reamkit 1.29.0 → 1.30.0

package/dist/esm/pdf-reader/struct-tree.js

@@ -23,17 +23,17 @@ function readStructTree(file) {
 		return pg instanceof Map ? pageMap.get(pg) : void 0;
 	};
 	const seen = /* @__PURE__ */ new Set();
+	const typeOf = roleResolver(file, stRoot);
 	const read = (value, parentPage) => {
 		const elem = file.resolve(value);
 		if (!(elem instanceof Map) || seen.has(elem) || seen.size > MAX_NODES) return void 0;
 		seen.add(elem);
 		const ownPage = pageIndexOf(elem.get("Pg") ?? PDF_NULL) ?? parentPage;
-		const mcids = [];
-		const children = [];
+		const content = [];
 		for (const kid of kidList(file, elem.get("K"))) {
 			const rk = file.resolve(kid);
 			if (typeof rk === "number") {
-				if (ownPage >= 0) mcids.push({
+				if (ownPage >= 0) content.push({
 					page: ownPage,
 					mcid: rk
 				});
@@ -42,22 +42,24 @@ function readStructTree(file) {
 				if (kind === "MCR") {
 					const m = rk.get("MCID");
 					const page = pageIndexOf(rk.get("Pg") ?? PDF_NULL) ?? ownPage;
-					if (typeof m === "number" && page >= 0) mcids.push({
+					if (typeof m === "number" && page >= 0) content.push({
 						page,
 						mcid: m
 					});
 				} else if (kind === "OBJR") {} else {
 					const child = read(rk, ownPage);
-					if (child) children.push(child);
+					if (!child) continue;
+					if (inline(file, rk, child.type)) content.push(...allMcids(child));
+					else content.push(child);
 				}
 			}
 		}
+		const type = typeOf(nameOf(elem.get("S")));
 		const alt = elem.get("Alt");
 		const { colSpan, rowSpan } = readSpans(file, elem.get("A") ?? PDF_NULL);
 		return {
-			type: nameOf(elem.get("S")),
-			mcids,
-			children,
+			type,
+			...settle(type, content),
 			...typeof alt === "string" ? { alt } : {},
 			...colSpan > 1 ? { colSpan } : {},
 			...rowSpan > 1 ? { rowSpan } : {}
@@ -71,6 +73,108 @@ function readStructTree(file) {
 		children: roots
 	};
 }
+/**
+* An element's own marked content and its child elements, as the node keeps
+* them.
+*
+* An element holds text of its own AND block elements only where the tree
+* mixes them — a heading carrying its number as a child, a paragraph with a
+* formula set between two of its sentences. Kept as two lists, the order
+* between them was gone and the element's own text was never read at all:
+* bug1937438_mml_from_latex.pdf's heading came back as "1" without "A small
+* example", and its sentence around a formula without its words. Each
+* stretch of the element's own content becomes a paragraph of its own
+* standing where it stood — except in a figure, whose marked content is its
+* picture.
+*/
+function settle(type, content) {
