted codeblock (child).
    if is_blank(text) then
      return ''
    end
    local code ---@type string
    if language then
      code = ('<pre><code class="language-%s">%s</code></pre>'):format(
        language,
        trim(trim_indent(text), 2)
      )
      language = nil
    else
      code = ('<pre>%s</pre>'):format(trim(trim_indent(text), 2))
    end
    return code
  elseif node_name == 'tag' then -- anchor, h4 pseudo-heading
    if root:has_error() then
      return text
    end
    local in_heading = vim.list_contains({ 'h1', 'h2', 'h3' }, parent)
    local h4 = not in_heading and not next_ and get_indent(node_text()) > 8 -- h4 pseudo-heading
    local cssclass = h4 and 'help-tag-right' or 'help-tag'
    local tagname = node_text(root:child(1), false)
    if vim.tbl_count(stats.first_tags) < 2 then
      -- Force the first 2 tags in the doc to be anchored at the main heading.
      table.insert(stats.first_tags, tagname)
      return ''
    end
    local el = 'span'
    local encoded_tagname = url_encode(tagname)
    local s = ('%s<%s id="%s" class="%s"><a href="#%s">%s</a></%s>'):format(
      ws(),
      el,
      encoded_tagname,
      cssclass,
      encoded_tagname,
      trimmed,
      el
    )
    if opt.old then
      s = fix_tab_after_conceal(s, node_text(root:next_sibling()))
    end

    if in_heading and prev ~= 'tag' then
      -- Start the <span> container for tags in a heading.
      -- This makes "justify-content:space-between" right-align the tags.
      --    <h2>foo bar<span>tag1 tag2</span></h2>
      return string.format('<span class="help-heading-tags">%s', s)
    elseif in_heading and next_ == nil then
      -- End the <span> container for tags in a heading.
      return string.format('%s</span>', s)
    end
    return s .. (h4 and '<br>' or '') -- HACK: <br> avoids h4 pseudo-heading mushing with text.
  elseif node_name == 'delimiter' or node_name == 'modeline' then
    return ''
  elseif node_name == 'ERROR' then
    if ignore_parse_error(opt.fname, trimmed) then
      return text
    end

    -- Store the raw text to give context to the bug report.
    local sample_text = level > 0 and getbuflinestr(root, opt.buf, 3) or '[top level!]'
    table.insert(stats.parse_errors, sample_text)
    return ('<a class="parse-error" target="_blank" title="Report bug... (parse error)" href="%s">%s</a>'):format(
      get_bug_url_vimdoc(opt.fname, opt.to_fname, sample_text),
      trimmed
    )
  else -- Unknown token.
    local sample_text = level > 0 and getbuflinestr(root, opt.buf, 3) or '[top level!]'
    return ('<a class="unknown-token" target="_blank" title="Report bug... (unhandled token "%s")" href="%s">%s</a>'):format(
      node_name,
      get_bug_url_nvim(opt.fname, opt.to_fname, sample_text, node_name),
      trimmed
    ),
      ('unknown-token:"%s"'):format(node_name)
  end
end

--- @param dir string e.g. '$VIMRUNTIME/doc'
--- @param include string[]|nil
--- @return string[]
local function get_helpfiles(dir, include)
  local rv = {}
  for f, type in vim.fs.dir(dir) do
    if
      vim.endswith(f, '.txt')
      and type == 'file'
      and (not include or vim.list_contains(include, f))
    then
      local fullpath = vim.fn.fnamemodify(('%s/%s'):format(dir, f), ':p')
      table.insert(rv, fullpath)
    end
  end
  return rv
end

--- Populates the helptags map.
local function get_helptags(help_dir)
  local m = {}
  -- Load a random help file to convince taglist() to do its job.
  vim.cmd(string.format('split %s/api.txt', help_dir))
  vim.cmd('lcd %:p:h')
  for _, item in ipairs(vim.fn.taglist('.*')) do
    if vim.endswith(item.filename, '.txt') then
      m[item.name] = item.filename
    end
  end
  vim.cmd('q!')
  return m
end

--- Use the vimdoc parser defined in the build, not whatever happens to be installed on the system.
local function ensure_runtimepath()
  if not vim.o.runtimepath:find('build/lib/nvim/') then
    vim.cmd [[set runtimepath^=./build/lib/nvim/]]
  end
end

