@jasy/pdf 1.0.0-alpha.10 → 1.0.0-alpha.12

package/dist/api/structure.d.ts

@@ -151,6 +151,13 @@ export interface RenderOptions {
     /** What to do when content is taller than a page and cannot break: `"error"` throws (default),
      *  `"warn"` logs and clips, `"ignore"` clips silently. It is always clipped either way. */
     onOverflow?: OverflowPolicy;
+    /**
+     * Called once, after the render, with the characters no font could draw - they were removed rather
+     * than emitted as `.notdef`, which PDF/A forbids. Deliberately a callback and not a policy: a
+     * missing tick mark must never stop an invoice from being produced, but the application should be
+     * able to SAY which character went missing, where a server log would never be read.
+     */
+    onMissingGlyphs?: (characters: string[]) => void;
     /** Encrypt the PDF with a password (AES-256, the newest standard). NOT compatible with PDF/A
      *  (ZUGFeRD invoices) - encrypting one throws, since PDF/A forbids encryption. */
     encrypt?: EncryptOptions;

package/dist/api/structure.js

@@ -143,10 +143,14 @@ export async function renderPdf(doc, options) {
     // A throwaway PDFDocument whose build() yields this tree, reusing the engine's standard
     // font registration + config handling (the constructor does both). Custom fonts are
     // registered here, before layout/render, so both the metrics and the backend see them.
+    // Captured so the caller can be told what had to be dropped: `render()` is static and builds its own
+    // instance, so this is the only place the object manager surfaces.
+    let manager;
     const Anon = class extends PDFDocument {
         constructor() {
             super(config);
             const om = this.objectManager;
+            manager = om;
             om.setCompress(options?.compress !== false); // FlateDecode streams by default
             om.setOverflowPolicy(options?.onOverflow ?? "error");
             om.setKerning(options?.kerning !== false); // on by default (experimental); `kerning:false` opts out
@@ -207,7 +211,11 @@ export async function renderPdf(doc, options) {
             return doc;
         }
     };
-    return Anon.render();
+    const pdf = await Anon.render();
+    const missing = manager?.getMissingGlyphs() ?? [];
+    if (missing.length > 0)
+        options?.onMissingGlyphs?.(missing.map((cp) => String.fromCodePoint(cp)));
+    return pdf;
 }
 /** Renders a `Document(...)` tree to PDF bytes (e.g. for a download / save dialog). */
 export async function renderToBytes(doc, options) {

package/dist/api/text.d.ts

@@ -3,6 +3,7 @@ import { TextOverflow } from "../text/line-breaker.js";
 import type { Direction } from "../text/bidi.js";
 import { ResolvedTextStyle, type TextTransform } from "../text/text-style.js";
 import { ColorInput } from "./color.js";
+import type { Hyphenator } from "../text/word-splitting.js";
 export type { TextOverflow };
 /** Text styling shared by `Text`, `Paragraph` and `span`. `bold`/`italic` are booleans
  *  (locked §7.3) and combine into one engine `FontStyle`. */
@@ -35,6 +36,13 @@ export interface TextStyle {
     wordSpacing?: number;
     /** CSS `text-indent`, in points: how far the FIRST line of a paragraph starts in. */
     textIndent?: number;
+    /** Split a word that is wider than its box, anywhere, no hyphen (CSS `overflow-wrap:
+     *  break-word`). The floor under `hyphenate`: an e-mail, an IBAN or an invoice number has no
+     *  valid hyphenation point anywhere. Off by default - a too-wide word overflows, as in CSS. */
+    breakWord?: boolean;
+    /** Language-aware splitting: given a word, return its parts. A hyphen is drawn at the break.
+     *  Pluggable, so no pattern data ships here: pass `hyphen`, `hyphenopoly`, or your own. */
+    hyphenate?: Hyphenator;
     /** CSS `vertical-align` for a SPAN: raises a footnote marker or lowers an index. It shifts the
      *  baseline only - pass a smaller `size` too if you want the browser's `<sup>` look. */
     verticalAlign?: VerticalTextAlign;
@@ -108,6 +116,13 @@ export interface TextDefaults {
     wordSpacing?: number;
     /** CSS `text-indent`, in points: how far the FIRST line of a paragraph starts in. */
     textIndent?: number;
+    /** Split a word that is wider than its box, anywhere, no hyphen (CSS `overflow-wrap:
+     *  break-word`). The floor under `hyphenate`: an e-mail, an IBAN or an invoice number has no
+     *  valid hyphenation point anywhere. Off by default - a too-wide word overflows, as in CSS. */
+    breakWord?: boolean;
+    /** Language-aware splitting: given a word, return its parts. A hyphen is drawn at the break.
+     *  Pluggable, so no pattern data ships here: pass `hyphen`, `hyphenopoly`, or your own. */
+    hyphenate?: Hyphenator;
 }
 /** Maps the `Text`-style option names onto a partial engine `ResolvedTextStyle` (only the set
  *  fields), for seeding the inheritance cascade. */

