rapier-markdown-kit 1.2.8 → 1.2.10

package/dist/kit/render-print.mjs

@@ -9,7 +9,6 @@ import {stripOneTerminator as _rapierWillStripOneTerminator} from '../agent/will
 import {quotedWillSpan} from '../spec/md-marks.mjs';
 const RAPIER_WILL_SENTINEL_MARK = '⁣';
 // SPDX-License-Identifier: MIT
-// The document's renderer. DOM, codecs and host state are explicit inputs.
 function createPrintRenderer(runtime) {
   const {Node, _rapierWillParse, crypto, document, globalThis} = runtime;
 function _rapierWillSentinelPlan(canonical, kind) {
@@ -46,8 +45,6 @@ function _rapierPrintWillAdmit(canonical, plan) {
 function _rapierPrintWillFont(markers, advanceOverride) {
 	const metrics = _rapierPrintWillMetrics(markers);
 	const size = metrics.size;
-	// The page's carrier face keeps the short marker advance. Searchable text is
-	// positioned at 500 units, which is also /DW, so this copy can be asked to match it.
 	const advance = advanceOverride ?? metrics.advance;
 	const scalars = Array.from(new Set(markers.flatMap(marker => Array.from(marker, ch => ch.codePointAt(0)))));
 	const order = scalars.filter(code => code < 0x10000).sort((a, b) => a - b)
@@ -75,8 +72,6 @@ function _rapierPrintWillFont(markers, advanceOverride) {
 	const cmap = narrow.length <= 0xffff
 		? cat(u16(0), u16(2), u16(3), u16(1), u32(20), u16(3), u16(10), u32(20 + narrow.length), narrow, full)
 		: cat(u16(0), u16(1), u16(3), u16(10), u32(12), full);
-	// The collinear contour has zero area and enough bounds for native print at 2px.
-	// Keep it below the baseline so a carrier at a page break does not straddle both page clips.
 	const bounds = [0, -400, 10, 0];
 	const outline = cat(u16(1), bounds.map(u16), u16(2), u16(0), [1, 1, 1], u16(0), u16(10), u16(-5), u16(bounds[1]), u16(bounds[3] - bounds[1]), u16((bounds[1] - bounds[3]) / 2), [0]);
 	const utf16 = text => Array.from(text, ch => u16(ch.charCodeAt(0)));
@@ -120,7 +115,6 @@ function _rapierPrintWillMetrics(markers) {
 }
 
 function _rapierPrintWillPlace(root, plan) {
-	// A marker the page holds inside other content (a code block, say) is quoted text, which a reader of the PDF's text could not tell from a carried marker.
 	const lost = () => { throw _rapierPrintWillLost('a marker stands inside other content, where a reader of the PDF\'s text could not tell it from a carried marker', 'Take the quoted marker out'); };
 	const measure = _rapierPrintWillMetrics(plan.markers);
 	const tokens = new Map(plan.markers.map((marker, index) => [plan.sentinel + index + RAPIER_WILL_SENTINEL_MARK, index]));
@@ -153,8 +147,6 @@ function _rapierPrintWillPlace(root, plan) {
 			carrier.textContent = plan.markers[row.index];
 			return carrier;
 		});
-		// A rule's own top margin is a line of the sheet's (none at the head of the page, where the sheet takes it off the first child): it
-		// moves to the run, so the gap the rule makes with the block above is the one it made.
 		if (next && next.tagName === 'HR') {
 			if (first.previousElementSibling || first.parentNode !== root) carriers[0].style.marginTop = 'var(--md-line)';
 			next.style.marginTop = '0';
@@ -177,18 +169,15 @@ function _rapierPrintWillPageCss() {
 		+ '}';
 }
 
-// The capture's own limits (agent/visual.mjs VISUAL_LIMITS): the longest edge and the most pixels one capture returns.
 const CAPTURE_EDGE = 4096, CAPTURE_PIXELS = 4 * 1024 * 1024;
 