--- Opens `fname` (or `text`, if given) in a buffer and gets a treesitter parser for the buffer contents.
---
--- @param fname string :help file to parse
--- @param text string? :help file contents
--- @param parser_path string? path to non-default vimdoc.so
--- @return vim.treesitter.LanguageTree, integer (lang_tree, bufnr)
local function parse_buf(fname, text, parser_path)
  local buf ---@type integer
  if text then
    vim.cmd('split new') -- Text contents.
    vim.api.nvim_put(vim.split(text, '\n'), '', false, false)
    vim.cmd('setfiletype help')
    -- vim.treesitter.language.add('vimdoc')
    buf = vim.api.nvim_get_current_buf()
  elseif type(fname) == 'string' then
    vim.cmd('split ' .. vim.fn.fnameescape(fname)) -- Filename.
    buf = vim.api.nvim_get_current_buf()
  else
    -- Left for debugging
    ---@diagnostic disable-next-line: no-unknown
    buf = fname
    vim.cmd('sbuffer ' .. tostring(fname)) -- Buffer number.
  end
  if parser_path then
    vim.treesitter.language.add('vimdoc', { path = parser_path })
  end
  local lang_tree = assert(vim.treesitter.get_parser(buf, nil, { error = false }))
  return lang_tree, buf
end

--- Validates one :help file `fname`:
---  - checks that |tag| links point to valid helptags.
---  - recursively counts parse errors ("ERROR" nodes)
---
--- @param fname string help file to validate
--- @param parser_path string? path to non-default vimdoc.so
--- @return { invalid_links: number, parse_errors: string[] }
local function validate_one(fname, parser_path)
  local stats = {
    parse_errors = {},
  }
  local lang_tree, buf = parse_buf(fname, nil, parser_path)
  for _, tree in ipairs(lang_tree:trees()) do
    visit_validate(tree:root(), 0, tree, { buf = buf, fname = fname }, stats)
  end
  lang_tree:destroy()
  vim.cmd.close()
  return stats
end

--- Generates HTML from one :help file `fname` and writes the result to `to_fname`.
---
--- @param fname string Source :help file.
--- @param text string|nil Source :help file contents, or nil to read `fname`.
--- @param to_fname string Destination .html file
--- @param old boolean Preformat paragraphs (for old :help files which are full of arbitrary whitespace)
--- @param parser_path string? path to non-default vimdoc.so
---
--- @return string html
--- @return table stats
local function gen_one(fname, text, to_fname, old, commit, parser_path)
  local stats = {
    noise_lines = {},
    parse_errors = {},
    first_tags = {}, -- Track the first few tags in doc.
  }
  local lang_tree, buf = parse_buf(fname, text, parser_path)
  ---@type nvim.gen_help_html.heading[]
  local headings = {} -- Headings (for ToC). 2-dimensional: h1 contains h2/h3.
  local title = to_titlecase(basename_noext(fname))

  local html = ([[
  <!DOCTYPE html>
  <html>
  <head>
    <meta charset="utf-8">
    <meta http-equiv="X-UA-Compatible" content="IE=edge">
    <meta name="viewport" content="width=device-width, initial-scale=1">
    <meta name="description" content="Neovim user documentation">

    <!-- algolia docsearch https://docsearch.algolia.com/docs/docsearch-v3/ -->
    <link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/@docsearch/css@3" />
    <link rel="preconnect" href="https://X185E15FPG-dsn.algolia.net" crossorigin />

    <link href="/css/bootstrap.min.css" rel="stylesheet">
    <link href="/css/main.css" rel="stylesheet">
    <link href="help.css" rel="stylesheet">
    <link href="/highlight/styles/neovim.min.css" rel="stylesheet">

    <script src="/highlight/highlight.min.js"></script>
    <script>hljs.highlightAll();</script>
    <title>%s - Neovim docs</title>
  </head>
  <body>
  ]]):format(title)

