@jasy/pdf 1.0.0-alpha.5 → 1.0.0-alpha.7

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (105) hide show
  1. package/dist/api/content.d.ts +6 -3
  2. package/dist/api/content.js +12 -3
  3. package/dist/api/descriptor.js +24 -1
  4. package/dist/api/dimension.d.ts +20 -0
  5. package/dist/api/dimension.js +16 -0
  6. package/dist/api/index.d.ts +2 -0
  7. package/dist/api/index.js +2 -0
  8. package/dist/api/layout.d.ts +112 -17
  9. package/dist/api/layout.js +118 -13
  10. package/dist/api/page-builder.d.ts +32 -0
  11. package/dist/api/page-builder.js +32 -0
  12. package/dist/api/structure.d.ts +5 -0
  13. package/dist/api/structure.js +1 -0
  14. package/dist/api/text.d.ts +24 -2
  15. package/dist/api/text.js +46 -4
  16. package/dist/elements/container-element.d.ts +26 -1
  17. package/dist/elements/container-element.js +79 -29
  18. package/dist/elements/image-element.d.ts +19 -1
  19. package/dist/elements/image-element.js +52 -11
  20. package/dist/elements/index.d.ts +8 -0
  21. package/dist/elements/index.js +8 -0
  22. package/dist/elements/layout/anchor-element.d.ts +23 -0
  23. package/dist/elements/layout/anchor-element.js +27 -0
  24. package/dist/elements/layout/bookmark-element.d.ts +27 -0
  25. package/dist/elements/layout/bookmark-element.js +30 -0
  26. package/dist/elements/layout/keep-together-element.d.ts +42 -0
  27. package/dist/elements/layout/keep-together-element.js +85 -0
  28. package/dist/elements/layout/link-element.d.ts +33 -0
  29. package/dist/elements/layout/link-element.js +39 -0
  30. package/dist/elements/layout/padding-element.d.ts +3 -0
  31. package/dist/elements/layout/padding-element.js +12 -0
  32. package/dist/elements/layout/page-break-element.d.ts +17 -0
  33. package/dist/elements/layout/page-break-element.js +24 -0
  34. package/dist/elements/layout/page-builder-element.d.ts +32 -0
  35. package/dist/elements/layout/page-builder-element.js +43 -0
  36. package/dist/elements/layout/positioned-element.d.ts +10 -7
  37. package/dist/elements/layout/positioned-element.js +21 -9
  38. package/dist/elements/layout/rotated-box-element.d.ts +35 -0
  39. package/dist/elements/layout/rotated-box-element.js +54 -0
  40. package/dist/elements/layout/rotated-element.d.ts +32 -0
  41. package/dist/elements/layout/rotated-element.js +41 -0
  42. package/dist/elements/page-element.d.ts +19 -1
  43. package/dist/elements/page-element.js +54 -17
  44. package/dist/elements/pdf-element.d.ts +59 -0
  45. package/dist/elements/pdf-element.js +51 -0
  46. package/dist/elements/rectangle-element.d.ts +20 -1
  47. package/dist/elements/rectangle-element.js +75 -18
  48. package/dist/elements/row-element.d.ts +19 -1
  49. package/dist/elements/row-element.js +45 -8
  50. package/dist/elements/text-element.d.ts +34 -4
  51. package/dist/elements/text-element.js +40 -17
  52. package/dist/ir/display-list.d.ts +60 -1
  53. package/dist/layout/box-constraints.d.ts +8 -0
  54. package/dist/layout/box-constraints.js +14 -0
  55. package/dist/layout/collect-images.d.ts +9 -0
  56. package/dist/layout/collect-images.js +34 -0
  57. package/dist/layout/fragmentation.d.ts +12 -0
  58. package/dist/layout/fragmentation.js +45 -8
  59. package/dist/renderer/anchor-renderer.d.ts +6 -0
  60. package/dist/renderer/anchor-renderer.js +11 -0
  61. package/dist/renderer/bookmark-renderer.d.ts +6 -0
  62. package/dist/renderer/bookmark-renderer.js +11 -0
  63. package/dist/renderer/keep-together-renderer.d.ts +10 -0
  64. package/dist/renderer/keep-together-renderer.js +12 -0
  65. package/dist/renderer/link-renderer.d.ts +6 -0
  66. package/dist/renderer/link-renderer.js +11 -0
  67. package/dist/renderer/page-break-renderer.d.ts +14 -0
  68. package/dist/renderer/page-break-renderer.js +15 -0
  69. package/dist/renderer/page-builder-renderer.d.ts +6 -0
  70. package/dist/renderer/page-builder-renderer.js +11 -0
  71. package/dist/renderer/page-renderer.js +37 -2
  72. package/dist/renderer/pdf-backend.d.ts +17 -0
  73. package/dist/renderer/pdf-backend.js +117 -7
  74. package/dist/renderer/pdf-document-class.js +3 -0
  75. package/dist/renderer/pdf-document-renderer.d.ts +16 -7
  76. package/dist/renderer/pdf-document-renderer.js +87 -36
  77. package/dist/renderer/pdf-renderer.js +53 -2
  78. package/dist/renderer/rotated-renderer.d.ts +7 -0
  79. package/dist/renderer/rotated-renderer.js +29 -0
  80. package/dist/renderer/text-renderer.d.ts +1 -1
  81. package/dist/renderer/text-renderer.js +123 -54
  82. package/dist/text/advance.d.ts +34 -0
  83. package/dist/text/advance.js +42 -0
  84. package/dist/text/line-breaker.d.ts +9 -8
  85. package/dist/text/line-breaker.js +28 -21
  86. package/dist/text/line-metrics.d.ts +49 -0
  87. package/dist/text/line-metrics.js +57 -0
  88. package/dist/text/text-decoration.d.ts +50 -0
  89. package/dist/text/text-decoration.js +62 -0