package/dist/api/text.js

@@ -100,6 +100,8 @@ export function Text(content, opts = {}) {
         textTransform: opts.textTransform,
         wordSpacing: opts.wordSpacing,
         textIndent: opts.textIndent,
+        breakWord: opts.breakWord,
+        hyphenate: opts.hyphenate,
         role: opts.role,
     });
 }
@@ -142,5 +144,9 @@ export function toTextStyleOverride(opts) {
         style.wordSpacing = opts.wordSpacing;
     if (opts.textIndent !== undefined)
         style.textIndent = opts.textIndent;
+    if (opts.breakWord !== undefined)
+        style.breakWord = opts.breakWord;
+    if (opts.hyphenate !== undefined)
+        style.hyphenate = opts.hyphenate;
     return style;
 }

package/dist/elements/text-element.d.ts

@@ -6,6 +6,7 @@ import { type TextTransform } from "../text/text-style.js";
 import { Fragmentable, FragmentResult } from "../layout/fragmentation.js";
 import { TextOverflow } from "../text/line-breaker.js";
 import { HorizontalAlignment, LayoutContext, SizedPDFElement } from "./pdf-element.js";
+import type { Hyphenator } from "../text/word-splitting.js";
 export interface TextSegment {
     content: string;
     fontStyle?: FontStyle;
@@ -74,6 +75,10 @@ interface TextElementParams {
     wordSpacing?: number;
     /** CSS `text-indent`, in points: how far the first line starts in. */
     textIndent?: number;
+    /** CSS `overflow-wrap: break-word` - split a word wider than its box, anywhere. */
+    breakWord?: boolean;
+    /** Language-aware splitting; a hyphen is drawn at the break. See text/word-splitting.ts. */
+    hyphenate?: Hyphenator;
     /** Accessibility role for the tagged structure tree (heading level or paragraph; default `"p"`). */
     role?: TextRole;
 }
@@ -93,6 +98,8 @@ export declare class TextElement extends SizedPDFElement implements Fragmentable
     private readonly rawTextTransform?;
     private readonly rawWordSpacing?;
     private readonly rawTextIndent?;
+    private readonly rawBreakWord?;
+    private readonly rawHyphenate?;
     private fontSize;
     private fontFamily;
     private fontStyle;
@@ -109,13 +116,15 @@ export declare class TextElement extends SizedPDFElement implements Fragmentable
     private textTransform;
     private wordSpacing;
     private textIndent;
+    private breakWord;
+    private hyphenate?;
     private content;
     private maxLines?;
     private orphans?;
     private widows?;
     private overflow;
     private readonly role?;
-    constructor({ fontSize, content, fontFamily, fontStyle, color, textAlignment, maxLines, orphans, widows, overflow, lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, fontFallback, textTransform, wordSpacing, textIndent, role, }: TextElementParams);
+    constructor({ fontSize, content, fontFamily, fontStyle, color, textAlignment, maxLines, orphans, widows, overflow, lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, fontFallback, textTransform, wordSpacing, textIndent, breakWord, hyphenate, role, }: TextElementParams);
     private resolveStyle;
     /** Justified lines may be squeezed to keep a word; every other alignment gets no slack. Read by the
      *  breaker in BOTH passes, or a measured line and a drawn one would disagree about where it ends. */
@@ -125,6 +134,13 @@ export declare class TextElement extends SizedPDFElement implements Fragmentable
      * same width in most fonts).
      */
     private display;
+    /**
+     * Drops or substitutes whatever the RESOLVED font cannot draw. Runs last, after the fallback stack
+     * has had its say, and before anything is measured - a character removed at draw time only would
+     * make the measured line and the drawn one disagree. A font that draws everything gets its input
+     * back unchanged, so an ordinary document is byte-identical.
+     */
+    private covered;
     /** The content with `text-transform` applied, and nothing else. */
     private recased;
     private lineOptions;
@@ -174,6 +190,8 @@ export declare class TextElement extends SizedPDFElement implements Fragmentable
         direction: Direction;
         wordSpacing: number;
         textIndent: number;
+        breakWord: boolean;
+        hyphenate: Hyphenator | undefined;
         role: TextRole | undefined;
     };
 }