  local logo_svg = [[
    <svg xmlns="http://www.w3.org/2000/svg" role="img" width="173" height="50" viewBox="0 0 742 214" aria-label="Neovim">
      <title>Neovim</title>
      <defs>
        <linearGradient x1="50%" y1="0%" x2="50%" y2="100%" id="a">
          <stop stop-color="#16B0ED" stop-opacity=".8" offset="0%" />
          <stop stop-color="#0F59B2" stop-opacity=".837" offset="100%" />
        </linearGradient>
        <linearGradient x1="50%" y1="0%" x2="50%" y2="100%" id="b">
          <stop stop-color="#7DB643" offset="0%" />
          <stop stop-color="#367533" offset="100%" />
        </linearGradient>
        <linearGradient x1="50%" y1="0%" x2="50%" y2="100%" id="c">
          <stop stop-color="#88C649" stop-opacity=".8" offset="0%" />
          <stop stop-color="#439240" stop-opacity=".84" offset="100%" />
        </linearGradient>
      </defs>
      <g fill="none" fill-rule="evenodd">
        <path
          d="M.027 45.459L45.224-.173v212.171L.027 166.894V45.459z"
          fill="url(#a)"
          transform="translate(1 1)"
        />
        <path
          d="M129.337 45.89L175.152-.149l-.928 212.146-45.197-45.104.31-121.005z"
          fill="url(#b)"
          transform="matrix(-1 0 0 1 305 1)"
        />
        <path
          d="M45.194-.137L162.7 179.173l-32.882 32.881L12.25 33.141 45.194-.137z"
          fill="url(#c)"
          transform="translate(1 1)"
        />
        <path
          d="M46.234 84.032l-.063 7.063-36.28-53.563 3.36-3.422 32.983 49.922z"
          fill-opacity=".13"
          fill="#000"
        />
        <g fill="#444">
          <path
            d="M227 154V64.44h4.655c1.55 0 2.445.75 2.685 2.25l.806 13.502c4.058-5.16 8.786-9.316 14.188-12.466 5.4-3.15 11.413-4.726 18.037-4.726 4.893 0 9.205.781 12.935 2.34 3.729 1.561 6.817 3.811 9.264 6.751 2.448 2.942 4.297 6.48 5.55 10.621 1.253 4.14 1.88 8.821 1.88 14.042V154h-8.504V96.754c0-8.402-1.91-14.987-5.729-19.757-3.82-4.771-9.667-7.156-17.544-7.156-5.851 0-11.28 1.516-16.292 4.545-5.013 3.032-9.489 7.187-13.427 12.467V154H227zM350.624 63c5.066 0 9.755.868 14.069 2.605 4.312 1.738 8.052 4.268 11.219 7.592s5.638 7.412 7.419 12.264C385.11 90.313 386 95.883 386 102.17c0 1.318-.195 2.216-.588 2.696-.393.48-1.01.719-1.851.719h-64.966v1.70c0 6.708.784 12.609 2.353 17.7 1.567 5.09 3.8 9.357 6.695 12.802 2.895 3.445 6.393 6.034 10.495 7.771 4.1 1.738 8.686 2.606 13.752 2.606 4.524 0 8.446-.494 11.762-1.483 3.317-.988 6.108-2.097 8.37-3.324 2.261-1.227 4.056-2.336 5.383-3.324 1.326-.988 2.292-1.482 2.895-1.482.784 0 1.388.3 1.81.898l2.352 2.875c-1.448 1.797-3.362 3.475-5.745 5.031-2.383 1.558-5.038 2.891-7.962 3.998-2.926 1.109-6.062 1.991-9.41 2.65a52.21 52.21 0 01-10.088.989c-6.152 0-11.762-1.064-16.828-3.19-5.067-2.125-9.415-5.225-13.043-9.298-3.63-4.074-6.435-9.06-8.415-14.96C310.99 121.655 310 114.9 310 107.294c0-6.408.92-12.323 2.76-17.744 1.84-5.421 4.493-10.093 7.961-14.016 3.467-3.922 7.72-6.991 12.758-9.209C338.513 64.11 344.229 63 350.624 63zm.573 6c-4.696 0-8.904.702-12.623 2.105-3.721 1.404-6.936 3.421-9.65 6.053-2.713 2.631-4.908 5.79-6.586 9.474S319.55 94.439 319 99h60c0-4.679-.672-8.874-2.013-12.588-1.343-3.712-3.232-6.856-5.67-9.43-2.44-2.571-5.367-4.545-8.782-5.92-3.413-1.374-7.192-2.062-11.338-2.062zM435.546 63c6.526 0 12.368 1.093 17.524 3.28 5.154 2.186 9.5 5.286 13.04 9.298 3.538 4.013 6.238 8.85 8.099 14.51 1.861 5.66 2.791 11.994 2.791 19.002 0 7.008-.932 13.327-2.791 18.957-1.861 5.631-4.561 10.452-8.099 14.465-3.54 4.012-7.886 7.097-13.04 9.254-5.156 2.156-10.998 3.234-17.524 3.234-6.529 0-12.369-1.078-17.525-3.234-5.155-2.157-9.517-5.242-13.085-9.254-3.57-4.013-6.285-8.836-8.145-14.465-1.861-5.63-2.791-11.95-2.791-18.957 0-7.008.93-13.342 2.791-19.002 1.861-5.66 4.576-10.496 8.145-14.51 3.568-4.012 7.93-7.112 13.085-9.299C423.177 64.094 429.017 63 435.546 63zm-.501 86c5.341 0 10.006-.918 13.997-2.757 3.99-1.838 7.32-4.474 9.992-7.909 2.67-3.435 4.664-7.576 5.986-12.428 1.317-4.85 1.98-10.288 1.98-16.316 0-5.965-.66-11.389-1.98-16.27-1.322-4.88-3.316-9.053-5.986-12.519-2.67-3.463-6-6.13-9.992-7.999-3.991-1.867-8.657-2.802-13.997-2.802s-10.008.935-13.997 2.802c-3.991 1.87-7.322 4.536-9.992 8-2.671 3.465-4.68 7.637-6.03 12.518-1.35 4.881-2.026 10.305-2.026 16.27 0 6.026.675 11.465 2.025 16.316 1.35 4.852 3.36 8.993 6.031 12.428 2.67 3.435 6 6.07 9.992 7.91 3.99 1.838 8.656 2.756 13.997 2.756z"
            fill="currentColor"
          />
          <path
            d="M530.57 152h-20.05L474 60h18.35c1.61 0 2.967.39 4.072 1.166 1.103.778 1.865 1.763 2.283 2.959l17.722 49.138a92.762 92.762 0 012.551 8.429c.686 2.751 1.298 5.5 1.835 8.25.537-2.75 1.148-5.499 1.835-8.25a77.713 77.713 0 012.64-8.429l18.171-49.138c.417-1.196 1.164-2.181 2.238-2.96 1.074-.776 2.356-1.165 3.849-1.165H567l-36.43 92zM572 61h23v92h-23zM610 153V60.443h13.624c2.887 0 4.78 1.354 5.682 4.06l1.443 6.856a52.7 52.7 0 015.097-4.962 32.732 32.732 0 015.683-3.879 30.731 30.731 0 016.496-2.57c2.314-.632 4.855-.948 7.624-.948 5.832 0 10.63 1.579 14.39 4.736 3.758 3.157 6.57 7.352 8.434 12.585 1.444-3.068 3.248-5.698 5.413-7.894 2.165-2.194 4.541-3.984 7.127-5.367a32.848 32.848 0 018.254-3.068 39.597 39.597 0 018.796-.992c5.111 0 9.653.783 13.622 2.345 3.97 1.565 7.307 3.849 10.014 6.857 2.706 3.007 4.766 6.675 6.18 11.005C739.29 83.537 740 88.5 740 94.092V153h-22.284V94.092c0-5.894-1.294-10.329-3.878-13.306-2.587-2.977-6.376-4.465-11.368-4.465-2.286 0-4.404.391-6.358 1.172a15.189 15.189 0 00-5.144 3.383c-1.473 1.474-2.631 3.324-3.474 5.548-.842 2.225-1.263 4.781-1.263 7.668V153h-22.37V94.092c0-6.194-1.249-10.704-3.744-13.532-2.497-2.825-6.18-4.24-11.051-4.24-3.19 0-6.18.798-8.976 2.391-2.799 1.593-5.399 3.775-7.804 6.54V153H610zM572 30h23v19h-23z"
            fill="currentColor"
            fill-opacity=".8"
          />
        </g>
      </g>
    </svg>
  ]]

