@jasy/pdf 1.0.0-alpha.1 → 1.0.0-alpha.11

package/dist/renderer/page-renderer.js

@@ -1,81 +1,119 @@
-"use strict";
-var __awaiter = (this && this.__awaiter) || function (thisArg, _arguments, P, generator) {
-    function adopt(value) { return value instanceof P ? value : new P(function (resolve) { resolve(value); }); }
-    return new (P || (P = Promise))(function (resolve, reject) {
-        function fulfilled(value) { try { step(generator.next(value)); } catch (e) { reject(e); } }
-        function rejected(value) { try { step(generator["throw"](value)); } catch (e) { reject(e); } }
-        function step(result) { result.done ? resolve(result.value) : adopt(result.value).then(fulfilled, rejected); }
-        step((generator = generator.apply(thisArg, _arguments || [])).next());
-    });
-};
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.PageRenderer = void 0;
-const page_sizes_1 = require("../constants/page-sizes");
-const renderer_registry_1 = require("../utils/renderer-registry");
-const pdf_config_1 = require("./pdf-config");
-const pdf_backend_1 = require("./pdf-backend");
-class PageRenderer {
-    static render(page, objectManager) {
-        return __awaiter(this, void 0, void 0, function* () {
-            var _a;
-            const { children, config, header, footer } = page.getProps();
-            // Header (top band) and footer (bottom band) sit around the body and repeat on every
-            // physical page; they are placed by `PageElement.calculateLayout` / the page driver.
-            const renderables = [...(header ? [header] : []), ...children, ...(footer ? [footer] : [])];
-            // Page geometry (also the MediaBox below). Needed up front because flipping the
-            // display list to PDF coordinates uses the page height. config is fully resolved
-            // by the layout pass; fall back to the document default rather than asserting.
-            let [width, height] = page_sizes_1.pageFormats[(_a = config === null || config === void 0 ? void 0 : config.pageSize) !== null && _a !== void 0 ? _a : page_sizes_1.PageSize.A4];
-            if ((config === null || config === void 0 ? void 0 : config.orientation) === pdf_config_1.Orientation.landscape) {
-                [width, height] = [height, width];
+import { pageFormats, PageSize } from "../constants/page-sizes.js";
+import { RendererRegistry } from "../utils/renderer-registry.js";
+import { Orientation } from "./pdf-config.js";
+import { PdfBackend } from "./pdf-backend.js";
+export class PageRenderer {
+    static async render(page, objectManager) {
+        const { children, config, header, footer } = page.getProps();
+        // Header (top band) and footer (bottom band) sit around the body and repeat on every
+        // physical page; they are placed by `PageElement.calculateLayout` / the page driver.
+        const renderables = [...(header ? [header] : []), ...children, ...(footer ? [footer] : [])];
+        // Page geometry (also the MediaBox below). Needed up front because flipping the
+        // display list to PDF coordinates uses the page height. config is fully resolved
+        // by the layout pass; fall back to the document default rather than asserting.
+        let [width, height] = config?.customSize ?? pageFormats[config?.pageSize ?? PageSize.A4];
+        if (config?.orientation === Orientation.landscape) {
+            [width, height] = [height, width];
+        }
+        // Collect the whole page as a display list (top-left coordinates), flip it to PDF
+        // coordinates at this one seam, then serialize once. Serializing registers the
+        // fonts/images used below, so it must run before the resource section.
+        const nodes = [];
+        for (const element of renderables) {
+            const renderer = RendererRegistry.getRenderer(element);
+            nodes.push(...(await renderer(element, objectManager)));
+        }
+        // Accessible tagging: one struct context per page (allocates its StructParents index), threaded into
+        // serialize so each drawable node is wrapped in marked content. Undefined = tagging off (byte-identical).
