fix(math): keep mixed equation tables instead of re-extracting TeX

src/elements/equation-table.ts

@@ -0,0 +1,31 @@
+import { isElement, isTextNode } from '../utils';
+import { isDirectTableChild } from '../utils/dom';
+
+export const EQUATION_TABLE_SELECTOR = 'table.ltx_equation, table.ltx_eqn_table, table.ltx_equationgroup';
+
+/** Only unwrap a whole equation, never cell fragments or alternative MathML. */
+export function getSingleEquationMath(table: Element): Element | null {
+	const maths = Array.from(table.querySelectorAll('math'))
+		.filter(math => isDirectTableChild(math, table) && !math.parentElement?.closest('math, annotation-xml'));
+	if (maths.length !== 1) return null;
+	const math = maths[0];
+	// Keep unconvertible sources and XML alternatives in their table context.
+	if (!math.getAttribute('data-latex')?.trim() || math.querySelector('annotation-xml, math')) return null;
+	// A generated text fallback is not evidence for promoting native inline math.
+	if (math.getAttribute('display') !== 'block' && !math.querySelector('annotation')) return null;
+	const rows = Array.from(table.querySelectorAll('tr')).filter(row => isDirectTableChild(row, table));
+	if (rows.length !== 1) return null;
+
+	const isEquationLayout = (node: Node): boolean => {
+		if (node === math) return true;
+		if (isTextNode(node)) return !node.textContent?.trim();
+		if (!isElement(node)) return node.nodeType === 8;
+		// Equation numbers are dropped, as for unwrapped equations before
+		if (node.classList.contains('ltx_eqn_eqno')) return true;
+		if (!['table', 'tbody', 'tr', 'td'].includes(node.nodeName.toLowerCase())) return false;
+		if (node !== table && node.nodeName.toLowerCase() === 'table') return false;
+		if (node.hasAttribute('colspan') || node.hasAttribute('rowspan')) return false;
+		return Array.from(node.childNodes).every(isEquationLayout);
+	};
+	return isEquationLayout(table) ? math : null;
+}

src/elements/math.base.ts

@@ -658,6 +658,17 @@ export function wrapRawLatexDelimiters(element: Element, doc: Document): void {
 
 export type LatexToMathML = (latex: string, isBlock: boolean, doc: Document) => Element | null;
 
+const hasPresentationalMathContent = (math: Element): boolean => {
+	const isPresentationalNode = (node: Node): boolean => {
+		if (isTextNode(node)) return Boolean(node.textContent?.trim());
+		if (!isElement(node)) return false;
+		const tag = node.tagName.toLowerCase();
+		if (tag === 'annotation' || tag === 'annotation-xml') return false;
+		return tag !== 'semantics' || Array.from(node.childNodes).some(isPresentationalNode);
+	};
+	return Array.from(math.childNodes).some(isPresentationalNode);
+};
+
 export const createMathElement = (
 	mathData: MathData | null,
 	latex: string | null,
@@ -676,7 +687,11 @@ export const createMathElement = (
 		const fragment = parseHTML(doc, mathData.mathml);
 		const mathContent = fragment.querySelector('math');
 		if (mathContent) {
-			transferContent(mathContent, cleanMathEl);
+			if (latex && !hasPresentationalMathContent(mathContent) && !mathContent.querySelector('annotation-xml')) {
+				cleanMathEl.textContent = latex;
+			} else {
+				transferContent(mathContent, cleanMathEl);
+			}
 		}
 	}
 	// If no MathML but we have LaTeX, convert it when possible or keep it as text

src/markdown.ts

@@ -814,18 +814,16 @@ export function createMarkdownContent(content: string, url: string) {
 	});
 