  local main = ''
  for _, tree in ipairs(lang_tree:trees()) do
    main = main
      .. (
        visit_node(
          tree:root(),
          0,
          tree,
          headings,
          { buf = buf, old = old, fname = fname, to_fname = to_fname, indent = 1 },
          stats
        )
      )
  end

  main = ([[
  <header class="container">
    <nav class="navbar navbar-expand-lg">
      <div class="container-fluid">
        <a href="/" class="navbar-brand" aria-label="logo">
          <!--TODO: use <img src="….svg"> here instead. Need one that has green lettering instead of gray. -->
          %s
          <!--<img src="https://neovim.io/logos/neovim-logo.svg" width="173" height="50" alt="Neovim" />-->
        </a>
        <div id="docsearch"></div> <!-- algolia docsearch https://docsearch.algolia.com/docs/docsearch-v3/ -->
      </div>
    </nav>
  </header>

  <div class="container golden-grid help-body">
  <div class="col-wide">
  <a name="%s" href="#%s"><h1 id="%s">%s</h1></a>
  <p>
    <i>
    Nvim <code>:help</code> pages, <a href="https://github.com/neovim/neovim/blob/master/scripts/gen_help_html.lua">generated</a>
    from <a href="https://github.com/neovim/neovim/blob/master/runtime/doc/%s">source</a>
    using the <a href="https://github.com/neovim/tree-sitter-vimdoc">tree-sitter-vimdoc</a> parser.
    </i>
  </p>
  <hr/>
  %s
  </div>
  ]]):format(
    logo_svg,
    stats.first_tags[1] or '',
    stats.first_tags[2] or '',
    stats.first_tags[2] or '',
    title,
    vim.fs.basename(fname),
    main
  )