-async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = false, filename, title, settings}, {capture, signal, current, maxBytes = 8 * 1024 * 1024, film, ink} = {}) {
+async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = false, filename, title, settings, templatePage = null}, {capture, signal, current, maxBytes = 8 * 1024 * 1024, film, ink} = {}) {
 	const refuse = code => { throw Object.assign(new Error(code), {code}); };
 	const check = () => {
 		if (signal?.aborted) refuse('cancelled');
 		if (current?.() === false) refuse('document_changed');
 	};
 	if (!root?.ownerDocument || typeof capture !== 'function') refuse('pdf_render_unavailable');
-	// The native printer chooses its default paper. A file made without a printer uses A4,
-	// with the same explicit document settings and default margins as the print artifact.
 	const paper = settings?.papersize || {width: 11906, height: 16838};
 	const width = paper.width / 20, height = paper.height / 20;
 	const margin = settings?.geometry ? settings.geometry.twips / 20 : null;
@@ -198,7 +187,7 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 	const bodyHeight = Math.floor((height - margins.top - margins.bottom) * 4 / 3);
 	if (bodyWidth < 1 || bodyHeight < 1 || bodyWidth * bodyHeight > 4 * 1024 * 1024) refuse('pdf_page_geometry_invalid');
 	const frame = document.createElement('iframe');
-	let vectors, nativeGlyphs, filmSession;
+	let vectors, backdrop = null, nativeGlyphs, filmSession;
 	let rejectAbort;
 	const aborted = new Promise((_, reject) => { rejectAbort = reject; });
 	aborted.catch(() => {});
@@ -221,8 +210,6 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 		check();
 		const into = frame.contentDocument;
 		if (!into?.body) refuse('pdf_render_unavailable');
-		// The capture and native printer use standards layout. An initial blank
-		// iframe otherwise measures preformatted inline struts in quirks mode.
 		into.open(); into.write('<!doctype html><html><head></head><body></body></html>'); into.close();
 		into.documentElement.setAttribute('data-rapier-theme', 'light');
 		into.documentElement.setAttribute('lang', documentLanguage(settings, document.defaultView?.navigator?.language));
@@ -246,7 +233,7 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 		strip.append(root); surface.append(strip); into.body.append(surface);
 		const main = root.querySelector('main');
 		if (!main) refuse('pdf_render_unavailable');
-		// A file for paper is light whatever the person's own scheme: the page's text takes its colour scheme from this attribute.
+		if (templatePage) for (const image of main.querySelectorAll('img')) if (image.getAttribute('src') === templatePage.url) (image.closest('p,figure') || image).remove();
 		main.setAttribute('data-md-theme', 'light');
 		for (const details of main.querySelectorAll('details')) details.open = true;
 		main.getBoundingClientRect();
@@ -254,12 +241,8 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 		await step(into.fonts.ready);
 		for (const image of main.querySelectorAll('img')) { image.loading = 'eager'; await step(image.decode()); }
 		check();
-		// The print sheet owns the words and styles. CSS fragmentation supplies the file's
-		// pages without a print dialog; its break constraints become column constraints here.
 		for (const node of [main, ...main.querySelectorAll('*')]) {
 			const style = into.defaultView.getComputedStyle(node);
-			// Clipped, out-of-flow source has no page ink. Leaving it in a multicolumn
-			// container can still create blank columns and expose its text to PDF search.
 			if (!node.closest('.rapier-will-carrier') && ['absolute', 'fixed'].includes(style.position) && /^rect\(/.test(style.clip)) {
 				const box = style.clip.slice(5, -1).trim().split(/[,\s]+/).map(Number.parseFloat);
 				if (box.length === 4 && box.every(Number.isFinite) && (box[1] <= box[3] || box[2] <= box[0])) {
@@ -273,17 +256,14 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 		}
 		for (const [name, value] of Object.entries({width: bodyWidth + 'px', height: bodyHeight + 'px',
 			'max-width': 'none', 'min-height': '0', padding: '0', margin: '0', overflow: 'visible',
-			'column-width': bodyWidth + 'px', 'column-gap': '0', 'column-count': 'auto', 'column-fill': 'auto'}))
+			'column-width': bodyWidth + 'px', 'column-gap': '0', 'column-count': 'auto', 'column-fill': 'auto', background: 'none'}))
 			main.style.setProperty(name, value, 'important');
 		globalThis.RapierMath?.fit(main);
 		applyPrintCapitals(main);
 		applyTypography(main, {all: true});
-		// The document's ink marks, drawn on the words where they now stand on the page.
 		ink?.(main);
 		const count = Math.max(1, Math.ceil((main.scrollWidth - 1) / bodyWidth));
 		if (!Number.isSafeInteger(count)) refuse('pdf_page_geometry_invalid');
-		// Captured document font bytes own a same-name family; factory glyphs must
-		// never replace their page paint. This uses the artifact, not a live catalog.
 		const customFamilies = new Set(fontCss ? fontRequests(main, {fontCss, css, settings}).map(face => face.family.toLowerCase()) : []);
 		nativeGlyphs = await collectPdfGlyphs(main, {width: bodyWidth, check,
 			faces: globalThis.RapierDocxFonts?.DOCX_FONTS?.filter(face => !customFamilies.has(face.family.toLowerCase()))});
@@ -304,11 +284,7 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 		const font = _rapierPrintWillFont(text, 500);
 		strip.style.width = count * bodyWidth + 'px';
 		strip.style.transform = 'none';
-		// One styled film is reused for every page window. Batches stay inside the capture edge
-		// and pixel limits. A picture no longer forces its own capture. Pixels are deflated once.
 		if (typeof film !== 'function') refuse('pdf_render_unavailable');
-		// step() can reject after the film resolves (the document changed, or the attempt was
-		// cancelled). The session would otherwise keep its abort listener with no owner.
 		let pendingFilm = null, abandonFilm = false;
 		const trackedFilm = Promise.resolve(film({root: surface, signal, current, frame: {width: count * bodyWidth, height: bodyHeight}})).then(session => {
 			if (abandonFilm) session?.close?.();
@@ -359,15 +335,24 @@ async function _rapierWritePrintPdf({root, css, fontCss = '', canonical, bom = f
 			first += span;
 		}
 		check();
-		const bytes = await writeRasterPdf({pages, textPages, linkPages, vectorPages: vectors.pages, canonical, bom, filename, title, font, width, height, margins,
+		if (templatePage) {
+			const sheetWidth = Math.round(width * 4 / 3), sheetHeight = Math.round(height * 4 / 3);
+			const holder = into.createElement('div'), picture = into.createElement('img');
+			holder.style.cssText = `position:fixed;left:0;top:0;width:${sheetWidth}px;height:${sheetHeight}px;pointer-events:none`;
+			picture.style.cssText = 'display:block;width:100%;height:100%'; picture.src = templatePage.url;
+			holder.append(picture); into.body.append(holder);
+			await step(picture.decode());
+			backdrop = await collectPdfVectors(holder, {width: sheetWidth, height: sheetHeight, count: 1, capture, signal, current, check, maxBytes,
+				sanitizeSvg: globalThis.RapierImageAssets?.sanitizeSvgText});
+			if (!backdrop.pages[0].length) refuse('pdf_template_border_unavailable');
+		}
+		const bytes = await writeRasterPdf({pages, textPages, linkPages, vectorPages: vectors.pages, canonical, bom, filename, title, font, width, height, margins, backdrop: backdrop?.pages[0] || null,
 			pixelWidth: bodyWidth, pixelHeight: bodyHeight, pageNumbers: settings?.pagestyle === 'plain', maxBytes, check});
 		check();
 		return {bytes, pages: count};
-	} finally { filmSession?.close(); nativeGlyphs?.restore(); vectors?.restore(); signal?.removeEventListener('abort', abort); frame.remove(); }
+	} finally { filmSession?.close(); nativeGlyphs?.restore(); vectors?.restore(); backdrop?.restore(); signal?.removeEventListener('abort', abort); frame.remove(); }
 }
 