 	function handleNestedEquations(element: GenericElement): string {
-		const mathElements = element.querySelectorAll('math');
-		if (mathElements.length === 0) return '';
-
-		return Array.from(mathElements).map(mathElement => {
-			const annotation = mathElement.querySelector('annotation[encoding="application/x-tex"]');
-			const latex = annotation?.textContent?.trim() || mathElement.getAttribute('alttext')?.trim();
-			if (latex) {
-				const isInline = mathElement.closest('.ltx_eqn_inline, .mwe-math-element-inline') !== null;
+		return Array.from(element.querySelectorAll('math'))
+			.filter(math => !math.parentElement?.closest('math'))
+			.map(math => {
+				const latex = extractLatex(asGenericElement(math));
+				if (!latex) return '';
+				const isInline = math.closest('.ltx_eqn_inline, .mwe-math-element-inline') !== null;
 				return formatMath(latex, !isInline);
-			}
-			return '';
-		}).join('\n\n');
+			})
+			.filter(Boolean)
+			.join('\n\n');
 	}
 
 	function cleanupTableHTML(element: GenericElement): string {

src/standardize.ts

@@ -16,6 +16,7 @@ import {
 
 import { DefuddleMetadata } from './types';
 import { mathRules, createCleanMathEl } from './elements/math';
+import { EQUATION_TABLE_SELECTOR, getSingleEquationMath } from './elements/equation-table';
 import { wrapRawLatexDelimiters, extractLatexFromImageSrc, LOOKS_LIKE_LATEX_RE } from './elements/math.base';
 import { codeBlockRules } from './elements/code';
 import { headingRules, removePermalinkAnchors, isPermalinkAnchor } from './elements/headings';
@@ -499,6 +500,7 @@ function stripUnwantedAttributes(element: Element, debug: boolean, extractorOutp
 				)) ||
 				// Preserve code block language classes, footnote backref class, and callout classes
 				(attrName === 'class' && (
+					(tag === 'table' && attrValue === 'ltx_eqn_table') ||
 					(tag === 'code' && attrValue.startsWith('language-')) ||
 					attrValue === 'footnote-backref' ||
 					/^callout(?:-|$)/.test(attrValue)
@@ -1381,54 +1383,22 @@ function standardizeElements(element: Element, doc: Document, subProfile?: Recor
 		outerCode.replaceWith(pre);
 	});
 
-	// arXiv LaTeXML: Convert equation tables to <math> elements before attribute stripping
-	const equationTables = Array.from(element.querySelectorAll('table.ltx_equation, table.ltx_eqn_table, table.ltx_equationgroup'));
-	const hasPresentationalMathContent = (mathEl: Element): boolean => {
-		const isPresentationalNode = (node: Node): boolean => {
-			if (isTextNode(node)) return Boolean(node.textContent?.trim());
-			if (!isElement(node)) return false;
-
-			const tag = node.tagName.toLowerCase();
-			if (tag === 'annotation' || tag === 'annotation-xml') return false;
-			if (tag === 'semantics') {
-				return Array.from(node.childNodes).some(isPresentationalNode);
+	// Math rules own normalization. The table adapter only unwraps a single
+	// whole equation; labels, fragments, alternatives and mixed content stay put.
+	const equationTables = Array.from(element.querySelectorAll(EQUATION_TABLE_SELECTOR));
+	equationTables.reverse().forEach(table => {
+		if (table.parentElement?.closest('math, annotation-xml')) return;
+		const math = getSingleEquationMath(table);
+		if (math) {
+			// Retain the existing standalone equation display policy.
+			if (table.classList.contains('ltx_equation') || table.classList.contains('ltx_equationgroup')) {
+				math.setAttribute('display', 'block');
 			}
-
-			return true;
-		};
-
-		return Array.from(mathEl.childNodes).some(isPresentationalNode);
-	};
-	equationTables.forEach(table => {
-		const mathElements = table.querySelectorAll('math');
-		if (mathElements.length === 0) return;
-
-		const fragment = doc.createDocumentFragment();
-		mathElements.forEach(mathEl => {
-			// Extract LaTeX from alttext or annotation
-			const alttext = mathEl.getAttribute('alttext');
-			const annotation = mathEl.querySelector('annotation[encoding="application/x-tex"]');
-			const latex = alttext || annotation?.textContent?.trim() || '';
-
-			if (!latex) return;
-
-			const isBlock = mathEl.getAttribute('display') === 'block' ||
-				table.classList.contains('ltx_equation') ||
-				table.classList.contains('ltx_equationgroup');
-
-			const cleanMath = mathEl.cloneNode(true) as Element;
-			cleanMath.setAttribute('xmlns', 'http://www.w3.org/1998/Math/MathML');
-			cleanMath.setAttribute('display', isBlock ? 'block' : 'inline');
-			cleanMath.setAttribute('data-latex', latex);
-			if (!hasPresentationalMathContent(cleanMath)) {
-				cleanMath.textContent = latex;
-			}
-			fragment.appendChild(cleanMath);
-		});
-
-		if (fragment.childNodes.length > 0) {
-			table.replaceWith(fragment);
+			table.replaceWith(math);
 			processedCount++;
+		} else {
+			// A narrow marker survives attribute stripping and layout cleanup.
+			table.setAttribute('class', 'ltx_eqn_table');
 		}
 	});
 