  ---@type string
  local toc = [[
    <div class="col-narrow toc">
      <div><a href="index.html">Main</a></div>
      <div><a href="vimindex.html">Commands index</a></div>
      <div><a href="quickref.html">Quick reference</a></div>
      <hr/>
  ]]

  local n = 0 -- Count of all headings + subheadings.
  for _, h1 in ipairs(headings) do
    n = n + 1 + #h1.subheadings
  end
  for _, h1 in ipairs(headings) do
    ---@type string
    toc = toc .. ('<div class="help-toc-h1"><a href="#%s">%s</a>\n'):format(h1.tag, h1.name)
    if n < 30 or #headings < 10 then -- Show subheadings only if there aren't too many.
      for _, h2 in ipairs(h1.subheadings) do
        toc = toc
          .. ('<div class="help-toc-h2"><a href="#%s">%s</a></div>\n'):format(h2.tag, h2.name)
      end
    end
    toc = toc .. '</div>'
  end
  toc = toc .. '</div>\n'

  local bug_url = get_bug_url_nvim(fname, to_fname, 'TODO', nil)
  local bug_link = string.format('(<a href="%s" target="_blank">report docs bug...</a>)', bug_url)

  local footer = ([[
  <footer>
    <div class="container flex">
      <div class="generator-stats">
        Generated at %s from <code><a href="https://github.com/neovim/neovim/commit/%s">%s</a></code>
      </div>
      <div class="generator-stats">
      parse_errors: %d %s | <span title="%s">noise_lines: %d</span>
      </div>
    <div>

    <!-- algolia docsearch https://docsearch.algolia.com/docs/docsearch-v3/ -->
    <script src="https://cdn.jsdelivr.net/npm/@docsearch/js@3"></script>
    <script type="module">
      docsearch({
        container: '#docsearch',
        appId: 'X185E15FPG',
        apiKey: 'b5e6b2f9c636b2b471303205e59832ed',
        indexName: 'nvim',
      });
    </script>

  </footer>
  ]]):format(
    os.date('%Y-%m-%d %H:%M'),
    commit,
    commit:sub(1, 7),
    #stats.parse_errors,
    bug_link,
    html_esc(table.concat(stats.noise_lines, '\n')),
    #stats.noise_lines
  )

  html = ('%s%s%s</div>\n%s</body>\n</html>\n'):format(html, main, toc, footer)
  vim.cmd('q!')
  lang_tree:destroy()
  return html, stats
end

local function gen_css(fname)
  local css = [[
    :root {
      --code-color: #004b4b;
      --tag-color: #095943;
    }
    @media (prefers-color-scheme: dark) {
      :root {
        --code-color: #00c243;
        --tag-color: #00b7b7;
      }
    }
    @media (min-width: 40em) {
      .toc {
        position: fixed;
        left: 67%;
      }
      .golden-grid {
          display: grid;
          grid-template-columns: 65% auto;
          grid-gap: 1em;
      }
    }
    @media (max-width: 40em) {
      .golden-grid {
        /* Disable grid for narrow viewport (mobile phone). */
        display: block;
      }
    }
    .toc {
      /* max-width: 12rem; */
      height: 85%;  /* Scroll if there are too many items. https://github.com/neovim/neovim.github.io/issues/297 */
      overflow: auto;  /* Scroll if there are too many items. https://github.com/neovim/neovim.github.io/issues/297 */
    }
    .toc > div {
      text-overflow: ellipsis;
      overflow: hidden;
      white-space: nowrap;
    }
    html {
      scroll-behavior: auto;
    }
    body {
      font-size: 18px;
      line-height: 1.5;
    }
    h1, h2, h3, h4, h5 {
      font-family: sans-serif;
      border-bottom: 1px solid var(--tag-color); /*rgba(0, 0, 0, .9);*/
    }
    h3, h4, h5 {
      border-bottom-style: dashed;
    }
    .help-column_heading {
      color: var(--code-color);
    }
    .help-body {
      padding-bottom: 2em;
    }
    .help-line {
      /* font-family: ui-monospace,SFMono-Regular,SF Mono,Menlo,Consolas,Liberation Mono,monospace; */
    }
    .help-li {
      white-space: normal;
      display: list-item;
      margin-left: 1.5rem; /* padding-left: 1rem; */
    }
    .help-para {
      padding-top: 10px;
      padding-bottom: 10px;
    }