+	const nodes = content.filter((c) => "type" in c);
+	const own = content.filter((c) => !("type" in c));
+	if (nodes.length === 0 || own.length === 0 || type === "Figure") return {
+		mcids: own,
+		children: nodes
+	};
+	const children = [];
+	let stretch = [];
+	const close = () => {
+		if (stretch.length > 0) children.push({
+			type: "P",
+			mcids: stretch,
+			children: []
+		});
+		stretch = [];
+	};
+	for (const c of content) if ("type" in c) {
+		close();
+		children.push(c);
+	} else stretch.push(c);
+	close();
+	return {
+		mcids: [],
+		children
+	};
+}
+/** Every marked content reference under a node, in reading order. */
+function allMcids(node) {
+	return [...node.mcids, ...node.children.flatMap(allMcids)];
+}
+/**
+* §14.8.4.4 — the inline-level structure types (PDF 2.0 adds `Em`, `Strong`,
+* `Sub`), a formula (§14.8.4.5.5, which is set in a line as often as out of
+* one), and a list label, which is the start of its item's line.
+*/
+var INLINE = new Set([
+	"Span",
+	"Quote",
+	"Reference",
+	"BibEntry",
+	"Code",
+	"Link",
+	"Annot",
+	"Ruby",
+	"RB",
+	"RT",
+	"RP",
+	"Warichu",
+	"WT",
+	"WP",
+	"Em",
+	"Strong",
+	"Sub",
+	"Formula",
+	"Lbl"
+]);
+/** The MathML namespace (ISO 32000-2 §14.8.6): an element in it is mathematics. */
+var MATHML = "http://www.w3.org/1998/Math/MathML";
+/** Whether an element is inline — by its standard type, or as MathML. */
+function inline(file, elem, type) {
+	if (INLINE.has(type)) return true;
+	const ns = file.resolve(elem.get("NS") ?? PDF_NULL);
+	const uri = ns instanceof Map ? file.resolve(ns.get("NS") ?? PDF_NULL) : void 0;
+	return typeof uri === "string" && uri === MATHML;
+}
+/**
+* §14.8.4.2 `/RoleMap` — a document's own structure types, mapped to the
+* standard ones they stand for, followed until a type maps no further. A
+* LaTeX document names its elements `section`, `text-unit` and
+* `section-number`, and maps them to `H1`, `Part` and `Span`: read by their
+* own names, none of them meant anything.
+*/
+function roleResolver(file, stRoot) {
+	const map = file.resolve(stRoot.get("RoleMap") ?? PDF_NULL);
+	if (!(map instanceof Map)) return (type) => type;
+	return (type) => {
+		let at = type;
+		for (let hop = 0; hop < MAX_ROLE_HOPS; hop++) {
+			const next = file.resolve(map.get(at) ?? PDF_NULL);
+			if (!(next instanceof PdfName) || next.value === at) break;
+			at = next.value;
+		}
+		return at;
+	};
+}
+/** How far a chain of role mappings is followed before it is taken for a cycle. */
+var MAX_ROLE_HOPS = 8;
 function kidList(file, kVal) {
 	if (kVal === void 0) return [];
 	const k = file.resolve(kVal);