@@ -1472,6 +1442,7 @@ function standardizeElements(element: Element, doc: Document, subProfile?: Recor
 	const tables = Array.from(element.querySelectorAll('table'));
 	tables.forEach(table => {
 		if (!table.parentNode) return;
+		if (table.classList.contains('ltx_eqn_table')) return;
 
 		const directCells = Array.from(table.querySelectorAll('td, th'))
 			.filter(cell => isDirectTableChild(cell, table));

tests/expected/math--equation-table-mathml.html

@@ -1 +1 @@
-<article>  <p>This article describes three mathematical expressions. Each equation includes its structured representation or its original source.</p> <math data-latex="x" display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>x</mi> <annotation>x</annotation></semantics></math> <math data-latex="y" display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>y</mi> <annotation>y</annotation></semantics></math> <math data-latex="z" display="block" xmlns="http://www.w3.org/1998/Math/MathML">z</math> <p>The expressions appear in order, with each equation occurring once. Structured MathML should remain available, while annotation-only equations retain a visible fallback.</p> </article>
+<article>  <p>This article describes three mathematical expressions. Each equation includes its structured representation or its original source.</p> <table class="ltx_eqn_table"><tbody> <tr><td><math data-latex="x" display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>x</mi> <annotation>x</annotation></semantics></math></td></tr> <tr><td><math data-latex="y" display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>y</mi> <annotation>y</annotation></semantics></math></td></tr> <tr><td><math data-latex="z" display="block" xmlns="http://www.w3.org/1998/Math/MathML">z</math></td></tr> </tbody></table> <p>The expressions appear in order, with each equation occurring once. Structured MathML should remain available, while annotation-only equations retain a visible fallback.</p> </article>

tests/expected/math--equation-table-mathml.md

@@ -12,13 +12,13 @@ This article describes three mathematical expressions. Each equation includes it
 $$
 x
 $$
- 
+
 $$
 y
 $$
- 
+
 $$
 z
 $$
 
-The expressions appear in order, with each equation occurring once. Structured MathML should remain available, while annotation-only equations retain a visible fallback.
+The expressions appear in order, with each equation occurring once. Structured MathML should remain available, while annotation-only equations retain a visible fallback.