-// Ranges locate the actual text fragments after column layout. Splitting a text node at
-// its line and page boundaries keeps a PDF search on the page that carries those words.
 async function _rapierPrintTextPages(main, pageWidth, count, check, origin = main.getBoundingClientRect(), nativeNodes = null) {
 	const into = main.ownerDocument;
 	const pages = Array.from({length: count}, () => []), walker = into.createTreeWalker(main, 5);
@@ -382,8 +367,6 @@ async function _rapierPrintTextPages(main, pageWidth, count, check, origin = mai
 		if (node.nodeType === 1 && !formula || node.nodeType === 3 && element.closest('.math-rendered[data-math-src]')) continue;
 		let value;
 		if (formula) {
-			// The math owner paints SVG paths and retains the equation's exact TeX alternative.
-			// Carry that semantic text at the SVG's measured position for search and copy.
 			value = formula.getAttribute('aria-label') || '';
 		} else value = node.data;
 		const style = into.defaultView.getComputedStyle(element);
@@ -397,14 +380,11 @@ async function _rapierPrintTextPages(main, pageWidth, count, check, origin = mai
 		const visit = (start, end) => {
 			if (++operations % 128 === 0) check();
 			if (!formula) { range.setStart(node, start); range.setEnd(node, end); }
-			// A tiny carrier font can have its advance rounded to zero by a browser. Its
-			// measured baseline still locates the marker; the font's authored metrics keep its text width.
 			const rects = Array.from(formula ? [formula.getBoundingClientRect()] : range.getClientRects()).filter(rect => rect.height > 0 && (rect.width > 0 || carrier));
 			if (!rects.length) { if (carrier) markerLost(); return; }
 			const first = rects[0], page = pageOf(first.left);
 			const oneLine = rects.every(rect => pageOf(rect.left) === page && pageOf(rect.width ? rect.right - .02 : rect.left) === page && Math.abs(rect.top - first.top) < .5);
 			let middle = formula ? start : Math.floor((start + end) / 2);
-			// A line never breaks inside a character: halve between whole ones (a letter and its marks, a conjunct, an emoji).
 			if (middle > start) middle = (characters ||= segmenter.segment(value)).containing(middle).index;
 			const cluster = native?.find(item => item.start < middle && item.end > middle);
 			if (cluster) middle = cluster.start > start ? cluster.start : cluster.end;
@@ -436,9 +416,6 @@ async function _rapierPrintTextPages(main, pageWidth, count, check, origin = mai
 		visit(0, value.length);
 	}
 	check();
-	// Trailing-carrier print CSS pulls a last marker up so it stays on the sheet.
-	// Keep that paint; put the searchable object after the preceding words so
-	// reading order matches the source (the marker itself has no visible ink).
 	const pageHeight = origin.height;
 	for (const page of pages) {
 		let prev = null;
@@ -454,8 +431,6 @@ async function _rapierPrintTextPages(main, pageWidth, count, check, origin = mai
 			prev = run;
 		}
 	}
-	// Each run carries the tags of its block's structure (outermost first): a PDF reader finds the document's headings, lists and
-	// tables from them rather than from positions.
 	const ROLES = {TABLE: 'Table', TR: 'TR', TD: 'TD', TH: 'TH', UL: 'L', OL: 'L', LI: 'LI', BLOCKQUOTE: 'BlockQuote', P: 'P', PRE: 'Code', FIGCAPTION: 'Caption',
 		H1: 'H1', H2: 'H2', H3: 'H3', H4: 'H4', H5: 'H5', H6: 'H6'};
 	const tags = block => {
@@ -498,8 +473,6 @@ function _rapierPrintLinkPages(main, width, height, count) {
   return {_rapierWillSentinelPlan, _rapierPrintWillAdmit, _rapierPrintWillFont, _rapierPrintWillMetrics, _rapierPrintWillPlace, _rapierPrintWillLost, _rapierPrintWillPageCss, _rapierWritePrintPdf};
 }
 
-// Where each scalar of an invisible text run stands, in character slots from the run's left edge: a mark shares its letter's
-// slot, a right-to-left word runs from the right, and so do the words of a right-to-left line (strong letters only; one line).
 function _rapierPrintSlots(text, rtl) {
 	const pieces = [];
 	let scalars = 0, count = 0;
@@ -520,9 +493,7 @@ function _rapierPrintSlots(text, rtl) {
 	return {slots, count};
 }
 
