Core Algorithm

HTML Structure of a Command Page

Each command page follows a consistent DITA-derived XHTML structure:

html

<body id="bgp_aggregate">
  <h1 class="title topictitle1">32.2 経路の集約の設定</h1>
  <div class="body refbody">

    <div class="section">
      <h2 class="title sectiontitle">[書式]</h2>
      <ul class="sl simple">
        <li class="sli">
          <span class="ph synph">
            <span class="keyword kwd">bgp</span>
            <span class="ph delim"> </span>
            <span class="keyword kwd">aggregate</span>
            ...
          </span>
        </li>
      </ul>
    </div>

    <div class="section">
      <h2 class="title sectiontitle">[設定値及び初期値]</h2>
      <!-- nested ul + table structure -->
    </div>

    <div class="section">
      <h2 class="title sectiontitle">[説明]</h2>
      <p class="p">...</p>
    </div>

  </div>
  <div class="related-links">
    <div class="parentlink"><a href="...">BGP</a></div>
  </div>
</body>

Section Dispatch

typescript

$('.section').each((_, section) => {
  const heading = $(section).find('h2.sectiontitle').first().text().trim()

  if (heading === '[書式]') {
    // → extractSyntax: collect li.sli text → fenced code block
  } else {
    // → sectionToMarkdown: walk children, tables → GFM, text → paragraphs
  }
})

Syntax Extraction (`extractSyntax`)

Each li.sli inside a [書式] section contains one syntax variant (e.g. the positive form and the no form). The raw text of each list item is collected and joined into a single fenced code block:

typescript

$section.find('li.sli').each((_, li) => {
  lines.push($(li).text().replace(/\s+/g, ' ').trim())
})
return '```\n' + lines.join('\n') + '\n```'

The span.keyword.kwd, span.ph.var, and span.ph.delim elements are all flattened to text — their CSS classes exist only for browser rendering.

Table Conversion (`tableToMarkdown`)

HTML tables are converted row by row. The first <tr> becomes the header row; a separator row of --- cells is inserted; remaining rows become body rows.

typescript

$table.find('tr').each((_, tr) => {
  const cells = $(tr).find('th, td').map((_, c) => $(c).text().trim()).get()
  rows.push(cells)
})
// rows[0] = header, rows[1..] = body

Limitation: rowspan and colspan attributes are not handled. Cells that span multiple rows will appear in the first row only; subsequent rows will have fewer columns than the header. The text content is never lost, but the visual layout may differ from the original.

Mixed-Content Sections

Some sections (especially [設定値及び初期値]) interleave plain text with tables:

<ul>
  <li>
    <var>ip_address/mask</var>
    <ul>
      <li>[設定値]: ...</li>
    </ul>
    <table>...</table>
    <li>[初期値]: ...</li>
  </li>
</ul>

sectionToMarkdown handles this by walking direct children and dispatching:

table elements → tableToMarkdown
elements containing nested tables → recurse one level, extract tables and non-table text separately
all other elements → .text() as plain paragraph

Core Algorithm ​

HTML Structure of a Command Page ​

Section Dispatch ​

Syntax Extraction (extractSyntax) ​

Table Conversion (tableToMarkdown) ​

Mixed-Content Sections ​

Core Algorithm

HTML Structure of a Command Page

Section Dispatch

Syntax Extraction (`extractSyntax`)

Table Conversion (`tableToMarkdown`)

Mixed-Content Sections