tests/expected/math--equation-table-ownership.html

@@ -0,0 +1 @@
+<article>  <p>Equations may use tables to arrange a whole expression, individual fragments, operators and labels. Extraction must not confuse a second representation with an independent equation.</p> <h2>Single equation</h2> <math data-latex="a" display="block" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>a</mi> <annotation>a</annotation></semantics></math> <h2>Label</h2> <table class="ltx_eqn_table"><tbody><tr><td><math data-latex="b" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>b</mi> <annotation>b</annotation></semantics></math></td> <td>(7)</td></tr></tbody></table> <h2>Rows and operators</h2> <table class="ltx_eqn_table"><tbody><tr><td><math data-latex="x" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>x</mi> <annotation>x</annotation></semantics></math></td> <td>=</td> <td><math data-latex="y" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>y</mi> <annotation>y</annotation></semantics></math></td> <td>(8a)</td></tr> <tr><td><math data-latex="z" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>z</mi> <annotation>z</annotation></semantics></math></td> <td>=</td> <td><math data-latex="w" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>w</mi> <annotation>w</annotation></semantics></math></td> <td>(8b)</td></tr></tbody></table> <h2>Native source without annotation</h2> <table class="ltx_eqn_table"><tbody><tr><td><math data-latex="c" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><semantics><mi>c</mi> <annotation>c</annotation></semantics></math></td></tr> <tr><td><math data-latex="d" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>d</mi></math></td></tr></tbody></table> <h2>Header and mixed content</h2> <table class="ltx_eqn_table"><tbody><tr><th>Expression</th> <th>Meaning</th></tr> <tr><td><math data-latex="g" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>g</mi></math></td> <td>Measured value</td></tr></tbody></table> <h2>Nested table</h2> <table class="ltx_eqn_table"><tbody><tr><td><math data-latex="h" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>h</mi></math></td> <td><table class="ltx_eqn_table"><tbody><tr><td><math data-latex="i" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>i</mi></math></td></tr></tbody></table></td> <td>(9)</td></tr></tbody></table> <h2>Ordinary data table</h2> <table><tbody><tr><th>Item</th> <th>Value</th></tr> <tr><td>Sample</td> <td>10</td></tr></tbody></table> <h2>Sanitized label</h2> <table class="ltx_eqn_table"><tbody><tr><td><math data-latex="j" display="inline" xmlns="http://www.w3.org/1998/Math/MathML"><mi>j</mi></math></td> <td><span title="label&quot; onmouseover=&quot;alert(1)">(10)</span></td></tr></tbody></table> <p>The single standalone equation retains the existing promotion policy. Tables containing fragments retain their cell boundaries rather than implying that every fragment is a complete display equation.</p> </article>

tests/expected/math--equation-table-ownership.md

@@ -0,0 +1,80 @@
+```json
+{
+  "title": "Equation table ownership",
+  "author": "",
+  "site": "example.com",
+  "published": ""
+}
+```
+
+Equations may use tables to arrange a whole expression, individual fragments, operators and labels. Extraction must not confuse a second representation with an independent equation.
+
+## Single equation
+
+$$
+a
+$$
+
+## Label
+
+$$
+b
+$$
+
+## Rows and operators
+
+$$
+x
+$$
+
+$$
+y
+$$
+
+$$
+z
+$$
+
+$$
+w
+$$
+
+## Native source without annotation
+
+$$
+c
+$$
+
+$$
+d
+$$
+
+## Header and mixed content
+
+$$
+g
+$$
+
+## Nested table
+
+$$
+h
+$$
+
+$$
+i
+$$
+
+## Ordinary data table
+
+| Item | Value |
+| --- | --- |
+| Sample | 10 |
+
+## Sanitized label
+
+$$
+j
+$$
+
+The single standalone equation retains the existing promotion policy. Tables containing fragments retain their cell boundaries rather than implying that every fragment is a complete display equation.