+        const structCtx = objectManager.struct.enabled ? objectManager.struct.beginPage() : undefined;
+        const flipped = PdfBackend.flipY(nodes, height);
+        const pageContent = PdfBackend.serialize(flipped, objectManager, structCtx);
+        // Add the page content as a new object (FlateDecode-compressed when enabled). The /Length is
+        // computed inside, with an explicit EOL before `endstream` (PDF/A clause 6.1.7.1).
+        const contentObjectNumber = objectManager.addContentStream(pageContent);
+        // Get the parent object number dynamically (linked with the page object)
+        const parentObjectNumber = objectManager.getParentObjectNumber(); // Get parent object number
+        // Page object with MediaBox
+        // - Get all fonts and add it to the page (reference)
+        objectManager.registerFont("Helvetica");
+        const fontReferences = [];
+        objectManager.getAllFontsRaw().forEach((value, _key) => {
+            const fontRef = `/F${value.fontIndex} ${value.resourceIndex} 0 R`;
+            fontReferences.push(fontRef);
+        });
+        // - Get all images and add it to the page (reference)
+        const imageReferences = [];
+        objectManager.getAllImagesRaw().forEach((value) => {
+            const imageRef = `/IM${value} ${value} 0 R`;
+            imageReferences.push(imageRef);
+        });
+        const imageCode = imageReferences.length > 0
+            ? "/ProcSet [/PDF /Text /ImageB /ImageC /ImageI] /XObject <<\n" +
+                imageReferences.join("\n") +
+                "\n>>\n"
+            : "";
+        // - Transparency (ExtGState) references, registered during serialize above
+        const extGStateReferences = [];
+        objectManager.getAllExtGStatesRaw().forEach((objectNumber, name) => {
+            extGStateReferences.push(`/${name} ${objectNumber} 0 R`);
+        });
+        const extGStateCode = extGStateReferences.length > 0
+            ? "/ExtGState <<\n" + extGStateReferences.join("\n") + "\n>>\n"
+            : "";
+        // - Gradient shadings (color-glyph fills), registered during serialize above
+        const shadingReferences = [];
+        objectManager.getAllShadingsRaw().forEach((objectNumber, name) => {
+            shadingReferences.push(`/${name} ${objectNumber} 0 R`);
+        });
+        const shadingCode = shadingReferences.length > 0 ? "/Shading <<\n" + shadingReferences.join("\n") + "\n>>\n" : "";
+        // Hyperlinks: a `Link` IR node drew nothing into the content stream - here its (already Y-flipped)
+        // rect + href become a /Link annotation object. `/Border [0 0 0]` hides the default visible frame;
+        // the URL is escaped like any PDF literal string. Refs are collected into the page's /Annots below.
+        const num = (n) => Number(n.toFixed(2));
+        const annotRefs = [];
+        for (const node of flipped) {
+            // A form field: build its Widget annotation + register it as an /AcroForm field (the logic lives in
+            // forms/acroform.ts), then reference it here in the page /Annots - the same side channel as a link.
+            if (node.type === "formfield") {
+                annotRefs.push(`${objectManager.acroform.addField(node, objectManager)} 0 R`);
+                continue;
             }
-            // Collect the whole page as a display list (top-left coordinates), flip it to PDF
-            // coordinates at this one seam, then serialize once. Serializing registers the
-            // fonts/images used below, so it must run before the resource section.
-            const nodes = [];
-            for (const element of renderables) {
-                const renderer = renderer_registry_1.RendererRegistry.getRenderer(element);
-                if (renderer) {
-                    nodes.push(...(yield renderer(element, objectManager)));
-                }
+            if (node.type !== "link")
+                continue;
+            const rect = `[${num(node.x)} ${num(node.y)} ${num(node.x + node.width)} ${num(node.y + node.height)}]`;
+            // Either an external URL (/URI) or an internal named destination (/GoTo, resolved via /Names /Dests).
+            // The Link IR guarantees exactly one is set; prefer `dest` when present.