    .old-help-para {
      padding-top: 10px;
      padding-bottom: 10px;
      /* Tabs are used for alignment in old docs, so we must match Vim's 8-char expectation. */
      tab-size: 8;
      white-space: pre-wrap;
      font-size: 16px;
      font-family: ui-monospace,SFMono-Regular,SF Mono,Menlo,Consolas,Liberation Mono,monospace;
      word-wrap: break-word;
    }
    .old-help-para pre, .old-help-para pre:hover {
      /* Text following <pre> is already visually separated by the linebreak. */
      margin-bottom: 0;
      /* Long lines that exceed the textwidth should not be wrapped (no "pre-wrap").
         Since text may overflow horizontally, we make the contents to be scrollable
         (only if necessary) to prevent overlapping with the navigation bar at the right. */
      white-space: pre;
      overflow-x: auto;
    }

    /* TODO: should this rule be deleted? help tags are rendered as <code> or <span>, not <a> */
    a.help-tag, a.help-tag:focus, a.help-tag:hover {
      color: inherit;
      text-decoration: none;
    }
    .help-tag {
      color: var(--tag-color);
    }
    /* Tag pseudo-header common in :help docs. */
    .help-tag-right {
      color: var(--tag-color);
      margin-left: auto;
      margin-right: 0;
      float: right;
      display: block;
    }
    .help-tag a,
    .help-tag-right a {
      color: inherit;
    }
    .help-tag a:not(:hover),
    .help-tag-right a:not(:hover) {
      text-decoration: none;
    }
    h1 .help-tag, h2 .help-tag, h3 .help-tag {
      font-size: smaller;
    }
    .help-heading {
      white-space: normal;
      display: flex;
      flex-flow: row wrap;
      justify-content: space-between;
      gap: 0 15px;
    }
    /* The (right-aligned) "tags" part of a section heading. */
    .help-heading-tags {
      margin-right: 10px;
    }
    .help-toc-h1 {
    }
    .help-toc-h2 {
      margin-left: 1em;
    }
    .parse-error {
      background-color: red;
    }
    .unknown-token {
      color: black;
      background-color: yellow;
    }
    code {
      color: var(--code-color);
      font-size: 16px;
    }
    pre {
      /* Tabs are used in codeblocks only for indentation, not alignment, so we can aggressively shrink them. */
      tab-size: 2;
      white-space: pre-wrap;
      line-height: 1.3;  /* Important for ascii art. */
      overflow: visible;
      /* font-family: ui-monospace,SFMono-Regular,SF Mono,Menlo,Consolas,Liberation Mono,monospace; */
      font-size: 16px;
      margin-top: 10px;
    }
    pre:last-child {
      margin-bottom: 0;
    }
    pre:hover {
      overflow: visible;
    }
    .generator-stats {
      color: gray;
      font-size: smaller;
    }
  ]]
  tofile(fname, css)
end

-- Testing

local function ok(cond, expected, actual, message)
  assert(
    (not expected and not actual) or (expected and actual),
    'if "expected" is given, "actual" is also required'
  )
  if expected then
    assert(
      cond,
      ('%sexpected %s, got: %s'):format(
        message and (message .. '\n') or '',
        vim.inspect(expected),
        vim.inspect(actual)
      )
    )
    return cond
  else
    return assert(cond)
  end
end
local function eq(expected, actual, message)
  return ok(vim.deep_equal(expected, actual), expected, actual, message)
end

function M._test()
  tagmap = get_helptags('$VIMRUNTIME/doc')
  helpfiles = get_helpfiles(vim.fs.normalize('$VIMRUNTIME/doc'))

  ok(vim.tbl_count(tagmap) > 3000, '>3000', vim.tbl_count(tagmap))
  ok(
    vim.endswith(tagmap['vim.diagnostic.set()'], 'diagnostic.txt'),
    tagmap['vim.diagnostic.set()'],
    'diagnostic.txt'
  )
  ok(vim.endswith(tagmap['%:s'], 'cmdline.txt'), tagmap['%:s'], 'cmdline.txt')
  ok(is_noise([[vim:tw=78:isk=!-~,^*,^\|,^\":ts=8:noet:ft=help:norl:]]))
  ok(is_noise([[          NVIM  REFERENCE  MANUAL     by  Thiago  de  Arruda      ]]))
  ok(not is_noise([[vim:tw=78]]))

  eq(0, get_indent('a'))
  eq(1, get_indent(' a'))
  eq(2, get_indent('  a\n  b\n  c\n'))
  eq(5, get_indent('     a\n      \n        b\n      c\n      d\n      e\n'))
  eq(
    'a\n        \n   b\n c\n d\n e\n',
    trim_indent('     a\n             \n        b\n      c\n      d\n      e\n')
  )