tests/fixtures/math--equation-table-mathml.html

@@ -6,7 +6,7 @@
 <article>
 <h1>Equation representations</h1>
 <p>This article describes three mathematical expressions. Each equation includes its structured representation or its original source.</p>
-<table class="ltx_equationgroup">
+<table class="ltx_equationgroup"><tbody>
 <tr><td><table class="ltx_equation"><tr><td>
 <math display="block" alttext="x"><semantics><mi>x</mi><annotation encoding="application/x-tex">x</annotation></semantics></math>
 </td></tr></table></td></tr>
@@ -16,7 +16,7 @@ <h1>Equation representations</h1>
 <tr><td><table class="ltx_equation"><tr><td>
 <math display="block"><semantics><annotation encoding="application/x-tex">z</annotation></semantics></math>
 </td></tr></table></td></tr>
-</table>
+</tbody></table>
 <p>The expressions appear in order, with each equation occurring once. Structured MathML should remain available, while annotation-only equations retain a visible fallback.</p>
 </article>
 </body>

tests/fixtures/math--equation-table-ownership.html

@@ -0,0 +1,23 @@
+<!-- {"url":"https://example.com/equation-tables"} -->
+<!doctype html>
+<html><head><title>Equation table ownership</title></head><body><article>
+<h1>Equation table ownership</h1>
+<p>Equations may use tables to arrange a whole expression, individual fragments, operators and labels. Extraction must not confuse a second representation with an independent equation.</p>
+<h2>Single equation</h2>
+<table class="ltx_equation"><tbody><tr><td><math display="inline"><semantics><mi>a</mi><annotation encoding="application/x-tex">a</annotation></semantics></math></td></tr></tbody></table>
+<h2>Label</h2>
+<table class="ltx_equation"><tbody><tr><td><math display="inline"><semantics><mi>b</mi><annotation encoding="application/x-tex">b</annotation></semantics></math></td><td>(7)</td></tr></tbody></table>
+<h2>Rows and operators</h2>
+<table class="ltx_equationgroup"><tbody><tr><td><math display="inline"><semantics><mi>x</mi><annotation encoding="application/x-tex">x</annotation></semantics></math></td><td>=</td><td><math display="inline"><semantics><mi>y</mi><annotation encoding="application/x-tex">y</annotation></semantics></math></td><td>(8a)</td></tr><tr><td><math display="inline"><semantics><mi>z</mi><annotation encoding="application/x-tex">z</annotation></semantics></math></td><td>=</td><td><math display="inline"><semantics><mi>w</mi><annotation encoding="application/x-tex">w</annotation></semantics></math></td><td>(8b)</td></tr></tbody></table>
+<h2>Native source without annotation</h2>
+<table class="ltx_equation"><tbody><tr><td><math display="inline"><semantics><mi>c</mi><annotation encoding="application/x-tex">c</annotation></semantics></math></td></tr><tr><td><math display="inline"><mi>d</mi></math></td></tr></tbody></table>
+<h2>Header and mixed content</h2>
+<table class="ltx_eqn_table"><tbody><tr><th>Expression</th><th>Meaning</th></tr><tr><td><math display="inline"><mi>g</mi></math></td><td>Measured value</td></tr></tbody></table>
+<h2>Nested table</h2>
+<table class="ltx_equation"><tbody><tr><td><math display="inline"><mi>h</mi></math></td><td><table class="ltx_equation"><tbody><tr><td><math display="inline"><mi>i</mi></math></td></tr></tbody></table></td><td>(9)</td></tr></tbody></table>
+<h2>Ordinary data table</h2>
+<table><tbody><tr><th>Item</th><th>Value</th></tr><tr><td>Sample</td><td>10</td></tr></tbody></table>
+<h2>Sanitized label</h2>
+<table class="ltx_equation" onclick="alert(1)"><tbody><tr><td><math display="inline" data-latex="j"><mi>j</mi></math></td><td><a href="javascript:alert(1)" onclick="alert(1)"><span title='label" onmouseover="alert(1)'>(10)</span></a><script>alert(1)</script></td></tr></tbody></table>
+<p>The single standalone equation retains the existing promotion policy. Tables containing fragments retain their cell boundaries rather than implying that every fragment is a complete display equation.</p>
+</article></body></html>