-// A standard PDF page tree, image streams, Unicode text and an associated source file.
-// It accepts already-rendered pages, so there is no second document grammar or layout engine.
-async function writeRasterPdf({pages, textPages, linkPages = [], vectorPages = [], canonical, bom = false, filename, title, font, width, height, margins, pixelWidth, pixelHeight, pageNumbers, maxBytes, check}) {
+async function writeRasterPdf({pages, textPages, linkPages = [], vectorPages = [], backdrop = null, canonical, bom = false, filename, title, font, width, height, margins, pixelWidth, pixelHeight, pageNumbers, maxBytes, check}) {
 	const encoder = new TextEncoder(), encode = value => encoder.encode(value);
 	const fail = code => { throw Object.assign(new Error(code), {code}); };
 	const concat = parts => {
@@ -589,7 +560,6 @@ async function writeRasterPdf({pages, textPages, linkPages = [], vectorPages = [
 		nativeResources.push(entry.name + ' ' + type + ' 0 R');
 	}
 	const sharedImages = new Map();
-	// The structure tree (PDF 1.7, 14.7): one element per tagged DOM element, each run marked with its page's next MCID.
 	const structs = new Map(), documentElement = {role: 'Document', kids: []}, parentTree = [];
 	const structOf = tags => {
 		let parent = documentElement;
@@ -607,12 +577,10 @@ async function writeRasterPdf({pages, textPages, linkPages = [], vectorPages = [
 		const held = pages[index], flate = !!held?.rgb;
 		const picture = stream('/Type /XObject /Subtype /Image /Width ' + pixelWidth + ' /Height ' + pixelHeight + ' /ColorSpace /DeviceRGB /BitsPerComponent 8 /Filter /' + (flate ? 'FlateDecode' : 'DCTDecode'), flate ? held.rgb : held);
 		pages[index] = null;
-		// The page's picture is how its tagged text looks, not a figure of its own: an artifact.
-		let commands = '/Artifact BMC\nq\n' + [decimal(pixelWidth * .75), 0, 0, decimal(pixelHeight * .75), decimal(margins.left), decimal(height - margins.top - pixelHeight * .75)].join(' ') + ' cm\n/Im0 Do\nQ\nEMC\n';
 		const pictures = ['/Im0 ' + picture + ' 0 R'], states = [];
-		for (const drawing of vectorPages[index] || []) {
-			commands += `q\n.75 0 0 -.75 ${decimal(margins.left)} ${decimal(height - margins.top)} cm\n0 0 ${pixelWidth} ${pixelHeight} re W n\n1 0 0 1 ${decimal(drawing.x)} 0 cm\n`;
-			commands += [drawing.clip.x, drawing.clip.y, drawing.clip.width, drawing.clip.height].map(decimal).join(' ') + ' re W n\n';
+		const emit = (drawing, left, top, clipWidth, clipHeight) => {
+			let out = `q\n.75 0 0 -.75 ${decimal(left)} ${decimal(top)} cm\n0 0 ${clipWidth} ${clipHeight} re W n\n1 0 0 1 ${decimal(drawing.x)} 0 cm\n`;
+			out += [drawing.clip.x, drawing.clip.y, drawing.clip.width, drawing.clip.height].map(decimal).join(' ') + ' re W n\n';
 			for (const piece of drawing.sequence) {
 				if (piece.image) {
 					let shared = sharedImages.get(piece.image);
@@ -626,21 +594,22 @@ async function writeRasterPdf({pages, textPages, linkPages = [], vectorPages = [
 						sharedImages.set(piece.image, shared);
 					}
 					if (!pictures.includes(shared.resource)) pictures.push(shared.resource);
-					commands += `q\n${[piece.width, 0, 0, -piece.height, piece.x, piece.y + piece.height].map(decimal).join(' ')} cm\n${shared.name} Do\nQ\n`;
+					out += `q\n${[piece.width, 0, 0, -piece.height, piece.x, piece.y + piece.height].map(decimal).join(' ')} cm\n${shared.name} Do\nQ\n`;
 				} else {
 					const state = '/Gs' + states.length;
 					states.push(state + ' ' + add('<< /Type /ExtGState /ca ' + decimal(piece.fillAlpha) + ' /CA ' + decimal(piece.strokeAlpha) + ' >>') + ' 0 R');
-					commands += 'q\n' + state + ' gs\n' + piece.commands + 'Q\n';
+					out += 'q\n' + state + ' gs\n' + piece.commands + 'Q\n';
 				}
 			}
-			commands += 'Q\n';
-		}
+			return out + 'Q\n';
+		};
+		let commands = '';
+		for (const drawing of backdrop || []) commands += emit(drawing, 0, height, Math.round(width * 4 / 3), Math.round(height * 4 / 3));
+		commands += '/Artifact BMC\nq\n' + [decimal(pixelWidth * .75), 0, 0, decimal(pixelHeight * .75), decimal(margins.left), decimal(height - margins.top - pixelHeight * .75)].join(' ') + ' cm\n/Im0 Do\nQ\nEMC\n';
+		for (const drawing of vectorPages[index] || []) commands += emit(drawing, margins.left, height - margins.top, pixelWidth, pixelHeight);
 		const textAlphas = new Map();
 		for (const run of textPages[index]) {
 			if (!run.text) continue;
-			// Keep one logical Unicode text object, resetting its font before each scalar's direction
-			// can change. Readers that ignore ActualText otherwise discard earlier words in mixed RTL/LTR.
-			// Express width in the font size: some readers ignore horizontal text scaling when joining.
 			const element = structOf(run.tags || []);
 			if (element) {element.kids.push({page: index, mcid: marked.length}); commands += `/${element.role} << /MCID ${marked.length} >> BDC\n`; marked.push(element);}
 			commands += `q\n/Span << /ActualText <feff${hex(run.text)}> >> BDC\nBT\n100 Tz\n`;
@@ -683,7 +652,7 @@ async function writeRasterPdf({pages, textPages, linkPages = [], vectorPages = [
 	const parents = add('<< /Nums [' + parentTree.map((marked, index) => index + ' [' + marked.map(element => ids.get(element) + ' 0 R').join(' ') + ']').join(' ') + '] >>');
 	put(tree, '<< /Type /StructTreeRoot /K [' + ids.get(documentElement) + ' 0 R] /ParentTree ' + parents + ' 0 R /ParentTreeNextKey ' + parentTree.length + ' >>');
 	put(1, '<< /Type /Catalog /Pages 2 0 R /MarkInfo << /Marked true >> /StructTreeRoot ' + tree + ' 0 R /Names << /EmbeddedFiles << /Names [<feff' + hex(filename) + '> ' + file + ' 0 R] >> >> /AF [' + file + ' 0 R] >>');
-	const info = add('<< /Producer (Rapier) /Title <feff' + hex(title || filename.replace(/\.[^.]*$/, '')) + '> >>');
+	const info = add('<< /Producer (Rapier) /Creator (Rapier) /Title <feff' + hex(title || filename.replace(/\.[^.]*$/, '')) + '> >>');
 	const header = encode('%PDF-1.7\n%\u00e2\u00e3\u00cf\u00d3\n'), offsets = new Array(objects.length);
 	let cursor = header.length;
 	for (let id = 1; id < objects.length; id++) {

