Mathematical typesetting: MathML in, accessible Formula tags out

The boxesandglue stack can now typeset mathematics. An OpenType-MATH engine renders presentation MathML to TeX-quality output, and, when the document output is set toPDF/UA, each formula is tagged as an accessible Formula structure element.

Write MathML, get a formula

Embed MathML directly in your HTML; htmlbag picks up <math> as foreign content and typesets it through the engine. No external converter is needed:

<style>
  @font-face { font-family: "LM Math"; src: url("latinmodern-math.otf"); }
  math { font-family: "LM Math"; font-size: 12pt; }
</style>

<p>The Pythagorean relation
<math>
  <msup><mi>a</mi><mn>2</mn></msup><mo>+</mo>
  <msup><mi>b</mi><mn>2</mn></msup><mo>=</mo>
  <msup><mi>c</mi><mn>2</mn></msup>
</math>
holds for every right triangle.</p>

The engine implements the OpenType MATH table: real per-glyph extents, style-aware fraction and script shifts, stretchy delimiters via MathVariants/GlyphAssembly, and big-operator limits in display mode. Identifier styling follows the MathML convention: a single-character <mi> defaults to math-italic, a multi-character one (a function name like sin) stays upright.

Supported presentation elements: <mrow>, <mi>, <mn>, <mo>, <mfrac> (and binomials via linethickness="0"), <msqrt>, <mroot>, <msup>, <msub>, <msubsup>, <munder>, <mover>, <munderover>, plus the transparent containers. See Mathematics (MathML) for the full list and the current scope boundary (matrices, <mtext> and extended mathvariant values are future work).

Accessible by default under PDF/UA

Set the conformance level and every formula is tagged automatically:

Each <math> becomes a Formula structure element carrying:

• /Alt — a plain-text fallback (from the MathML alttext attribute, or the token content concatenated in order).
• /AF — under PDF/UA-2 — the MathML source embedded as an associated file (application/mathml+xml, /AFRelationship /Supplement), so a MathML-aware reader can speak the math semantically.

Because a formula sits inside a sentence, the renderer splits the paragraph’s marked content so the reading order is text · formula · text, not text · text · formula. This rests on a new reading-order /K serialization in the structure tree that interleaves marked content, object references, and child elements by document position.

Showcase

boxesandglue-examples/glu/html/mathml renders five formulas: Pythagoras, the quadratic formula, a Gauss-summation in display mode, an indexed n-th root, and a derivative (all as as PDF/UA-2). pdfa11y reports Verdict: PASS, including the dedicated MathML checks (the Formula has a math representation, and the associated file is well-formed MathML with the right subtype and relationship).

glu boxesandglue-examples/glu/html/mathml/mathml.html
pdfa11y mathml.pdf      # Verdict: PASS

New HTML metadata: <meta name="pdf-format">

HTML documents have no frontmatter, so requesting a conformance level previously meant passing --format on every run. HTML documents can now declare it inline with <meta name="pdf-format" content="PDF/UA-2"> (mirroring the Markdown format: key), and HTML mode now also derives the document title from the <title> element. The glu HTML mode page has the details.

Reading more

• Mathematics (MathML) — supported elements, mathvariant rules, scope.
• PDF/UA tagging — Mathematics (Formula) — how the Formula element, /Alt, and the MathML associated file are emitted.
• glu HTML mode — conformance metadata and a MathML example.

What’s next

Still missing is a decent interface to Markdown. You might want to write your code in TeX/LaTeX style, for example

\[ \sum_{k=0}^{n} k^2 \]

This will be a bit of work, but I am working on it.