+            const action = node.dest !== undefined
+                ? `/A << /S /GoTo /D ${objectManager.pdfString(node.dest)} >>`
+                : `/A << /S /URI /URI ${objectManager.pdfString(node.href ?? "")} >>`;
+            const annot = objectManager.addObject(`<< /Type /Annot /Subtype /Link /Rect ${rect} /Border [0 0 0] ${action} >>`);
+            annotRefs.push(`${annot} 0 R`);
+        }
+        const annotsCode = annotRefs.length > 0 ? ` /Annots [${annotRefs.join(" ")}]` : "";
+        // Tagging only: /StructParents links this page's marked content to the ParentTree, /Tabs /S makes the
+        // logical structure order the tab/reading order (a PDF/UA requirement).
+        const structAttrs = structCtx ? ` /StructParents ${structCtx.structParents} /Tabs /S` : "";
+        const pageObject = `<< /Type /Page /Parent ${parentObjectNumber} 0 R /Contents ${contentObjectNumber} 0 R /Resources <<\n/Font <<\n${fontReferences.join("\n")}\n>>\n${imageCode}${extGStateCode}${shadingCode}>>\n/MediaBox [0 0 ${width} ${height}]${structAttrs}${annotsCode} >>`;
+        // Add page as new object; register its object number with the struct tree (for /Pg refs), then return it.
+        const pageObjNum = objectManager.addObject(pageObject);
+        if (structCtx)
+            objectManager.struct.setPageObject(structCtx.structParents, pageObjNum);
+        // Bookmarks + named destinations: an `Outline`/`Anchor` IR node drew nothing; record it now that the
+        // page's object number is known. `node.y` is already Y-flipped to the page-space scroll target.
+        // PDFRenderer builds /Outlines and /Names /Dests at the end.
+        for (const node of flipped) {
+            if (node.type === "outline") {
+                objectManager.outline.add({
+                    title: node.title,
+                    level: node.level,
+                    pageObjNum,
+                    top: node.y,
+                });
             }
-            const pageContent = pdf_backend_1.PdfBackend.serialize(pdf_backend_1.PdfBackend.flipY(nodes, height), objectManager);
-            // Add the page content as a new object (FlateDecode-compressed when enabled). The /Length is
-            // computed inside, with an explicit EOL before `endstream` (PDF/A clause 6.1.7.1).
-            const contentObjectNumber = objectManager.addContentStream(pageContent);
-            // Get the parent object number dynamically (linked with the page object)
-            const parentObjectNumber = objectManager.getParentObjectNumber(); // Get parent object number
-            // Page object with MediaBox
-            // - Get all fonts and add it to the page (reference)
-            objectManager.registerFont("Helvetica");
-            const fontReferences = [];
-            objectManager.getAllFontsRaw().forEach((value, _key) => {
-                const fontRef = `/F${value.fontIndex} ${value.resourceIndex} 0 R`;
-                fontReferences.push(fontRef);
-            });
-            // - Get all images and add it to the page (reference)
-            const imageReferences = [];
-            objectManager.getAllImagesRaw().forEach((value) => {
-                const imageRef = `/IM${value} ${value} 0 R`;
-                imageReferences.push(imageRef);
-            });
-            const imageCode = imageReferences.length > 0
-                ? "/ProcSet [/PDF /Text /ImageB /ImageC /ImageI] /XObject <<\n" +
-                    imageReferences.join("\n") +
-                    "\n>>\n"
-                : "";
-            // - Transparency (ExtGState) references, registered during serialize above
-            const extGStateReferences = [];
-            objectManager.getAllExtGStatesRaw().forEach((objectNumber, name) => {
-                extGStateReferences.push(`/${name} ${objectNumber} 0 R`);
-            });
-            const extGStateCode = extGStateReferences.length > 0
-                ? "/ExtGState <<\n" + extGStateReferences.join("\n") + "\n>>\n"
-                : "";
-            const pageObject = `<< /Type /Page /Parent ${parentObjectNumber} 0 R /Contents ${contentObjectNumber} 0 R /Resources <<\n/Font <<\n${fontReferences.join("\n")}\n>>\n${imageCode}${extGStateCode}>>\n/MediaBox [0 0 ${width} ${height}] >>`;
-            // Add page as new object and return the page number
-            return objectManager.addObject(pageObject);
-        });
+            else if (node.type === "anchor") {
+                objectManager.dests.add(node.name, { pageObjNum, top: node.y });
+            }
+        }
+        return pageObjNum;
     }
 }
-exports.PageRenderer = PageRenderer;

package/dist/renderer/pdf-backend.d.ts

@@ -1,5 +1,6 @@
-import { IRNode } from "../ir/display-list";
-import { PDFObjectManager } from "../utils/pdf-object-manager";
+import { IRNode } from "../ir/display-list.js";
+import { PDFObjectManager } from "../utils/pdf-object-manager.js";
+import type { PageStructContext } from "../utils/struct-tree.js";
 /**
  * PDF backend - turns display-list primitives into content-stream operators.