package/dist/kit/render-sanitize.mjs

@@ -1,10 +1,7 @@
 // SPDX-License-Identifier: MIT
 import {typography} from '../spec/md-marks.mjs';
 const RAPIER_SANITIZE_FORBID_TAGS = Object.freeze(['form', 'style', 'dialog', 'template', 'iframe']);
-// Invokers (DOMPurify 3.4.16 allows them) act on another element by id, and a document shares the page with the editor's own
-// controls: a document's button could open or close them.
 const RAPIER_SANITIZE_FORBID_ATTR = Object.freeze(['action', 'formaction', 'form', 'for', 'command', 'commandfor', 'popover', 'popovertarget', 'popovertargetaction']);
-// The document profiles use one allowlist in a browser and in an inert server parser.
 const RAPIER_SANITIZE_PROFILES = Object.freeze({
 	htmlTags: Object.freeze(["a","abbr","acronym","address","area","article","aside","audio","b","bdi","bdo","big","blink","blockquote","body","br","button","canvas","caption","center","cite","code","col","colgroup","content","data","datalist","dd","decorator","del","details","dfn","dialog","dir","div","dl","dt","element","em","fieldset","figcaption","figure","font","footer","form","h1","h2","h3","h4","h5","h6","head","header","hgroup","hr","html","i","img","input","ins","kbd","label","legend","li","main","map","mark","marquee","menu","menuitem","meter","nav","nobr","ol","optgroup","option","output","p","picture","pre","progress","q","rp","rt","ruby","s","samp","search","section","select","shadow","slot","small","source","spacer","span","strike","strong","style","sub","summary","sup","table","tbody","td","template","textarea","tfoot","th","thead","time","tr","track","tt","u","ul","var","video","wbr"]),
 	svgTags: Object.freeze(["svg","a","altglyph","altglyphdef","altglyphitem","animatecolor","animatemotion","animatetransform","circle","clippath","defs","desc","ellipse","enterkeyhint","exportparts","filter","font","g","glyph","glyphref","hkern","image","inputmode","line","lineargradient","marker","mask","metadata","mpath","part","path","pattern","polygon","polyline","radialgradient","rect","stop","style","switch","symbol","text","textpath","title","tref","tspan","view","vkern"]),
@@ -56,7 +53,6 @@ const DIAGRAM_LABEL_SVG_NAMESPACE = 'http://www.w3.org/2000/svg';
 const DIAGRAM_LABEL_HTML_NAMESPACE = 'http://www.w3.org/1999/xhtml';
 const DIAGRAM_LABEL_MATH_NAMESPACE = 'http://www.w3.org/1998/Math/MathML';
 
-// The profile each DOMPurify instance holds (setConfig), shared by every renderer made over it.
 const purifyConfigured = new WeakMap();
 
 function _rapierImageDataAttributeAllowed(tag, attribute, value, verify) {
@@ -64,9 +60,6 @@ function _rapierImageDataAttributeAllowed(tag, attribute, value, verify) {
 		&& /^data:image\/(?:png|jpeg|webp|jxl);base64,/i.test(value)) && verify(value);
 }
 
-// Snapshot current math text before a Markdown writer collapses whitespace. Rendered SVG
-// glyphs have no textContent: put their retained source in this inert copy so the writer's
-// blank-node rule cannot discard an equation or its containing paragraph/list item.
 function prepareMathSource(input, DOMParser) {
 	const html = typeof input === 'string';
 	const MARK = 'span[data-rapier-math-source],span.math-rendered[data-math-src]';
@@ -82,18 +75,17 @@ function prepareMathSource(input, DOMParser) {
 			try { node.textContent = decodeURIComponent(node.getAttribute('data-math-src')); } catch (_) {}
 			continue;
 		}
+		try {
+			const held = node.getAttribute('data-rapier-math-source'), source = decodeURIComponent(held);
+			if (held && (source === node.textContent || source.replace(/\r\n?/g, '\n').replace(/\0/g, '') === node.textContent)) continue;
+		} catch (_) {}
 		try { node.setAttribute('data-rapier-math-source', encodeURIComponent(node.textContent || '')); }
 		catch (_) { node.removeAttribute('data-rapier-math-source'); }
 	}
 	return sourceRoot ? root.outerHTML : root;
 }
 
-// An inert parser and a string sanitizer use the same profile as DOMPurify. Neither port may
-// execute document scripts or fetch resources. HTML5 parsing precedes filtering so foreign
-// content, entities and malformed table markup cross the same boundary as a browser's parser.
 function createInertHtmlSanitizer({parseFragment, serialize, sanitizeHtml, safeRasterDataUrl, chromeOwnsId = id => /^rapier-(?!anchor-)/.test(id)}) {
-	// chromeOwnsId defaults to _rapierChromeOwnsId's prefix rule: every page Rapier writes carries rapier- elements
-	// (rapier-document, rapier-ledger). The factories are serialized into written pages, so neither reads a module constant.
 	if (![parseFragment, serialize, sanitizeHtml, safeRasterDataUrl, chromeOwnsId].every(port => typeof port === 'function'))
 		throw new TypeError('The inert HTML sanitizer requires every parser and filtering port');
 	const uri = /^(?:(?:(?:f|ht)tps?|mailto|tel|callto|sms|cid|xmpp|matrix):|[^a-z]|[a-z+.\-]+(?:[^a-z+.\-:]|$))/i;
@@ -103,8 +95,6 @@ function createInertHtmlSanitizer({parseFragment, serialize, sanitizeHtml, safeR
 	const svgTags = new Set(RAPIER_SANITIZE_PROFILES.svgTags), mathTags = new Set(RAPIER_SANITIZE_PROFILES.mathTags);
 	const mathIntegration = new Set(['mi', 'mo', 'mn', 'ms', 'mtext']), htmlIntegration = new Set(['annotation-xml']);
 	const sharedHtmlSvg = new Set(['title', 'style', 'font', 'a', 'script']);
-	// DOMPurify's default namespace admission applies in addition to its tag profile. In
-	// particular a MathML name parsed as HTML, or HTML inside SVG desc, is not admitted.
 	const validNamespace = (node, parent) => {
 		const tag = node.tagName.toLowerCase(), parentTag = (parent.tagName || 'template').toLowerCase();
 		const namespace = node.namespaceURI, parentNamespace = parent.namespaceURI || htmlNamespace;
@@ -158,6 +148,7 @@ function createInertHtmlSanitizer({parseFragment, serialize, sanitizeHtml, safeR
 					const key = name.toLowerCase();
 					const value = name === 'value' ? inputValue : inputValue.trim();
 					if (forbiddenAttributes.has(key) || key.startsWith('on')) continue;
+					if (key === 'id' && chromeOwnsId(value)) continue;
 					if (/((--!?|])>)|<\/(style|script|title|xmp|textarea|noscript|iframe|noembed|noframes)/i.test(value)
 						|| key === 'attributename' && value.includes('href')) continue;
 					const data = /^data-[\-\w.\u00b7-\uffff]+$/.test(key), aria = /^aria-[\-\w]+$/.test(key);
@@ -179,13 +170,9 @@ function createInertHtmlSanitizer({parseFragment, serialize, sanitizeHtml, safeR
 	};
 }
 