package/dist/esm/pdf-reader/tagged.js

@@ -2,10 +2,11 @@ import { pt } from "../core/ir/units.js";
 import { ResourceStore } from "../core/ir/resources.js";
 import { FEATURES } from "../core/ir/features.js";
 import { collectEmbeddedFonts } from "./embedded-fonts.js";
+import { collectFaceFamilies } from "./font.js";
 import { extractPageText } from "./text.js";
 import { collectPageVectors } from "./vector.js";
 import { displayOf, placeRuns, placeVectors } from "./display.js";
-import { buildFlowDoc, dedupeLosses, imageBlock, paragraphBlock, paragraphFromRuns, sectionFromPdfPages, shapeBlock, withMeasuredMargins } from "./flow-build.js";
+import { buildFlowDoc, dedupeLosses, imageBlock, paragraphBlock, paragraphFromRuns, sectionFromPdfPages, shapeBlock, spaceAfter, withMeasuredMargins } from "./flow-build.js";
 import { collectPageImages } from "./images.js";
 import { markDrawnRules } from "./text-rules.js";
 import { endedParagraph } from "./layout.js";
@@ -78,10 +79,14 @@ function reconstructTaggedPdf(file) {
 	const textOf = (node) => squash(node.mcids.map(({ page, mcid }) => runsOfMcid(page, mcid).map((r) => r.text).join("")).join(" "));
 	const spansOf = (node) => {
 		const spans = [];
-		node.mcids.forEach(({ page, mcid }, i) => {
-			if (i > 0) spans.push({ text: " " });
-			for (const run of runsOfMcid(page, mcid)) spans.push(spanOf(run));
-		});
+		let last;
+		for (const { page, mcid } of node.mcids) {
+			const runs = runsOfMcid(page, mcid);
+			const first = runs[0];
+			if (last !== void 0 && first !== void 0 && spacedApart(last, first)) spans.push(spaceAfter(spans[spans.length - 1]));
+			for (const run of runs) spans.push(spanOf(run));
+			last = runs[runs.length - 1] ?? last;
+		}
 		return spans;
 	};
 	/** One run as the span that carries everything the page showed it with. */
@@ -156,8 +161,10 @@ function reconstructTaggedPdf(file) {
 			groups[groups.length - 1].push(line);
 			prev = line;
 		}
+		const ends = (spans) => /\s$/u.test(spans.at(-1)?.text ?? "");
+		const opens = (spans) => /^\s/u.test(spans[0]?.text ?? "");
 		return groups.map((g) => ({
-			spans: g.flatMap((l, i) => i > 0 ? [{ text: " " }, ...l.spans] : [...l.spans]),
+			spans: g.flatMap((l, i) => i > 0 && !ends(g[i - 1].spans) && !opens(l.spans) ? [spaceAfter(g[i - 1].spans.at(-1)), ...l.spans] : [...l.spans]),
 			set: {
 				top: g[0].y,
 				bottom: g[g.length - 1].y,
@@ -323,7 +330,7 @@ function reconstructTaggedPdf(file) {
 	const reached = [...claimed].reduce((n, r) => n + r.text.length, 0);
 	if (onPage > 0 && reached * 2 < onPage) return void 0;
 	return {
-		doc: buildFlowDoc(body, resources, withMeasuredMargins(sectionFromPdfPages(pages), shown, placedRuns, pageImages.map((p) => p.images)), collectEmbeddedFonts(file, pages, imageLosses)),
+		doc: buildFlowDoc(body, resources, withMeasuredMargins(sectionFromPdfPages(pages), shown, placedRuns, pageImages.map((p) => p.images)), collectEmbeddedFonts(file, pages, imageLosses), [], void 0, collectFaceFamilies(file, pages)),
 		losses: imageLosses
 	};
 }
@@ -414,6 +421,19 @@ function equalGrid(raw) {
 		grid: Array.from({ length: numCols }, () => w)
 	};
 }
+/**
+* Whether the page shows a space between one run and the next: the second
+* starts another line, or stands clear of the first by more than a kern —
+* and neither already carries the space.
+*/
+function spacedApart(prev, next) {
+	if (/\s$/u.test(prev.text) || /^\s/u.test(next.text)) return false;
+	const size = Math.max(prev.fontSizePt, next.fontSizePt, 1);
+	if (Math.abs(prev.y - next.y) > size * .5) return true;
+	return next.x - prev.endX > size * MCID_SPACE_EM;
+}
+/** A gap between two stretches of marked content this wide, in ems, is a word space. */
+var MCID_SPACE_EM = .15;
 function squash(text) {
 	return text.replace(/\s+/g, " ").trim();
 }

package/dist/esm/pdf-reader/text-rules.js

@@ -95,7 +95,7 @@ var SAME_HEIGHT_PT = .5;
 * separate table rules is not.
 */
 function joinRules(vectors) {
-	const rules = vectors.filter((v) => isRule(v));
+	const rules = vectors.filter((v) => v.glyph !== true && isRule(v));
 	const byRow = /* @__PURE__ */ new Map();
 	for (const v of rules) {
 		const mid = (v.minY + v.maxY) / 2;