package/dist/elements/text-element.js

@@ -10,6 +10,8 @@ import { singleLineWidth, MAX_SPACE_SHRINK, wrapStringIntoLines, breakSegmentsIn
 import { adjustForOrphansWidows, DEFAULT_ORPHANS, DEFAULT_WIDOWS, } from "../text/orphans-widows.js";
 import { lineBoxForSegmentLine, lineBoxForString } from "../text/line-metrics.js";
 import { HorizontalAlignment, SizedPDFElement } from "./pdf-element.js";
+import { normalizeContent } from "../text/whitespace.js";
+import { coverText } from "../text/glyph-coverage.js";
 /** How far a run sits off the baseline, as a fraction of its font size (up is positive). */
 export const VERTICAL_TEXT_SHIFT = {
     baseline: 0,
@@ -28,7 +30,7 @@ function finitePoints(value, name) {
     return value;
 }
 export class TextElement extends SizedPDFElement {
-    constructor({ fontSize, content, fontFamily, fontStyle, color, textAlignment, maxLines, orphans, widows, overflow = "clip", lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, fontFallback, textTransform, wordSpacing, textIndent, role, }) {
+    constructor({ fontSize, content, fontFamily, fontStyle, color, textAlignment, maxLines, orphans, widows, overflow = "clip", lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, fontFallback, textTransform, wordSpacing, textIndent, breakWord, hyphenate, role, }) {
         super({ x: 0, y: 0 });
         this.role = role;
         this.rawFontSize = fontSize;
@@ -46,7 +48,11 @@ export class TextElement extends SizedPDFElement {
         this.rawTextTransform = textTransform;
         this.rawWordSpacing = finitePoints(wordSpacing, "wordSpacing");
         this.rawTextIndent = finitePoints(textIndent, "textIndent");
-        this.content = content;
+        this.rawBreakWord = breakWord;
+        this.rawHyphenate = hyphenate;
+        // Control characters have no glyph and would be DRAWN as .notdef boxes; `\n` survives
+        // because the breaker treats it as a hard break. See text/whitespace.ts.
+        this.content = normalizeContent(content);
         this.maxLines = maxLines;
         this.orphans = orphans;
         this.widows = widows;
@@ -69,8 +75,10 @@ export class TextElement extends SizedPDFElement {
         // A SPAN may bring its own stack, so the element's being empty is not enough to skip the pass.
         const anyStack = this.fontFallback.length > 0 ||
             (typeof cased !== "string" && cased.some((seg) => (seg.fontFallback?.length ?? 0) > 0));
-        if (!anyStack || !metrics)
+        if (!metrics)
             return cased;
+        if (!anyStack)
+            return this.covered(cased, metrics);
         // Font fallback turns the content into spans - one per family - which every later pass already
         // knows how to handle. Nothing new downstream.
         const pieces = typeof cased === "string" ? [{ content: cased }] : cased;
@@ -83,7 +91,28 @@ export class TextElement extends SizedPDFElement {
                 for (const run of runs)
                     out.push({ ...seg, content: run.text, fontFamily: run.fontFamily });
         }
-        return out;
+        return this.covered(out, metrics);
+    }
+    /**
+     * Drops or substitutes whatever the RESOLVED font cannot draw. Runs last, after the fallback stack
+     * has had its say, and before anything is measured - a character removed at draw time only would
+     * make the measured line and the drawn one disagree. A font that draws everything gets its input
+     * back unchanged, so an ordinary document is byte-identical.
+     */
+    covered(content, metrics) {
+        const report = (dropped) => metrics.reportMissingGlyph?.(dropped);
+        if (typeof content === "string") {
+            const { text, dropped } = coverText(content, this.fontFamily, this.fontStyle, metrics);
+            if (dropped.length > 0)
+                report(dropped);
+            return text;
+        }
+        return content.map((seg) => {
+            const { text, dropped } = coverText(seg.content, seg.fontFamily ?? this.fontFamily, seg.fontStyle ?? this.fontStyle, metrics);
+            if (dropped.length > 0)
+                report(dropped);
+            return text === seg.content ? seg : { ...seg, content: text };
+        });
     }
     /** The content with `text-transform` applied, and nothing else. */
     recased() {
@@ -104,6 +133,7 @@ export class TextElement extends SizedPDFElement {
         return {
             wordSpacing: this.wordSpacing,
             indent: this.textIndent,
+            splitting: { breakWord: this.breakWord, hyphenate: this.hyphenate },
             // Justified lines may be squeezed to keep a word; every other alignment gets no slack.
             shrink: this.textAlignment === HorizontalAlignment.justify ? MAX_SPACE_SHRINK : 0,
         };
@@ -124,6 +154,8 @@ export class TextElement extends SizedPDFElement {
         this.textTransform = this.rawTextTransform ?? ts.textTransform;
         this.wordSpacing = this.rawWordSpacing ?? ts.wordSpacing;
         this.textIndent = this.rawTextIndent ?? ts.textIndent;
+        this.breakWord = this.rawBreakWord ?? ts.breakWord;
+        this.hyphenate = this.rawHyphenate ?? ts.hyphenate;
     }
     /**
      * Splits the paragraph at line boxes (Slice 1). The lines that fit in `maxHeight` stay;
@@ -281,6 +313,8 @@ export class TextElement extends SizedPDFElement {
             direction: this.direction,
             wordSpacing: this.wordSpacing,
             textIndent: this.textIndent,
+            breakWord: this.breakWord,
+            hyphenate: this.hyphenate,
             role: this.role,
         };
     }

