｢‍｣ Lingenic

HTML in HSV

Parallel-parseable structured documents

HSV enables fully parallel processing of HTML-like structured documents. Represent HTML as HSV, split at any delimiter, parse chunks on separate cores. No speculation, no state synchronization, linear scaling.

Proof-of-concept: go-html-parallel — parallel HTML parsing in ~50 lines of Go.

The Problem with HTML Parsing

HTML and XML parsing has been sequential for 30 years. Research into parallel HTML parsing achieved limited results:

Why so little progress? The inherent complexity of HTML's stateful parsing model:

• State machine: Track whether you're inside a tag, attribute, comment, CDATA...
• Tag matching: Opening tags must match closing tags
• Escape sequences: &, <, >, "
• Context-dependent: < means different things in different places

You cannot split an HTML file in the middle and parse the chunks independently. The parser must process byte-by-byte from the start.

HTML in HSV

HSV represents HTML-like structures using its own delimiters:

<div class="container">
  <p>Hello world</p>
  <a href="https://example.com">
    Click here
  </a>
</div>

[STX] html:div [US] [SSA]
  html:class [US] container [RS]
  html:p [US] Hello world [RS]
  html:a [US] [SSA]
    html:href [US] https://example.com [RS]
    html:text [US] Click here
  [ESA]
[ESA] [ETX]

The structure uses HSV control characters:

• SSA / ESA (0x86/0x87) — nesting (like opening/closing tags)
• [RS] (RS) — sibling elements
• [US] (US) — attribute/value pairs
• html: prefix — namespace convention

Parallel Parsing

HTML-in-HSV inherits HSV's parallel-parseable structure.

Find a delimiter (SSA, ESA, RS, US), split, parse chunks on separate cores. Same as any HSV data.

No Escaping

Quotes, angles, and ampersands are just content:

<p>Use &lt;div&gt; for containers</p>
<p>A &amp; B &amp; C</p>
<a href="?a=1&amp;b=2">Link</a>

[STX] html:p [US] Use 
 for containers [FS]
html:p [US] A & B & C [FS]
html:a [US] [SSA] html:href [US] ?a=1&b=2 [RS] html:text [US] Link [ESA] [ETX]

The control characters (0x86, 0x87, 0x1E, 0x1F) never appear in normal text, so no escaping is ever needed.

Rich Text Content

Multi-line content with formatting:

[STX] html:article [US] [SSA]
  html:h1 [US] The Title [RS]
  html:p [US] First paragraph with "quotes" and . [RS]
  html:p [US] Second paragraph.
This continues on a new line.
And another. [RS]
  html:blockquote [US] [SSA]
    html:p [US] A nested quote with special chars: <>&"'
  [ESA]
[ESA] [ETX]

Newlines are literal. Quotes are literal. Everything except the reserved control characters is data.

AST Representation

HTML-in-HSV is essentially an AST (Abstract Syntax Tree) format:

[STX] tag [US] div [RS]
  attr:class [US] container [RS]
  children [US] [SSA]
    tag [US] p [RS] text [US] Hello [RS]
    tag [US] a [RS] attr:href [US] /link [RS] text [US] Click
  [ESA] [ETX]

This makes it ideal for:

• Template engines: Compile templates to HSV, render in parallel
• DOM manipulation: Parse once, transform, serialize
• Static site generators: Process pages in parallel
• Document storage: Store structured content without escaping nightmares

Converting HTML to HSV

The mapping is straightforward:

Summary

HTML-in-HSV gives you:

• Parallel parsing of structured documents
• No escaping for quotes, angles, ampersands
• AST-level representation
• Same HSV tooling for documents and data

30 years of sequential HTML parsing. HSV changes that.