-// The document's renderer. DOM, codecs and host state are explicit inputs.
 function createRenderSanitizer(runtime) {
   const {CSSStyleSheet, DOMPurify, RAPIER_RASTER_DATA_URL_RE, URL, _rapierCssDeclaration, _rapierDropRemoteDeclarations, _rapierRemoteContent, _rapierRuleDescriptorIsRemote, _rapierSanitizeRuntime, document, globalThis, location} = runtime;
 
-// Only the diagram reader and its styled export admit static foreignObject labels. The
-// ordinary SVG reader still rejects HTML subdocuments. The export class selects scope;
-// every descendant is checked again, so a copied class never confers authority.
 function diagramLabelRoot(node) {
 	for (let current = node; current?.nodeType === 1; current = current.parentElement) {
 		if (current.localName?.toLowerCase() === 'foreignobject') return current;
@@ -205,7 +192,6 @@ function diagramLabelStyle(property, value) {
 	declaration.setProperty(property, value);
 	const parsed = declaration.getPropertyValue(property);
 	if (!parsed || /(?:url|image-set|element|paint|attr)\s*\(/i.test(parsed)) return false;
-	// SVG paint accepts resource references; a custom property must not smuggle one in.
 	return !/^(?:fill|stroke)$/.test(property) || !/var\s*\(/i.test(parsed);
 }
 
@@ -279,14 +265,6 @@ function _rapierInstallSanitizeHooks() {
 			&& _rapierImageDataAttributeAllowed(tag, data.attrName.toLowerCase(), compact, _rapierSafeRasterDataUrl);
 		if (!rasterAllowed) data.keepAttr = false;
 	});
-	// Paste, render and export sanitise: a remote <img src> would fetch the instant its markup lands in the page,
-	// so the src the hook below is about to strip never survives. Move it to a data-* attribute first
-	// (DOMPurify already leaves data-* alone, and the sanitizer's own inert document never fetches it
-	// either), so the picture and its alt text still reach Turndown's rapierImage rule as a live
-	// pasted-page picture and write out `![alt](url)` -- not vanish -- and an <img> in a document's own
-	// HTML (a table kept as HTML) is written again with its URL: the raw profile admits no data-*, so
-	// only the three this hook sets on the node it holds back stay. A picture inside a link is the same
-	// node with the same fix: the outer <a> rule reads this img's own converted markdown as its content.
 	DOMPurify.addHook('uponSanitizeElement', node => {
 		if (_rapierRemoteContent.allowed || !/^(?:paste|render|raw|export)$/.test(_rapierSanitizeRuntime.context)) return;
 		if (!node || node.nodeName !== 'IMG' || node.hasAttribute('data-rapier-remote-src')) return;
@@ -308,18 +286,12 @@ function _rapierInstallSanitizeHooks() {
 			node.setAttribute('data-rapier-remote-src', href);
 		}
 	});
-	// A list from Word is named by the style of its paragraph and of the span that holds its marker (mso-list,
-	// spec/html-reading.mjs), and the CSS pass below empties a style of every property the browser does not
-	// know: a marker's style is only `mso-list:Ignore`. The paste carries both as data-* first, where they stay,
-	// for the conversion that reads them (_rapierPasteWordLists).
 	DOMPurify.addHook('uponSanitizeElement', node => {
 		if (_rapierSanitizeRuntime.context !== 'paste' || !node || !/^(?:P|LI|H[1-6]|SPAN)$/.test(node.nodeName)) return;
-		// Only the style names them: a page cannot bring these attributes of its own.
 		node.removeAttribute('data-rapier-word-level');
 		node.removeAttribute('data-rapier-word-marker');
 		const style = node.getAttribute('style');
 		if (!style || !/mso-list/i.test(style)) return;
-		// Word's list reading is the editor's (AGPL) grammar; a host without it keeps the paste's paragraphs as they are.
 		const spec = globalThis.RapierMarkdownSpec;
 		if (!spec) return;
 		const {wordListLevel, wordListIsMarker} = spec;
@@ -334,9 +306,7 @@ function _rapierInstallSanitizeHooks() {
 			data.keepAttr = !!data.attrValue;
 			return;
 		}
-		// A document's own HTML never carries an id the chrome owns, remote content allowed or not: the
-		// allow is about what the page fetches, and this rule is about what the page's chrome finds.
-		if (String(data.attrName || '').toLowerCase() === 'id' && _rapierSanitizeRuntime.context === 'raw' && _rapierChromeOwnsId(value)) { data.keepAttr = false; return; }
+		if (String(data.attrName || '').toLowerCase() === 'id' && _rapierChromeOwnsId(value)) { data.keepAttr = false; return; }
 		if (_rapierRemoteContent.allowed || _rapierSanitizeRuntime.context === 'source') return;
 		if (String(data.attrName || '').toLowerCase() === 'style') {
 			const cleaned = _rapierStyleWithoutRemoteUrls(value);
@@ -344,7 +314,6 @@ function _rapierInstallSanitizeHooks() {
 			return;
 		}
 		const attribute = String(data.attrName || '').toLowerCase();
-		// These attributes carry authored text, never resource requests.
 		if (attribute === 'alt' || attribute === 'title') return;
 		if (/^data-rapier-remote-(?:src|alt|title)$/.test(attribute) && node && node._rapierHeldBack === true) { data.forceKeepAttr = true; return; }
 		if (attribute.startsWith('data-')) return;
@@ -363,9 +332,6 @@ function _rapierInstallSanitizeHooks() {
 			data.forceKeepAttr = true;
 		}
 	});
-	// A link that opens a new tab never hands that tab a window.opener or a Referer: the editor's own
-	// click router already opens with noopener,noreferrer; the sanitized markup that Share and the
-	// standalone export carry gets the same rule in its bytes.
 	DOMPurify.addHook('afterSanitizeAttributes', node => {
 		if (!node || !/^(?:A|AREA)$/i.test(node.nodeName) || !node.hasAttribute('target')) return;
 		if (String(node.getAttribute('target') || '').trim().toLowerCase() !== '_blank') return;
@@ -387,25 +353,17 @@ function sanitizeRapierHtml(value, context = 'render') {
 	return sanitize(value, context, false);
 }
 