  local fixed_url, removed_chars = fix_url('https://example.com).')
  eq('https://example.com', fixed_url)
  eq(').', removed_chars)
  fixed_url, removed_chars = fix_url('https://example.com.)')
  eq('https://example.com.', fixed_url)
  eq(')', removed_chars)
  fixed_url, removed_chars = fix_url('https://example.com.')
  eq('https://example.com', fixed_url)
  eq('.', removed_chars)
  fixed_url, removed_chars = fix_url('https://example.com)')
  eq('https://example.com', fixed_url)
  eq(')', removed_chars)
  fixed_url, removed_chars = fix_url('https://example.com')
  eq('https://example.com', fixed_url)
  eq('', removed_chars)

  print('all tests passed.\n')
end

--- @class nvim.gen_help_html.gen_result
--- @field helpfiles string[] list of generated HTML files, from the source docs {include}
--- @field err_count integer number of parse errors in :help docs
--- @field invalid_links table<string, any>

--- Generates HTML from :help docs located in `help_dir` and writes the result in `to_dir`.
---
--- Example:
---
---   gen('$VIMRUNTIME/doc', '/path/to/neovim.github.io/_site/doc/', {'api.txt', 'autocmd.txt', 'channel.txt'}, nil)
---
--- @param help_dir string Source directory containing the :help files. Must run `make helptags` first.
--- @param to_dir string Target directory where the .html files will be written.
--- @param include string[]|nil Process only these filenames. Example: {'api.txt', 'autocmd.txt', 'channel.txt'}
---
--- @return nvim.gen_help_html.gen_result result
function M.gen(help_dir, to_dir, include, commit, parser_path)
  vim.validate('help_dir', help_dir, function(d)
    return vim.fn.isdirectory(vim.fs.normalize(d)) == 1
  end, 'valid directory')
  vim.validate('to_dir', to_dir, 'string')
  vim.validate('include', include, 'table', true)
  vim.validate('commit', commit, 'string', true)
  vim.validate('parser_path', parser_path, function(f)
    return vim.fn.filereadable(vim.fs.normalize(f)) == 1
  end, true, 'valid vimdoc.{so,dll} filepath')

  local err_count = 0
  local redirects_count = 0
  ensure_runtimepath()
  tagmap = get_helptags(vim.fs.normalize(help_dir))
  helpfiles = get_helpfiles(help_dir, include)
  to_dir = vim.fs.normalize(to_dir)
  parser_path = parser_path and vim.fs.normalize(parser_path) or nil

  print(('output dir: %s\n\n'):format(to_dir))
  vim.fn.mkdir(to_dir, 'p')
  gen_css(('%s/help.css'):format(to_dir))

  for _, f in ipairs(helpfiles) do
    -- "foo.txt"
    local helpfile = vim.fs.basename(f)
    -- "to/dir/foo.html"
    local to_fname = ('%s/%s'):format(to_dir, get_helppage(helpfile))
    local html, stats =
      gen_one(f, nil, to_fname, not new_layout[helpfile], commit or '?', parser_path)
    tofile(to_fname, html)
    print(
      ('generated (%-2s errors): %-15s => %s'):format(
        #stats.parse_errors,
        helpfile,
        vim.fs.basename(to_fname)
      )
    )

    -- Generate redirect pages for renamed help files.
    local helpfile_tag = (helpfile:gsub('%.txt$', ''))
    local redirect_from = redirects[helpfile_tag]
    if redirect_from then
      local redirect_text = ([[
*%s*      Nvim

This document moved to: |%s|

==============================================================================
This document moved to: |%s|

This document moved to: |%s|

==============================================================================
 vim:tw=78:ts=8:ft=help:norl:
      ]]):format(
        redirect_from,
        helpfile_tag,
        helpfile_tag,
        helpfile_tag,
        helpfile_tag,
        helpfile_tag
      )
      local redirect_to = ('%s/%s'):format(to_dir, get_helppage(redirect_from))
      local redirect_html, _ =
        gen_one(redirect_from, redirect_text, redirect_to, false, commit or '?', parser_path)
      assert(redirect_html:find(helpfile_tag))
      tofile(redirect_to, redirect_html)

      print(
        ('generated (redirect) : %-15s => %s'):format(
          redirect_from .. '.txt',
          vim.fs.basename(to_fname)
        )
      )
      redirects_count = redirects_count + 1
    end

    err_count = err_count + #stats.parse_errors
  end

  print(('\ngenerated %d html pages'):format(#helpfiles + redirects_count))
  print(('total errors: %d'):format(err_count))
  print(('invalid tags: %s'):format(vim.inspect(invalid_links)))
  assert(#(include or {}) > 0 or redirects_count == vim.tbl_count(redirects)) -- sanity check
  print(('redirects: %d'):format(redirects_count))
  print('\n')