  *
@@ -17,14 +18,34 @@ export declare class PdfBackend {
      * directly. (image is migrated in a later slice and still arrives pre-flipped.)
      */
     static flipY(nodes: IRNode[], pageHeight: number): IRNode[];
-    /** Serialize a whole display list into one content stream (page-level entry point). */
-    static serialize(nodes: IRNode[], om: PDFObjectManager): string;
+    /** Serialize a whole display list into one content stream (page-level entry point). When a
+     *  `PageStructContext` is given (accessible tagging on), each drawable node is wrapped in a marked-content
+     *  sequence (`/Role <</MCID n>> BDC … EMC`, or `/Artifact` when untagged); graphics-state and empty nodes
+     *  pass through untouched, so untagged output stays byte-identical. */
+    /**
+     * Every number reaching the backend must be finite. A non-finite coordinate is a layout bug, but
+     * silence turns it into a corrupt PDF: `Infinity` is written into the operator stream verbatim, and a
+     * viewer then discards everything from that point on - including the footer, which is drawn last. That
+     * failure is invisible until somebody opens the file, so we refuse instead.
+     *
+     * Checks the node's numbers, not the serialized text: a document containing the WORD "Infinity" is
+     * perfectly legal and must not trip this.
+     */
+    private static assertFinite;
+    static serialize(nodes: IRNode[], om: PDFObjectManager, struct?: PageStructContext): string;
     /**
      * Escapes a string for a PDF literal string `( ... )`. The backslash must be doubled
      * first, then the parentheses that delimit the string. Without this, a ")" in the
      * text closes the string early and the remaining characters leak as raw operators.
      */
     static escapePdfString(text: string): string;
+    /**
+     * Builds the `[ ... ]` operand of a `TJ` from a run and its per-gap kern adjustments (em/1000, the
+     * sign the font declares - negative tightens). A `TJ` number moves the pen LEFT, so it is the
+     * NEGATED kern. Consecutive un-kerned glyphs stay in one encoded chunk, so `[(T) 40 (otal)]` rather
+     * than one chunk per glyph. `kerns[i]` is the adjustment AFTER code point `i`.
+     */
+    static kernedArray(text: string, kerns: number[], encode: (chunk: string) => string): string;
     /**
      * Returns the `/GSx gs` operator that selects a transparency state, or `""` when both
      * alphas are fully opaque. Opaque draws emit nothing here, so output stays
@@ -37,6 +58,12 @@ export declare class PdfBackend {
      * (kappa ≈ 0.5523). Returns m/l/c/h ops WITHOUT the paint operator.
      */
     private static roundedRectPath;
+    /**
+     * Clamp the four radii into the box. Each is capped at half the box, and two radii sharing an edge
+     * are scaled down together when they would overlap - the CSS rule, and what keeps the path from
+     * folding back on itself.
+     */
+    private static clampRadii;
     /**
      * Serialize a single display-list node to its content-stream operators.
      * `om` is used only by primitives that allocate PDF resources (images, fonts).