-// The same parser, profile and hooks admit both results. A private renderer can
-// adopt the sanitized nodes without serializing and parsing them again.
 function sanitizeRapierFragment(value, context = 'render') {
 	return sanitize(value, context, true);
 }
 
 function sanitize(value, context, fragment) {
-	// 'export' is the render profile under the export hooks; 'raw' is a document's own HTML (initMarkdownIt, html_block):
-	// the render profile without data-* attributes, which Rapier's delegated handlers read as their own controls, and
-	// (the hook) without an id the chrome owns. 'source' is Markdown source HTML, written and never shown: the render
-	// profile's admission, and the block on remote content stays the reader's (a remote picture's URL is the person's words).
 	const selected = context === 'raw' || context === 'export' || context === 'source' || RAPIER_SANITIZE_OPTIONS[context] ? context : 'render';
 	const labels = selected === 'diagram' || selected === 'export', key = selected + (fragment ? ' fragment' : '');
 	let options = purifyOptions.get(key);
 	if (!options) {
 		options = selected === 'raw' ? { ...RAPIER_SANITIZE_OPTIONS.render, ALLOW_DATA_ATTR: false } : RAPIER_SANITIZE_OPTIONS[selected] || RAPIER_SANITIZE_OPTIONS.render;
 		if (selected === 'source') options = {...options, ADD_DATA_URI_TAGS: ['image']};
-		// DOMPurify's USE_PROFILES replaces explicit allowlists. Add MathML to the
-		// selected profile without discarding its admitted HTML and SVG vocabulary.
 		if (labels) options = {...options, ALLOWED_TAGS: [...new Set([...options.ALLOWED_TAGS, ...RAPIER_SANITIZE_PROFILES.mathTags])],
 			ALLOWED_ATTR: [...new Set([...options.ALLOWED_ATTR, ...RAPIER_SANITIZE_PROFILES.mathAttributes])], ADD_TAGS: ['foreignobject'], ADD_DATA_URI_TAGS: ['image'],
 			HTML_INTEGRATION_POINTS: {foreignobject: true}, FORBID_TAGS: options.FORBID_TAGS.filter(tag => tag !== 'foreignobject')};
@@ -417,8 +375,6 @@ function sanitize(value, context, fragment) {
 		DOMPurify.addHook('uponSanitizeAttribute', diagramLabelAttribute);
 		DOMPurify.addHook('afterSanitizeAttributes', diagramLabelContained);
 	}
-	// DOMPurify reparses a per-call configuration on every call (its cache compares its own copy), about 0.45 ms
-	// each in Chrome 155. A profile is set once and kept until another profile is asked for; every caller comes here.
 	if (purifyConfigured.get(DOMPurify) !== options) { DOMPurify.setConfig(options); purifyConfigured.set(DOMPurify, options); }
 	const previous = _rapierSanitizeRuntime.context;
 	_rapierSanitizeRuntime.context = selected;
@@ -444,15 +400,10 @@ function escapeRapierHtmlText(value) {
 }
 
 function _rapierChromeOwnsId(id) {
-	// Every id the shell's own markup carries, read once before any document renders (initMarkdownIt primes it),
-	// and every rapier- id: the chrome makes its dialogs, overlays, parts and plug-in list later, under that prefix
-	// (rapier-anchor- is the renderer's own name for a repeated document anchor, and stays the document's).
 	if (!_rapierSanitizeRuntime.chromeIds) _rapierSanitizeRuntime.chromeIds = new Set(Array.from(typeof document?.querySelectorAll === 'function' ? document.querySelectorAll('[id]') : [], element => element.id));
 	return _rapierSanitizeRuntime.chromeIds.has(String(id)) || /^rapier-(?!anchor-)/.test(id);
 }
 
-// A link whose words or address carry a bidi control, or whose host is spelled in punycode, can show a destination it does
-// not have. Opening one names its real address first; the answer is that address, or '' when the link opens at once.
 const RAPIER_BIDI_CONTROL = /[\u061c\u200e\u200f\u202a-\u202e\u2066-\u2069]/;
 function _rapierLinkDestinationToConfirm(href, text = '') {
 	let url, address = String(href || '');
@@ -484,9 +435,6 @@ function _rapierCssPresentationIsRemote(property, value) {
 	return declaration.length ? _rapierDropRemoteDeclarations(declaration) : false;
 }
 
-// A reference with no scheme and no host of its own ("photo.png", "#top", each word of a class) resolves on the base's scheme and
-// host whatever its path, so its answer is the base's, found once for each base and page: a long document's render asked
-// this of every token of every attribute, a URL built each time.
 const relativeReference = {base: null, here: null, origin: ''};
 function _rapierRemoteSubresourceOrigin(value) {
 	const raw = String(value == null ? '' : value)
@@ -544,7 +492,7 @@ function _rapierDropRemoteRules(rules, owner) {
 }
 
 function _rapierVerifyRasterBytes(bytes, mime) {
-	if (mime === 'image/jxl') return globalThis.RapierImageAssets?.isJxl(bytes) === true; // without the editor's codec a JPEG XL picture is refused
+	if (mime === 'image/jxl') return globalThis.RapierImageAssets?.isJxl(bytes) === true;
 