  90. package/dist/text/text-style.d.ts +13 -3
  91. package/dist/text/text-style.js +10 -3
  92. package/dist/utils/afm-parser.d.ts +34 -0
  93. package/dist/utils/afm-parser.js +73 -4
  94. package/dist/utils/dest-registry.d.ts +19 -0
  95. package/dist/utils/dest-registry.js +32 -0
  96. package/dist/utils/flex-layout.d.ts +10 -3
  97. package/dist/utils/flex-layout.js +26 -6
  98. package/dist/utils/font-metrics.d.ts +14 -0
  99. package/dist/utils/outline.d.ts +21 -0
  100. package/dist/utils/outline.js +80 -0
  101. package/dist/utils/pdf-object-manager.d.ts +33 -1
  102. package/dist/utils/pdf-object-manager.js +0 -0
  103. package/dist/utils/ttf-parser.d.ts +53 -0
  104. package/dist/utils/ttf-parser.js +368 -0
  105. package/package.json +1 -1
@@ -22,6 +22,16 @@ export declare class PdfBackend {
22
22
  * `PageStructContext` is given (accessible tagging on), each drawable node is wrapped in a marked-content
23
23
  * sequence (`/Role <</MCID n>> BDC … EMC`, or `/Artifact` when untagged); graphics-state and empty nodes
24
24
  * pass through untouched, so untagged output stays byte-identical. */
25
+ /**
26
+ * Every number reaching the backend must be finite. A non-finite coordinate is a layout bug, but
27
+ * silence turns it into a corrupt PDF: `Infinity` is written into the operator stream verbatim, and a
28
+ * viewer then discards everything from that point on - including the footer, which is drawn last. That
29
+ * failure is invisible until somebody opens the file, so we refuse instead.
30
+ *
31
+ * Checks the node's numbers, not the serialized text: a document containing the WORD "Infinity" is
32
+ * perfectly legal and must not trip this.
33
+ */
34
+ private static assertFinite;
25
35
  static serialize(nodes: IRNode[], om: PDFObjectManager, struct?: PageStructContext): string;
26
36
  /**
27
37
  * Escapes a string for a PDF literal string `( ... )`. The backslash must be doubled
@@ -29,6 +39,13 @@ export declare class PdfBackend {
29
39
  * text closes the string early and the remaining characters leak as raw operators.
30
40
  */
31
41
  static escapePdfString(text: string): string;
42
+ /**
43
+ * Builds the `[ ... ]` operand of a `TJ` from a run and its per-gap kern adjustments (em/1000, the
44
+ * sign the font declares - negative tightens). A `TJ` number moves the pen LEFT, so it is the
45
+ * NEGATED kern. Consecutive un-kerned glyphs stay in one encoded chunk, so `[(T) 40 (otal)]` rather
46
+ * than one chunk per glyph. `kerns[i]` is the adjustment AFTER code point `i`.
47
+ */
48
+ static kernedArray(text: string, kerns: number[], encode: (chunk: string) => string): string;
32
49
  /**
33
50
  * Returns the `/GSx gs` operator that selects a transparency state, or `""` when both
34
51
  * alphas are fully opaque. Opaque draws emit nothing here, so output stays
@@ -62,6 +62,28 @@ export class PdfBackend {
62
62
  : { ...node.fill, y0: pageHeight - node.fill.y0, y1: pageHeight - node.fill.y1 };
63
63
  return { ...node, commands, fill };
64
64
  }
65
+ case "link":
66
+ // A link's clickable rect flips around its bottom edge, exactly like a rect.
67
+ return { ...node, y: pageHeight - node.y - node.height };
68
+ case "outline":
69
+ // An outline anchor is a single point (the target's top); flip it like a text baseline.
70
+ return { ...node, y: pageHeight - node.y };
71
+ case "anchor":
72
+ // A named-destination anchor is a single point (the target's top), flipped like a baseline.
73
+ return { ...node, y: pageHeight - node.y };
74
+ case "transform-push": {
75
+ // The child nodes are flipped from top-left to bottom-left by F = [1,0,0,-1,0,H] (F is its
76
+ // own inverse). For a matrix authored in top-left space to act correctly on them, conjugate
77
+ // it by that flip: M_pdf = F · M · F. Worked out for [a,b,c,d,e,f] with H = pageHeight:
78
+ const [a, b, c, d, e, f] = node.matrix;
79
+ const H = pageHeight;
80
+ return {
81
+ ...node,
82
+ matrix: [a, -b, -c, d, H * c + e, H - H * d - f],
83
+ };
84
+ }
85
+ case "transform-pop":
86
+ return node; // no coordinates to flip
65
87
  default: {
66
88
  const unknown = node;
67
89
  return unknown;
@@ -73,11 +95,49 @@ export class PdfBackend {
73
95
  * `PageStructContext` is given (accessible tagging on), each drawable node is wrapped in a marked-content
74
96
  * sequence (`/Role <</MCID n>> BDC … EMC`, or `/Artifact` when untagged); graphics-state and empty nodes
75
97
  * pass through untouched, so untagged output stays byte-identical. */
98
+ /**
99
+ * Every number reaching the backend must be finite. A non-finite coordinate is a layout bug, but
100
+ * silence turns it into a corrupt PDF: `Infinity` is written into the operator stream verbatim, and a
101
+ * viewer then discards everything from that point on - including the footer, which is drawn last. That
102
+ * failure is invisible until somebody opens the file, so we refuse instead.