package/dist/renderer/text-renderer.js

@@ -18,17 +18,25 @@ export class TextRenderer {
             const box = lineBoxForString(objectManager, fontFamily, fontStyle, fontSize, lineHeight);
             return lines.length * box.height;
         }
-        // Segments: each line's box comes from the fonts actually on it.
-        const defaults = { fontFamily, fontSize, fontStyle, letterSpacing };
+        // Segments: each line's box comes from the fonts actually on it. `splitting` has to travel with
+        // them for the same reason `letterSpacing` does: it changes where a line ENDS, so leaving it out
+        // measures a paragraph at one line count and draws it at another.
+        const defaults = {
+            fontFamily,
+            fontSize,
+            fontStyle,
+            letterSpacing,
+            splitting: lineOptions.splitting,
+        };
         const lines = breakSegmentsIntoLines(content, defaults, maxWidth, objectManager, maxLines, overflow);
         return lines.reduce((total, line) => total + lineBoxForSegmentLine(line, defaults, objectManager, lineHeight).height, 0);
     }
     static async render(textElement, objectManager) {
-        const { x, y, width, fontSize, color, content, fontFamily, fontStyle, textAlignment, maxLines, overflow, lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, wordSpacing, textIndent, role, } = textElement.getProps();
+        const { x, y, width, fontSize, color, content, fontFamily, fontStyle, textAlignment, maxLines, overflow, lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, wordSpacing, textIndent, breakWord, hyphenate, role, } = textElement.getProps();
         // Component -> display list. Wrapping and positioning stay here; the backend
         // turns each run into BT/Tf/Td/Tj/ET. The wrapping algorithm is unchanged from
         // the original renderer - unifying it into the engine is Phase 3.
-        const { runs, links, decorations } = TextRenderer._buildRuns(content, fontSize, fontFamily, fontStyle, objectManager, width ?? Number.NaN, textAlignment, color, x, y, maxLines, overflow, lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, wordSpacing, textIndent);
+        const { runs, links, decorations } = TextRenderer._buildRuns(content, fontSize, fontFamily, fontStyle, objectManager, width ?? Number.NaN, textAlignment, color, x, y, maxLines, overflow, lineHeight, underline, strikethrough, skipInk, letterSpacing, direction, wordSpacing, textIndent, breakWord, hyphenate);
         // Accessible tagging: this whole text block is one structure element (a paragraph P, or a heading
         // from `role`); each run's text lands under it. Emoji sub-nodes produced below stay untagged (drawn
         // as artifacts). Keyed by structId so a block split across pages stays one element.
@@ -224,7 +232,7 @@ export class TextRenderer {
     // Lay the content out into absolutely-positioned text runs. Glyph positions match
     // the previous hand-written operators exactly (verified by pixel-identical render);
     // only the output form changed from PDF strings to `TextRun`s.
-    static _buildRuns(content, fontSize, fontFamily, fontStyle, objectManager, maxWidth, textAlignment, color, initialX, yPosition, maxLines, overflow, lineHeight, underline = false, strikethrough = false, skipInk = false, letterSpacing = 0, direction = "ltr", wordSpacing = 0, textIndent = 0) {
+    static _buildRuns(content, fontSize, fontFamily, fontStyle, objectManager, maxWidth, textAlignment, color, initialX, yPosition, maxLines, overflow, lineHeight, underline = false, strikethrough = false, skipInk = false, letterSpacing = 0, direction = "ltr", wordSpacing = 0, textIndent = 0, breakWord = false, hyphenate) {
         const runs = [];
         // /Link annotation rects for any `href` spans, collected as we place segments. Empty for the
         // common (no-href) case, so plain text keeps producing exactly the runs it did before.
@@ -371,6 +379,7 @@ export class TextRenderer {
                 wordSpacing,
                 indent: textIndent,
                 shrink: align === HorizontalAlignment.justify ? MAX_SPACE_SHRINK : 0,
+                splitting: { breakWord, hyphenate },
             });
             // yPosition is the top of the text box (top-left). The line box seats its own baseline; lines
             // then stack by that box's height. Same numbers as calculateTextHeight, by construction.
@@ -457,7 +466,13 @@ export class TextRenderer {
         // The seam flips the whole thing to PDF space. `letterSpacing` MUST be in the defaults so a span
         // that does not override it wraps and aligns with the element's spacing - the same defaults the
         // measure path (calculateTextHeight) uses, or segmented spaced text mis-wraps (measured != drawn).
-        const defaults = { fontFamily, fontSize, fontStyle, letterSpacing };
+        const defaults = {
+            fontFamily,
+            fontSize,
+            fontStyle,
+            letterSpacing,
+            splitting: { breakWord, hyphenate },
+        };
         let top = yPosition;
         // Materialised, because justification has to know which line is the LAST one of the paragraph.
         const segmentLines = [