  --- @type nvim.gen_help_html.gen_result
  return {
    helpfiles = helpfiles,
    err_count = err_count,
    invalid_links = invalid_links,
  }
end

--- @class nvim.gen_help_html.validate_result
--- @field helpfiles integer number of generated helpfiles
--- @field err_count integer number of parse errors
--- @field parse_errors table<string, string[]>
--- @field invalid_links table<string, any> invalid tags in :help docs
--- @field invalid_urls table<string, any> invalid URLs in :help docs
--- @field invalid_spelling table<string, table<string, string>> invalid spelling in :help docs

--- Validates all :help files found in `help_dir`:
---  - checks that |tag| links point to valid helptags.
---  - recursively counts parse errors ("ERROR" nodes)
---
--- This is 10x faster than gen(), for use in CI.
---
--- @return nvim.gen_help_html.validate_result result
function M.validate(help_dir, include, parser_path)
  vim.validate('help_dir', help_dir, function(d)
    return vim.fn.isdirectory(vim.fs.normalize(d)) == 1
  end, 'valid directory')
  vim.validate('include', include, 'table', true)
  vim.validate('parser_path', parser_path, function(f)
    return vim.fn.filereadable(vim.fs.normalize(f)) == 1
  end, true, 'valid vimdoc.{so,dll} filepath')
  local err_count = 0 ---@type integer
  local files_to_errors = {} ---@type table<string, string[]>
  ensure_runtimepath()
  tagmap = get_helptags(vim.fs.normalize(help_dir))
  helpfiles = get_helpfiles(help_dir, include)
  parser_path = parser_path and vim.fs.normalize(parser_path) or nil

  for _, f in ipairs(helpfiles) do
    local helpfile = vim.fs.basename(f)
    local rv = validate_one(f, parser_path)
    print(('validated (%-4s errors): %s'):format(#rv.parse_errors, helpfile))
    if #rv.parse_errors > 0 then
      files_to_errors[helpfile] = rv.parse_errors
      vim.print(('%s'):format(vim.iter(rv.parse_errors):fold('', function(s, v)
        return s .. '\n    ' .. v
      end)))
    end
    err_count = err_count + #rv.parse_errors
  end

  ---@type nvim.gen_help_html.validate_result
  return {
    helpfiles = #helpfiles,
    err_count = err_count,
    parse_errors = files_to_errors,
    invalid_links = invalid_links,
    invalid_urls = invalid_urls,
    invalid_spelling = invalid_spelling,
  }
end

--- Validates vimdoc files on $VIMRUNTIME. and print human-readable error messages if fails.
---
--- If this fails, try these steps (in order):
--- 1. Fix/cleanup the :help docs.
--- 2. Fix the parser: https://github.com/neovim/tree-sitter-vimdoc
--- 3. File a parser bug, and adjust the tolerance of this test in the meantime.
---
--- @param help_dir? string e.g. '$VIMRUNTIME/doc' or './runtime/doc'
function M.run_validate(help_dir)
  help_dir = vim.fs.normalize(help_dir or '$VIMRUNTIME/doc')
  print('doc path = ' .. vim.uv.fs_realpath(help_dir))

  local rv = M.validate(help_dir)

  -- Check that we actually found helpfiles.
  ok(rv.helpfiles > 100, '>100 :help files', rv.helpfiles)

  eq({}, rv.parse_errors, 'no parse errors')
  eq(0, rv.err_count, 'no parse errors')
  eq({}, rv.invalid_links, 'invalid tags in :help docs')
  eq({}, rv.invalid_urls, 'invalid URLs in :help docs')
  eq(
    {},
    rv.invalid_spelling,
    'invalid spelling in :help docs (see spell_dict in scripts/gen_help_html.lua)'
  )
end

--- Test-generates HTML from docs.
---
--- 1. Test that gen_help_html.lua actually works.
--- 2. Test that parse errors did not increase wildly. Because we explicitly test only a few
---    :help files, we can be precise about the tolerances here.
--- @param help_dir? string e.g. '$VIMRUNTIME/doc' or './runtime/doc'
function M.test_gen(help_dir)
  local tmpdir = vim.fs.dirname(vim.fn.tempname())
  help_dir = vim.fs.normalize(help_dir or '$VIMRUNTIME/doc')
  print('doc path = ' .. vim.uv.fs_realpath(help_dir))

  -- Because gen() is slow (~30s), this test is limited to a few files.
  local input = { 'help.txt', 'index.txt', 'nvim.txt' }
  local rv = M.gen(help_dir, tmpdir, input)
  eq(#input, #rv.helpfiles)
  eq(0, rv.err_count, 'parse errors in :help docs')
  eq({}, rv.invalid_links, 'invalid tags in :help docs')
end

return M