103
+ *
104
+ * Checks the node's numbers, not the serialized text: a document containing the WORD "Infinity" is
105
+ * perfectly legal and must not trip this.
106
+ */
107
+ static assertFinite(node) {
108
+ const finite = (value) => {
109
+ if (typeof value === "number")
110
+ return Number.isFinite(value);
111
+ if (Array.isArray(value))
112
+ return value.every(finite);
113
+ if (value !== null && typeof value === "object")
114
+ return Object.values(value).every(finite);
115
+ return true;
116
+ };
117
+ if (!finite(node)) {
118
+ throw new Error(`Layout produced a non-finite number on a "${node.type}" node, refusing to emit a corrupt ` +
119
+ `content stream. This is a bug in an element's calculateLayout (a size resolved to Infinity ` +
120
+ `or NaN), not in your document.`);
121
+ }
122
+ }
76
123
  static serialize(nodes, om, struct) {
77
124
  return nodes
78
125
  .map((node) => {
126
+ PdfBackend.assertFinite(node);
79
127
  const ops = PdfBackend.serializeNode(node, om);
80
- if (!struct || node.type === "clip-push" || node.type === "clip-pop" || ops === "")
128
+ // Two kinds of node skip the marked-content wrapper, both because they carry no tag and never
129
+ // draw: GRAPHICS-STATE nodes (clip / transform push+pop) only save/restore state around the
130
+ // drawables they enclose, and SIDE-CHANNEL nodes (link, outline, anchor) become annotations /
131
+ // catalog entries instead of content-stream operators.
132
+ if (!struct ||
133
+ node.type === "clip-push" ||
134
+ node.type === "clip-pop" ||
135
+ node.type === "transform-push" ||
136
+ node.type === "transform-pop" ||
137
+ node.type === "link" ||
138
+ node.type === "outline" ||
139
+ node.type === "anchor" ||
140
+ ops === "")
81
141
  return ops;
82
142
  const { open, close } = struct.mark(node.tag);
83
143
  return open + ops + close;
@@ -92,6 +152,27 @@ export class PdfBackend {
92
152
  static escapePdfString(text) {
93
153
  return text.replace(/\\/g, "\\\\").replace(/\(/g, "\\(").replace(/\)/g, "\\)");
94
154
  }
155
+ /**
156
+ * Builds the `[ ... ]` operand of a `TJ` from a run and its per-gap kern adjustments (em/1000, the
157
+ * sign the font declares - negative tightens). A `TJ` number moves the pen LEFT, so it is the
158
+ * NEGATED kern. Consecutive un-kerned glyphs stay in one encoded chunk, so `[(T) 40 (otal)]` rather
159
+ * than one chunk per glyph. `kerns[i]` is the adjustment AFTER code point `i`.
160
+ */
161
+ static kernedArray(text, kerns, encode) {
162
+ const chars = [...text];
163
+ let out = "";
164
+ let chunk = chars[0] ?? "";
165
+ for (let i = 0; i < kerns.length; i++) {
166
+ if (kerns[i] === 0) {
167
+ chunk += chars[i + 1];
168
+ }
169
+ else {
170
+ out += encode(chunk) + ` ${-kerns[i]} `;
171
+ chunk = chars[i + 1];
172
+ }
173
+ }
174
+ return `[${out}${encode(chunk)}]`;
175
+ }
95
176
  /**
96
177
  * Returns the `/GSx gs` operator that selects a transparency state, or `""` when both
97
178
  * alphas are fully opaque. Opaque draws emit nothing here, so output stays
@@ -182,18 +263,31 @@ export class PdfBackend {
182
263
  const font = isCustom
183
264
  ? om.getCustomFontResource(node.fontFamily, node.fontStyle)
184
265
  : om.registerFont(node.fontFamily, node.fontStyle);
185
- const textOp = isCustom
186
- ? `<${om.encodeCustomText(node.fontFamily, node.text, node.fontStyle)}>`
187
- : `(${PdfBackend.escapePdfString(node.text)})`;
266
+ const encode = (t) => isCustom
267
+ ? `<${om.encodeCustomText(node.fontFamily, t, node.fontStyle)}>`
268
+ : `(${PdfBackend.escapePdfString(t)})`;
269
+ // Kerning (if the document has it on): a `TJ` array with the font's per-pair adjustments
270
+ // between glyph chunks, measured into the layout by the SAME numbers (runAdvance). A plain
271
+ // `Tj` when the run has no kerning, byte-identical.