package/dist/esm/pdf-reader/text.js

@@ -2,10 +2,10 @@ import { PDF_NULL, PdfName, PdfStream } from "../pdf/objects.js";
 import { buildColorSpaceMap, buildShadingMap } from "./shading.js";
 import { IDENTITY, interpretContent, multiply } from "./content.js";
 import { textMarkupOf } from "./annot-draw.js";
-import { collectPageAppearances } from "./annots.js";
-import { hiddenProperties, hiddenXObject } from "./optional-content.js";
+import { appearanceContent, collectPageAppearances } from "./annots.js";
 import { buildContentFont } from "./font.js";
-import { patternTint } from "./pattern-tint.js";
+import { hiddenProperties, hiddenXObject } from "./optional-content.js";
+import { patternTint, tintedHex } from "./pattern-tint.js";
 //#region src/pdf-reader/text.ts
 var MAX_FORM_DEPTH = 8;
 /**
@@ -22,7 +22,12 @@ var MAX_FORM_DEPTH = 8;
 function extractPageText(file, page) {
 	const runs = [];
 	collectRuns(file, page.resources, file.pageContent(page), IDENTITY, 0, /* @__PURE__ */ new Set(), runs);
-	for (const appearance of collectPageAppearances(file, page)) collectRuns(file, appearance.resources ?? page.resources, file.streamData(appearance.stream), appearance.ctm, 1, new Set([appearance.stream]), runs);
+	const own = runs.length;
+	for (const appearance of collectPageAppearances(file, page)) collectRuns(file, appearance.resources ?? page.resources, appearanceContent(file, appearance), appearance.ctm, 1, new Set([appearance.stream]), runs);
+	for (let i = own; i < runs.length; i++) runs[i] = {
+		...runs[i],
+		annotation: true
+	};
 	const links = collectLinks(file, page);
 	const marks = collectTextMarkup(file, page);
 	const shown = withoutRestrikes(runs);
@@ -244,16 +249,6 @@ function withPatternColour(file, resources, run, visiting) {
 		visiting.delete(stream);
 	}
 }
-/** A colour laid over white paper at `coverage` strength, as a 6-hex string. */
-function tintedHex(colorHex, coverage) {
-	const k = Math.min(1, Math.max(0, coverage));
-	if (k >= 1) return colorHex;
-	const channel = (at) => {
-		const c = Number.parseInt(colorHex.slice(at, at + 2), 16);
-		return Math.round(255 - (255 - (Number.isFinite(c) ? c : 0)) * k).toString(16).toUpperCase().padStart(2, "0");
-	};
-	return `${channel(0)}${channel(2)}${channel(4)}`;
-}
 /**
 * The `/Font` resources of one dictionary, built into interpreter fonts. Shared
 * with the path and picture walks, which need them for one thing only: a Type 3

package/dist/esm/pdf-reader/vector.d.ts

@@ -1,5 +1,5 @@
+import { PageGradient } from './shading.js';
 import { PathSeg } from './content.js';
-import { ShapeGradient } from '../core/vector.js';
 import { Loss } from '../core/ir/index.js';
 import { PdfFile, PdfPage } from './document.js';
 /**
@@ -21,7 +21,7 @@ export interface PdfVector {
     /** Present iff a qualifying solid fill survived (EP10). */
     readonly fillHex?: string;
     /** Present iff a shading-pattern fill survived (EP16c). */
-    readonly gradient?: ShapeGradient;
+    readonly gradient?: PageGradient;
     /** §11.6.4.4 — how opaque the fill is, when the page asked for less than all. */
     readonly alpha?: number;
     /**
@@ -33,6 +33,12 @@ export interface PdfVector {
     readonly strokeHex?: string;
     /** Stroke width in page-space points (EP11). */
     readonly lineWidth?: number;
+    /** §8.4.3.6 — the stroke's dash lengths in page-space points, where it is dashed. */
+    readonly dash?: ReadonlyArray<number>;
+    /** §8.4.3.3 — the stroke's cap, where it is not the butt cap. */
+    readonly cap?: 'round' | 'square';
+    /** §11.6.4.4 `/CA` — how opaque the stroke is, where less than all. */
+    readonly strokeAlpha?: number;
     readonly minX: number;
     readonly minY: number;
     readonly maxX: number;