 	// declaration, doctype or generator/license comment before <svg itself, of whatever length its
 

package/dist/kit/render-styles.mjs

@@ -1,12 +1,8 @@
 // SPDX-License-Identifier: MIT
-// The document's renderer. DOM, codecs and host state are explicit inputs.
 function createRenderStyles(runtime) {
   const {_rapierArtifactAccent, _rapierStyleText} = runtime;
 function _rapierArtifactThemeCss(houseDefaults = false) {
-	// Geometry and the selected accent belong to this page; palette and type belong to the
 	// MIT reference sheet. Static code highlighting reads the app token names.
-	// The live editor resolves System to Light before applying the reference heading tokens;
-	// a carried System page keeps following the reader's device, including those light tokens.
 	const accent = houseDefaults ? '' : ':root,.md-render{--md-color-accent:' + _rapierArtifactAccent() + '}';
 	return `
 ${accent}
@@ -82,7 +78,6 @@ function _rapierArtifactStyles(theme, includeFonts, printMode, houseDefaults = f
 	else if (includeFonts) parts.push(_rapierStyleText('rapier-font-style'));
 	parts.push(_rapierStyleText('rapier-content-style'));
 	parts.push(_rapierArtifactThemeCss(houseDefaults));
-	// Local faces keep an offline fallback. The HTML writer may append a small factory Sans subset.
 	if (!includeFonts && fontCss === null) parts.push(`@font-face{font-family:'Geist/Roboto';src:local('Roboto');font-weight:400 500;size-adjust:105%;ascent-override:95.7%;descent-override:28.1%;line-gap-override:0%}
 @font-face{font-family:'Geist/Roboto';src:local('Roboto Bold');font-weight:600 700;size-adjust:105%;ascent-override:95.7%;descent-override:28.1%;line-gap-override:0%}
 @font-face{font-family:'Geist/Helvetica Neue';src:local('Helvetica Neue');font-weight:400 500;size-adjust:103.8%;ascent-override:96.8%;descent-override:28.4%;line-gap-override:0%}
@@ -102,8 +97,6 @@ function _rapierArtifactStyles(theme, includeFonts, printMode, houseDefaults = f
 :root,.md-render{--md-font-sans:'Geist','Geist/Roboto','Geist/Helvetica Neue','Geist/Segoe UI','Geist/Arial','Geist/Noto Sans','Geist/Ubuntu','Geist/Liberation Sans','Geist/DejaVu Sans',system-ui,sans-serif}`);
 	const highlightCss = _rapierStyleText('rapier-highlight-style');
 	parts.push(highlightCss);
-	// A page written for the system theme has no body.light to switch: under a light system
-	// setting each palette takes its light values from the sheet's own body.light rule.
 	if (theme === 'system') for (const [, palette, light] of highlightCss.matchAll(/html\[data-highlights="(standard|accent)"\] body\.light\s*\{([^}]*)\}/g))
 		parts.push('@media (prefers-color-scheme:light){html[data-rapier-theme="system"][data-highlights="' + palette + '"] body{' + light.trim() + '}}');
 	if (printMode) parts.push(_rapierArtifactPrintCss());

package/dist/kit/render-work.mjs

@@ -1,5 +1,4 @@
 // SPDX-License-Identifier: MIT
-// The synchronous and cooperative writers consume the same document transformations.
 export function finish(iterator) {
   let next;
   do { next = iterator.next(); } while (!next.done);
@@ -19,8 +18,6 @@ export async function pause(work, fraction, label) {
   check(work);
 }
 
-// This locates complete fragments; the existing HTML parser and sanitizer still admit
-// every byte. Uncertain nesting stays together instead of guessing a new boundary.
 export function htmlBoundary() {
   const stack = [], voids = new Set('area base br col embed hr img input link meta param source track wbr'.split(' '));
   const rawTags = new Set('script style textarea title xmp iframe noembed noframes'.split(' '));
@@ -101,9 +98,6 @@ export function* cloneTree(root) {
   return copy;
 }
 
-// Null entries are traversal checkpoints, never fragment boundaries. A large
-// nested unit is serialized cooperatively but is still admitted as one complete
-// HTML/namespace unit by the synchronous parser and sanitizer.
 export function* htmlParts(root, limit = 16384, consume = false) {
   limit = Math.max(2, limit);
   const text = root.ownerDocument.createElement('div');
@@ -135,11 +129,7 @@ export function* htmlParts(root, limit = 16384, consume = false) {
       if (at.nodeType === 1) {
         if (small(at)) pieces.push(at.outerHTML);
         else {
-          // Native shallow serialization owns tag/attribute spelling and
-          // escaping. Native text serialization below keeps its parent context.
           const empty = at.cloneNode(false);
-          // A foreign element with children must retain its native closing tag
-          // even in an inert DOM that self-closes an empty SVG element.
           if (at.firstChild) empty.appendChild(root.ownerDocument.createTextNode(''));
           const shell = empty.outerHTML, close = /<\/[^<>]+>$/.exec(shell)?.[0] || '';
           if (close && at.firstChild) {
@@ -182,7 +172,6 @@ export function* htmlParts(root, limit = 16384, consume = false) {
 export function admitHtml(root, html, sanitize, fragment = null, context = 'export') {
   if (fragment) root.appendChild(fragment(html, context));
   else {
-    // Inert string-only hosts retain their explicitly supported admission path.
     const part = root.ownerDocument.createElement('div');
     part.innerHTML = sanitize(html, context);
     while (part.firstChild) root.appendChild(part.firstChild);
@@ -208,7 +197,6 @@ export function* serializeTree(root, sanitize = null, trim = false) {
   return trim ? result.trim() : result;
 }
 
-// Chunk boundaries never split a UTF-16 pair; TextEncoder remains the byte authority.
 export function* encodeUtf8Steps(source, size = 65536) {
   const encoder = new TextEncoder(), parts = [];
   let length = 0;