272
+ const kerns = om.kerningEnabled
273
+ ? om.getKernPairs(node.text, node.fontFamily, node.fontStyle)
274
+ : [];
275
+ const showOp = kerns.some((k) => k !== 0)
276
+ ? `${PdfBackend.kernedArray(node.text, kerns, encode)} TJ`
277
+ : `${encode(node.text)} Tj`;
278
+ // letterSpacing is the `Tc` operator: extra advance after EVERY glyph (an embedded
279
+ // Identity-H font too - that is `Tw`, word spacing, which only touches single-byte code 32).
280
+ const spacing = node.letterSpacing ? `${node.letterSpacing.toFixed(3)} Tc\n` : "";
188
281
  const block = `BT\n` +
189
282
  `${node.color.toPDFColorString()} rg ` +
190
283
  `/F${font.fontIndex} ${node.fontSize} Tf ` +
191
284
  `${node.x.toFixed(3)} ${node.y.toFixed(3)} Td ` +
192
- `${textOp} Tj\n` +
285
+ `${showOp}\n` +
193
286
  `ET\n`;
194
- // Transparent text gets an isolating q/Q + gs; opaque text is byte-identical.
195
287
  const gs = PdfBackend.alphaPrefix(om, node.color.getAlpha(), 1);
196
- return gs ? `q\n${gs}${block}Q\n` : block;
288
+ // `Tc` is graphics state and would leak into the next run, so isolate it in q/Q - the same
289
+ // reason transparent text is isolated. A run with neither stays byte-identical.
290
+ return gs || spacing ? `q\n${gs}${spacing}${block}Q\n` : block;
197
291
  }
198
292
  case "image": {
199
293
  // The backend owns PDF resource creation: register the XObject (using the
@@ -247,6 +341,22 @@ export class PdfBackend {
247
341
  }
248
342
  case "clip-pop":
249
343
  return `Q\n`;
344
+ case "transform-push": {
345
+ // Save the graphics state, then apply the affine; everything until transform-pop paints through it.
346
+ const f = (n) => n.toFixed(3);
347
+ return `q\n${node.matrix.map(f).join(" ")} cm\n`;
348
+ }
349
+ case "transform-pop":
350
+ return `Q\n`;
351
+ case "link":
352
+ // A link draws nothing in the content stream - it becomes a page /Annot (built in PageRenderer).
353
+ return "";
354
+ case "outline":
355
+ // An outline anchor draws nothing - it becomes a /Outlines entry (built in PageRenderer/PDFRenderer).
356
+ return "";
357
+ case "anchor":
358
+ // A named destination draws nothing - it becomes a /Names /Dests entry (built in PageRenderer/PDFRenderer).
359
+ return "";
250
360
  default: {
251
361
  // Exhaustiveness guard: if a new IRNode variant is added, this fails to compile.
252
362
  const unknown = node;
@@ -88,6 +88,9 @@ export class PDFDocument {
88
88
  }
89
89
  if (config)
90
90
  this._objectManager.changePDFConfig(config);
91
+ // Kerning is on by default at the document level (the `renderToBytes` path sets the same). A
92
+ // bare `PDFObjectManager` still defaults OFF, so unit tests keep predictable un-kerned metrics.
93
+ this._objectManager.setKerning(true);
91
94
  }
92
95
  get objectManager() {
93
96
  return this._objectManager;
@@ -4,16 +4,25 @@ import { PDFObjectManager } from "../utils/pdf-object-manager.js";
4
4
  export declare class PDFDocumentRenderer {
5
5
  static render(document: PDFDocumentElement, objectManager: PDFObjectManager, ctx: LayoutContext): Promise<number>;
6
6
  /**
7
- * Renders one logical page, paginating it into one or more physical pages.
7
+ * The page geometry a physical page will use. `PageElement.calculateLayout` has already folded the
8
+ * document defaults into every page's config by the time we get here (`PDFRenderer` lays the document
9
+ * out before it calls us, and it is our only caller). Merging again is therefore a no-op today - it is
10
+ * here so this stays correct BY CONSTRUCTION rather than by call order: were that order ever changed,
11
+ * an unmerged config would otherwise make `resolvePageSize` fall back to A4 and quietly report the
12
+ * wrong page size.
13
+ */
14
+ private static configOf;
15
+ /**
16
+ * Splits one logical page into the physical pages it will occupy, without rendering.
8
17
  *
9
- * Slice 0 only auto-paginates the common shape - a page whose single child is a
10
- * fragmentation context (a Container). Anything else renders as a single page on the
11
- * unchanged path, so non-overflowing documents stay byte-identical to pre-Slice-0.
18
+ * Only the common shape auto-paginates - a page whose single child is a fragmentation context
19
+ * (a Container). Anything else stays a single page on the unchanged path, so non-overflowing
20
+ * documents keep producing byte-identical output.
12
21
  */
13
- private static renderLogicalPage;
22
+ private static paginateLogicalPage;
14
23
  /**
15
- * Lays out one fragment on a fresh page of the same geometry and renders it. The
16
- * header/footer are attached to every physical page so they repeat; `PageElement`
24
+ * Lays out one physical page and renders it. A fragment gets a fresh `PageElement` of the same
25
+ * geometry; the header/footer are attached to every physical page so they repeat, and `PageElement`
17
26
  * re-places them and the body in the band between.