package/dist/esm/pdf-reader/vector.js

@@ -2,9 +2,10 @@ import { PDF_NULL, PdfName, PdfStream } from "../pdf/objects.js";
 import { FEATURES } from "../core/ir/features.js";
 import { buildAlphaMap, buildColorSpaceMap, buildShadingMap, gradientShading, shadingTypeOf } from "./shading.js";
 import { IDENTITY, interpretContent, multiply } from "./content.js";
-import { collectPageAppearances } from "./annots.js";
+import { appearanceContent, collectPageAppearances } from "./annots.js";
 import { hiddenProperties, hiddenXObject } from "./optional-content.js";
 import { buildFonts } from "./text.js";
+import { patternTint, tintedHex } from "./pattern-tint.js";
 //#region src/pdf-reader/vector.ts
 function paintedVectors(file, page) {
 	const out = [];
@@ -22,6 +23,31 @@ function paintedVectors(file, page) {
 		stateCache.set(resources, made);
 		return made;
 	};
+	const tints = /* @__PURE__ */ new Map();
+	const tinted = (resources, vector) => {
+		const name = vector.patternName;
+		if (name === void 0) return vector;
+		let known = tints.get(resources);
+		if (!known) tints.set(resources, known = /* @__PURE__ */ new Map());
+		const key = `${name}\u0000${vector.patternPaint ?? ""}`;
+		let hex = known.get(key);
+		if (!known.has(key)) {
+			let tint;
+			try {
+				tint = patternTint(file, resources, name, vector.patternPaint);
+			} catch {
+				tint = void 0;
+			}
+			hex = tint ? tintedHex(tint.colorHex, tint.coverage) : void 0;
+			known.set(key, hex);
+		}
+		if (hex === void 0) return vector;
+		const { patternName: _named, patternPaint: _paint, ...plain } = vector;
+		return {
+			...plain,
+			fillHex: hex
+		};
+	};
 	const walk = (resources, content, baseCtm, depth, prefix) => {
 		if (out.length >= MAX_VECTORS) return;
 		const xobjects = resources ? file.get(resources, "XObject") : PDF_NULL;
@@ -34,7 +60,7 @@ function paintedVectors(file, page) {
 			const found = dict instanceof Map ? file.resolve(dict.get(paint.name) ?? PDF_NULL) : PDF_NULL;
 			const sh = found instanceof PdfStream ? found.dict : found instanceof Map ? found : void 0;
 			if (sh && shadingTypeOf(file, sh) === 1) continue;
-			const gradient = sh ? gradientShading(file, sh) : void 0;
+			const gradient = sh ? gradientShading(file, sh, paint.ctm) : void 0;
 			if (paint.masked) {
 				bareShadings++;
 				continue;
@@ -74,7 +100,7 @@ function paintedVectors(file, page) {
 			if (out.length >= MAX_VECTORS) return;
 			if (event.vector) {
 				out.push({
-					...event.vector,
+					...tinted(resources, event.vector),
 					orderKey: [...prefix, event.order]
 				});
 				continue;
@@ -102,7 +128,7 @@ function paintedVectors(file, page) {
 	};
 	walk(page.resources, file.pageContent(page), IDENTITY, 0, []);
 	collectPageAppearances(file, page).forEach((appearance, index) => {
-		walk(appearance.resources ?? page.resources, file.streamData(appearance.stream), appearance.ctm, 1, [Number.MAX_SAFE_INTEGER, index]);
+		walk(appearance.resources ?? page.resources, appearanceContent(file, appearance), appearance.ctm, 1, [Number.MAX_SAFE_INTEGER, index]);
 	});
 	return {
 		placements: out,
@@ -196,7 +222,10 @@ function collectPageVectors(file, page, occupied = []) {
 			...filled && v.darkens === true ? { darkens: true } : {},
 			...stroked ? {
 				strokeHex: v.strokeHex,
-				...v.lineWidth !== void 0 ? { lineWidth: v.lineWidth } : {}
+				...v.lineWidth !== void 0 ? { lineWidth: v.lineWidth } : {},
+				...v.dash !== void 0 ? { dash: v.dash } : {},
+				...v.cap !== void 0 ? { cap: v.cap } : {},
+				...v.strokeAlpha !== void 0 ? { strokeAlpha: v.strokeAlpha } : {}
 			} : {},
 			...b,
 			...v.glyph === true ? { glyph: true } : {},