18
27
  */
19
28
  private static renderPhysicalPage;
@@ -1,21 +1,36 @@
1
- import { layoutPageBands, PageElement } from "../elements/page-element.js";
1
+ import { layoutPageBands, pageFrame, PageElement, resolvePageSize, } from "../elements/page-element.js";
2
2
  import { BoxConstraints } from "../layout/box-constraints.js";
3
- import { isFragmentable } from "../layout/fragmentation.js";
3
+ import { isFragmentable, reportOverflow } from "../layout/fragmentation.js";
4
4
  import { PageRenderer } from "./page-renderer.js";
5
5
  export class PDFDocumentRenderer {
6
6
  static async render(document, objectManager, ctx) {
7
- const pageNumbers = [];
8
7
  // Add the pages object first... we need its object number (resources). The count is
9
8
  // a placeholder; it is replaced below with the real (post-pagination) page count.
10
9
  const pagesObject = `<< /Type /Pages /Kids [] /Count ${document.getProps().children.length} >>`;
11
10
  const pagesObjectNumber = objectManager.addObject(pagesObject);
12
11
  // Now set the given object number all its childs
13
12
  objectManager.setParentObjectNumber(pagesObjectNumber);
14
- // The page driver: each logical PageElement may produce SEVERAL physical PDF pages
15
- // when its content overflows (Slice 0: whole children reflow to the next page).
16
- for (let page of document.getProps().children) {
17
- const numbers = await PDFDocumentRenderer.renderLogicalPage(page, objectManager, ctx);
18
- pageNumbers.push(...numbers);
13
+ // Pass A - paginate the whole document WITHOUT drawing anything. Fragmentation is synchronous and
14
+ // pure (`{ fitted, remainder }`), so the full list of physical pages can be settled up front. Only
15
+ // then is the total page count known, which is what page numbers in a header/footer need.
16
+ const physicalPages = [];
17
+ for (const page of document.getProps().children) {
18
+ physicalPages.push(...PDFDocumentRenderer.paginateLogicalPage(page, ctx));
19
+ }
20
+ // Pass B - draw them, in order, so the emitted page objects keep their previous numbering. Each page
21
+ // is laid out again with its own PageInfo, which is how a `PageBuilder` anywhere in the tree (header,
22
+ // footer or body) learns its final page number and the now-known document total.
23
+ const pageNumbers = [];
24
+ for (const [index, physical] of physicalPages.entries()) {
25
+ const pageInfo = {
26
+ pageNumber: index + 1,
27
+ pageCount: physicalPages.length,
28
+ pageSize: resolvePageSize(PDFDocumentRenderer.configOf(physical, ctx)),
29
+ };
30
+ pageNumbers.push(await PDFDocumentRenderer.renderPhysicalPage(physical, objectManager, {
31
+ ...ctx,
32
+ pageInfo,
33
+ }));
19
34
  }
20
35
  // We must update the pages object with the real physical page numbers and count.
21
36
  const updatedPagesObject = `<< /Type /Pages /Kids [${pageNumbers
@@ -26,65 +41,101 @@ export class PDFDocumentRenderer {
26
41
  return pagesObjectNumber;
27
42
  }
28
43
  /**
29
- * Renders one logical page, paginating it into one or more physical pages.
44
+ * The page geometry a physical page will use. `PageElement.calculateLayout` has already folded the
45
+ * document defaults into every page's config by the time we get here (`PDFRenderer` lays the document
46
+ * out before it calls us, and it is our only caller). Merging again is therefore a no-op today - it is
47
+ * here so this stays correct BY CONSTRUCTION rather than by call order: were that order ever changed,
48
+ * an unmerged config would otherwise make `resolvePageSize` fall back to A4 and quietly report the
49
+ * wrong page size.
50
+ */
51
+ static configOf(physical, ctx) {
52
+ const config = physical.kind === "whole" ? physical.page.getProps().config : physical.config;
53
+ return { ...ctx.pageConfig, ...config };
54
+ }
55
+ /**
56
+ * Splits one logical page into the physical pages it will occupy, without rendering.
30
57
  *
31
- * Slice 0 only auto-paginates the common shape - a page whose single child is a
32
- * fragmentation context (a Container). Anything else renders as a single page on the
33
- * unchanged path, so non-overflowing documents stay byte-identical to pre-Slice-0.
58
+ * Only the common shape auto-paginates - a page whose single child is a fragmentation context
59
+ * (a Container). Anything else stays a single page on the unchanged path, so non-overflowing
60
+ * documents keep producing byte-identical output.
34
61
  */
35
- static async renderLogicalPage(page, objectManager, ctx) {
62
+ static paginateLogicalPage(page, ctx) {
36
63
  const { children, config, header, footer } = page.getProps();
37
64
  if (children.length !== 1 || !isFragmentable(children[0])) {
38
- return [await PageRenderer.render(page, objectManager)];
65
+ return [{ kind: "whole", page }];
39
66
  }
40
67
  // Header/footer repeat on every physical page, so the body only ever flows into the
41
68
  // band between them. Resolve that band once (config is already merged by pass 1).
69
+ // The frame is a throwaway: this pass only wants the band HEIGHTS, and an out-of-flow child
70
+ // contributes none. It exists because a `Positioned` in a band demands one; pass 2 builds the
71
+ // real frame (in `PageElement`) and drains it.
42
72
  const pageCtx = {
43
73
  metrics: ctx.metrics,
44
74
  pageConfig: config,
45
75
  textStyle: ctx.textStyle,
46
76
  onOverflow: ctx.onOverflow,
77
+ frame: pageFrame(config),
47
78
  };
48
79
  const { bodyWidth: width, bodyHeight: height } = layoutPageBands(config, header, footer, pageCtx);
49
- const numbers = [];
80
+ // A `keepTogether` group compares its height against a FULL fresh page (this band), so thread it in.
81
+ pageCtx.pageBodyHeight = height;
82
+ const pages = [];
50
83
  let region = children[0];
51
84
  let isFirstRegion = true;
52
85
  while (region) {
53
86
  if (!isFragmentable(region)) {
54
87
  // A non-fragmentable remainder is placed whole on its own page.
55
- numbers.push(await PDFDocumentRenderer.renderPhysicalPage(config, region, header, footer, objectManager, ctx));
88
+ pages.push({ kind: "fragment", config, content: region, header, footer });
56
89
  break;
57
90
  }
58
- const { fitted, remainder } = region.fragment(height, width, ctx);
59
- // Everything fits on one page: render the ORIGINAL page so output is unchanged.
60
- // Measuring inside fragment() laid the children out at the origin to size them, so
61
- // re-run the page layout first to restore their real positions (deterministic).
62
- if (isFirstRegion && remainder === null) {
63
- page.calculateLayout(new BoxConstraints(), { x: 0, y: 0 }, ctx);
64
- numbers.push(await PageRenderer.render(page, objectManager));
91
+ // pageCtx, not the document ctx: a descendant that reads `pageConfig` while being measured (a
92
+ // `PageBuilder` sizing its provisional build) must see THIS page's geometry, not the document default.
93
+ const { fitted, remainder, forceBreak } = region.fragment(height, width, pageCtx);
94
+ // Everything fits on one page: keep the ORIGINAL page so output is unchanged. Measuring inside
95
+ // fragment() left its children at the measuring origin; pass B lays the page out again, which
96
+ // restores their real positions (layout is deterministic). A consumed trailing `PageBreak`
97
+ // (forceBreak with no remainder) is the exception: render the packed `fitted` instead, which no
98
+ // longer holds the break marker, so the original with the break does not leak into the render pass.
99
+ if (isFirstRegion && remainder === null && !forceBreak) {
100
+ pages.push({ kind: "whole", page });
65
101
  break;
66
102
  }
67
- if (fitted) {
68
- numbers.push(await PDFDocumentRenderer.renderPhysicalPage(config, fitted, header, footer, objectManager, ctx));
103
+ // TERMINATION GUARANTEE. Every page here has the full body height, so `fitted === null` means
104
+ // nothing fit even on a whole page - the region did not get smaller. Advancing to the
105
+ // (identical) remainder would loop forever. So we stop: place the region whole (clipped to the
106
+ // page) and surface it per the overflow policy. A step that shrinks nothing ends the loop, for
107
+ // ANY region - an oversized unbreakable block, a future keep-together group, or an engine bug.
108
+ if (fitted === null) {
109
+ // Measure the region's true height for the message (rare path, so the extra pass is free);
110
+ // pass B lays it out again, so this mutation does not leak.
111
+ const needed = region.calculateLayout(BoxConstraints.loose(width, Infinity), { x: 0, y: 0 }, pageCtx).height;
112
+ reportOverflow(region, needed, height, pageCtx.onOverflow ?? "ignore");
113
+ pages.push({ kind: "fragment", config, content: region, header, footer });
114
+ break;
69
115
  }
116
+ pages.push({ kind: "fragment", config, content: fitted, header, footer });
70
117
  region = remainder;
71
118
  isFirstRegion = false;
72
119
  }
73
- return numbers;
120
+ return pages;
74
121
  }
75
122
  /**
76
- * Lays out one fragment on a fresh page of the same geometry and renders it. The
77
- * header/footer are attached to every physical page so they repeat; `PageElement`
123
+ * Lays out one physical page and renders it. A fragment gets a fresh `PageElement` of the same
124
+ * geometry; the header/footer are attached to every physical page so they repeat, and `PageElement`
78
125
  * re-places them and the body in the band between.
79
126
  */
80
- static async renderPhysicalPage(config, content, header, footer, objectManager, ctx) {
81
- const physicalPage = new PageElement({
82
- config,
83
- header,
84
- footer,
85
- children: [content],
127
+ static async renderPhysicalPage(physical, objectManager, ctx) {
128
+ if (physical.kind === "whole") {
129
+ physical.page.calculateLayout(new BoxConstraints(), { x: 0, y: 0 }, ctx);
130
+ return PageRenderer.render(physical.page, objectManager);
131
+ }
132
+ const page = new PageElement({
133
+ config: physical.config,
134
+ header: physical.header,
135
+ footer: physical.footer,
136
+ children: [physical.content],
86
137
  });
87
- physicalPage.calculateLayout(new BoxConstraints(), { x: 0, y: 0 }, ctx);
88
- return PageRenderer.render(physicalPage, objectManager);
138
+ page.calculateLayout(new BoxConstraints(), { x: 0, y: 0 }, ctx);
139
+ return PageRenderer.render(page, objectManager);
89
140
  }
90
141
  }
@@ -21,7 +21,24 @@ import { PositionedElement } from "../elements/layout/positioned-element.js";
21
21
  import { PositionedRenderer } from "./positioned-renderer.js";
22
22
  import { StructGroup } from "../elements/layout/struct-group.js";
23
23
  import { StructGroupRenderer } from "./struct-group-renderer.js";
24
+ import { LinkElement } from "../elements/layout/link-element.js";
25
+ import { LinkRenderer } from "./link-renderer.js";
26
+ import { BookmarkElement } from "../elements/layout/bookmark-element.js";
27
+ import { BookmarkRenderer } from "./bookmark-renderer.js";
28
+ import { AnchorElement } from "../elements/layout/anchor-element.js";
29
+ import { PageBreakElement } from "../elements/layout/page-break-element.js";
30
+ import { AnchorRenderer } from "./anchor-renderer.js";
31
+ import { PageBreakRenderer } from "./page-break-renderer.js";
32
+ import { KeepTogetherElement } from "../elements/layout/keep-together-element.js";
33
+ import { KeepTogetherRenderer } from "./keep-together-renderer.js";
34
+ import { PageBuilderElement } from "../elements/layout/page-builder-element.js";
35
+ import { PageBuilderRenderer } from "./page-builder-renderer.js";
36
+ import { RotatedElement } from "../elements/layout/rotated-element.js";
37
+ import { RotatedBoxElement } from "../elements/layout/rotated-box-element.js";
38
+ import { RotatedRenderer } from "./rotated-renderer.js";
39
+ import { PdfBackend } from "./pdf-backend.js";
24
40
  import { BoxConstraints } from "../layout/box-constraints.js";
41
+ import { collectImageElements } from "../layout/collect-images.js";
25
42
  import { DEFAULT_TEXT_STYLE, mergeTextStyle } from "../text/text-style.js";
26
43
  export class PDFRenderer {
27
44
  static async render(document, objectManager) {
@@ -32,6 +49,8 @@ export class PDFRenderer {
32
49
  RendererRegistry.register(RectangleElement, RectangleRenderer.render);
33
50
  RendererRegistry.register(ExpandedElement, ExpandedRenderer.render);
34
51
  RendererRegistry.register(PaddingElement, PaddingRenderer.render);
52
+ RendererRegistry.register(PageBreakElement, PageBreakRenderer.render);
53
+ RendererRegistry.register(KeepTogetherElement, KeepTogetherRenderer.render);
35
54
  RendererRegistry.register(DefaultTextStyleElement, DefaultTextStyleRenderer.render);
36
55
  RendererRegistry.register(ImageElement, ImageRenderer.render);
37
56
  RendererRegistry.register(LineElement, LineRenderer.render);
@@ -39,6 +58,12 @@ export class PDFRenderer {
39
58
  RendererRegistry.register(DeferredElement, DeferredRenderer.render);
40
59
  RendererRegistry.register(PositionedElement, PositionedRenderer.render);
41
60
  RendererRegistry.register(StructGroup, StructGroupRenderer.render);
61
+ RendererRegistry.register(LinkElement, LinkRenderer.render);
62
+ RendererRegistry.register(BookmarkElement, BookmarkRenderer.render);
63
+ RendererRegistry.register(AnchorElement, AnchorRenderer.render);
64
+ RendererRegistry.register(PageBuilderElement, PageBuilderRenderer.render);
65
+ RendererRegistry.register(RotatedElement, RotatedRenderer.render);
66
+ RendererRegistry.register(RotatedBoxElement, RotatedRenderer.render);
42
67
  let pdfContent = "";
43
68
  // Header: version line + the PDF/A binary marker (the object manager owns it so its length
44
69
  // matches the xref offset calculation).
@@ -51,6 +76,11 @@ export class PDFRenderer {
51
76
  textStyle: mergeTextStyle(DEFAULT_TEXT_STYLE, document.getDefaultTextStyle()),
52
77
  onOverflow: objectManager.getOverflowPolicy(),
53
78
  };
79
+ // Pre-layout: resolve every image's intrinsic pixel size (async), so the synchronous layout can
80
+ // give a width-only image a proportional height. init() is idempotent, so the renderer reuses it.
81
+ const images = collectImageElements(document);
82
+ if (images.length > 0)
83
+ await Promise.all(images.map((img) => img.resolveIntrinsicSize()));
54
84
  document.calculateLayout(new BoxConstraints(), { x: 0, y: 0 }, ctx);
55
85
  // Render pages and contents (the driver paginates overflowing pages).
56
86
  await PDFDocumentRenderer.render(document, objectManager, ctx);
@@ -66,11 +96,20 @@ export class PDFRenderer {
66
96
  if (outputIntent) {
67
97
  catalogParts.push(`/OutputIntents [${outputIntent} 0 R]`);
68
98
  }
99
+ // The catalog has exactly one /Names dict; embedded files and named destinations are two entries in
100
+ // it, so collect both and emit the /Names once (below).
101
+ const namesParts = [];
69
102
  const attachments = objectManager.getAttachments();
70
103
  if (attachments.length > 0) {
71
- const names = attachments.map((a) => `(${a.name}) ${a.filespec} 0 R`).join(" ");
104
+ // The name-tree key is a PDF literal string, so it needs the same escaping the /Filespec's /F and
105
+ // /UF get in attachFile(). Unescaped, a ")" or "\" in a file name closes the string early and the
106
+ // rest of it leaks out as raw operators.
107
+ const names = attachments
108
+ .map((a) => `(${PdfBackend.escapePdfString(a.name)}) ${a.filespec} 0 R`)
109
+ .join(" ");
72
110
  const af = attachments.map((a) => `${a.filespec} 0 R`).join(" ");
73
- catalogParts.push(`/AF [${af}]`, `/Names << /EmbeddedFiles << /Names [${names}] >> >>`);
111
+ catalogParts.push(`/AF [${af}]`);
112
+ namesParts.push(`/EmbeddedFiles << /Names [${names}] >>`);
74
113
  }
75
114
  // Accessible tagging: finalize the structure tree (emits StructTreeRoot + StructElems + ParentTree) and
76
115
  // fold /MarkInfo + /StructTreeRoot + /Lang into the catalog. Returns "" (no-op) when tagging is off.
@@ -80,6 +119,18 @@ export class PDFRenderer {
80
119
  // PDF/UA requires the viewer to show the document title (not the file name).
81
120
  catalogParts.push("/ViewerPreferences << /DisplayDocTitle true >>");
82
121
  }
122
+ // Document outline (bookmarks): emit the /Outlines tree collected during page rendering. Returns
123
+ // "" (no-op) when no Bookmark was placed, so a plain document's catalog is unchanged.
124
+ const outlineCatalog = objectManager.outline.finalize(objectManager);
125
+ if (outlineCatalog)
126
+ catalogParts.push(outlineCatalog);
127
+ // Named destinations (internal-link targets): a /Dests entry in the shared /Names dict.
128
+ const destsNames = objectManager.dests.finalize();
129
+ if (destsNames)
130
+ namesParts.push(destsNames);
131
+ // Emit the single /Names dict if either embedded files or destinations contributed to it.
132
+ if (namesParts.length > 0)
133
+ catalogParts.push(`/Names << ${namesParts.join(" ")} >>`);
83
134
  const catalogObject = `<< ${catalogParts.join(" ")} >>`;
84
135
  objectManager.addObject(catalogObject);
85
136
  // Now that the render pass has revealed which glyphs each embedded font uses, fill the reserved
@@ -0,0 +1,7 @@
1
+ import { PDFObjectManager } from "../utils/pdf-object-manager.js";
2
+ import { RotatedElement } from "../elements/layout/rotated-element.js";
3
+ import { RotatedBoxElement } from "../elements/layout/rotated-box-element.js";
4
+ import { IRNode } from "../ir/display-list.js";
5
+ export declare class RotatedRenderer {
6
+ static render(element: RotatedElement | RotatedBoxElement, objectManager: PDFObjectManager): Promise<IRNode[]>;
7
+ }
@@ -0,0 +1,29 @@
1
+ import { RendererRegistry } from "../utils/renderer-registry.js";
2
+ export class RotatedRenderer {
3
+ // Both `Rotated` (paint-only) and `RotatedBox` (layout-aware quarter-turns) expose the same
4
+ // `{ angle, child, x, y, width, height }` and rotate the child around the box center - shared here.
5
+ static async render(element, objectManager) {
6
+ const { angle, child, x, y, width, height } = element.getProps();
7
+ // Rotate around the box CENTER, in the engine's top-left coordinates (the IR->backend seam flips
8
+ // the matrix to PDF space per page). A positive angle reads clockwise on the page.
9
+ const cx = x + width / 2;
10
+ const cy = y + height / 2;
11
+ const theta = (angle * Math.PI) / 180;
12
+ const cos = Math.cos(theta);
13
+ const sin = Math.sin(theta);
14
+ // M = T(cx,cy) · R(theta) · T(-cx,-cy) as a PDF cm [a,b,c,d,e,f]: the linear part is the rotation,
15
+ // the translation keeps the center fixed.
16
+ const matrix = [
17
+ cos,
18
+ sin,
19
+ -sin,
20
+ cos,
21
+ cx - cos * cx + sin * cy,
22
+ cy - sin * cx - cos * cy,
23
+ ];
24
+ const renderer = RendererRegistry.getRenderer(child);
25
+ const childNodes = renderer ? await renderer(child, objectManager) : [];
26
+ // Wrap the child's drawing between a push (apply the matrix) and a pop (restore).
27
+ return [{ type: "transform-push", matrix }, ...childNodes, { type: "transform-pop" }];
28
+ }
29
+ }
@@ -4,7 +4,7 @@ import type { FontMetrics } from "../utils/font-metrics.js";
4
4
  import { IRNode } from "../ir/display-list.js";
5
5
  import { TextOverflow } from "../text/line-breaker.js";
6
6
  export declare class TextRenderer {
7
- static calculateTextHeight(content: string | TextSegment[], fontSize: number, fontFamily: string, fontStyle: FontStyle, objectManager: FontMetrics, maxWidth: number, maxLines?: number, overflow?: TextOverflow, lineHeight?: number): number;
7
+ static calculateTextHeight(content: string | TextSegment[], fontSize: number, fontFamily: string, fontStyle: FontStyle, objectManager: FontMetrics, maxWidth: number, maxLines?: number, overflow?: TextOverflow, lineHeight?: number, letterSpacing?: number): number;
8
8
  static render(textElement: TextElement, objectManager: PDFObjectManager): Promise<IRNode[]>;
9
9
  private static _expandColorGlyphs;
10
10
  private static _glyphToPath;