package/dist/text/glyph-coverage.d.ts

@@ -0,0 +1,15 @@
+import type { FontMetrics } from "../utils/font-metrics.js";
+import type { FontStyle } from "../utils/pdf-object-manager.js";
+export interface CoverageResult {
+    /** The text as it can actually be drawn. */
+    text: string;
+    /** Code points that had to go - empty for every document that never hits this. */
+    dropped: number[];
+}
+/**
+ * Replaces or removes what `fontFamily` cannot draw. Returns the input unchanged (and no allocation
+ * beyond the scan) when everything is drawable, which is the overwhelmingly common case.
+ */
+export declare function coverText(text: string, fontFamily: string, fontStyle: FontStyle, metrics: FontMetrics): CoverageResult;
+/** `U+2713` - how a dropped code point is named to the caller. */
+export declare function formatCodePoint(cp: number): string;

package/dist/text/glyph-coverage.js

@@ -0,0 +1,77 @@
+/**
+ * A code point the chosen font cannot draw becomes glyph 0 - which IS `.notdef`, and referencing it is
+ * forbidden by PDF/A (ISO 19005-3, 6.2.11.8). So an invoice carrying a "✓" in its description is a
+ * conforming file right up until someone validates it.
+ *
+ * The check has to be DYNAMIC, against the font actually resolved: which characters are missing is a
+ * property of that font, not of Unicode. `@jasy/e-invoice` embeds Liberation, and its gaps look
+ * arbitrary from the outside - "→" draws, "⇒" does not; "₴" draws, "₽" does not.
+ *
+ * What we do about it, in this order:
+ *   1. The font has it        -> keep, untouched. This is every ordinary document, unchanged.
+ *   2. A plain equivalent MEANS the same and the font has THAT -> substitute.
+ *   3. Otherwise              -> drop it, and report it, so the application can say which character
+ *                                went missing instead of shipping a box or a broken file.
+ */
+/**
+ * Substitutes that keep the meaning. Deliberately tiny: only where dropping would CHANGE the text
+ * rather than thin it. `E‑Rechnung` with a non-breaking hyphen must not become `ERechnung`.
+ *
+ * Not a list of "what Word pastes" - most of that (quotes, dashes, the ellipsis) is in WinAnsi and in
+ * every normal font, so it never reaches rule 3. Entries earn their place by being measured, not
+ * guessed: each one is a code point seen missing from a real embedded font.
+ */
+const EQUIVALENTS = {
+    0x2010: "-", // hyphen
+    0x2011: "-", // non-breaking hyphen - the one that turns E‑Rechnung into ERechnung
+    0x2012: "-", // figure dash
+    0x2015: "-", // horizontal bar
+    0x2007: " ", // figure space
+    0x2008: " ", // punctuation space
+    0x2009: " ", // thin space
+    0x200a: " ", // hair space
+    0x202f: " ", // narrow no-break space
+    0x2032: "'", // prime
+    0x2033: '"', // double prime
+    0x02bc: "'", // modifier letter apostrophe
+};
+/**
+ * Replaces or removes what `fontFamily` cannot draw. Returns the input unchanged (and no allocation
+ * beyond the scan) when everything is drawable, which is the overwhelmingly common case.
+ */
+export function coverText(text, fontFamily, fontStyle, metrics) {
+    let out = null; // built only once something has to change
+    const dropped = [];
+    let index = 0;
+    for (const char of text) {
+        const cp = char.codePointAt(0);
+        // A space and a line break are handled by the breaker, never drawn as a glyph.
+        const drawable = cp === 0x20 ||
+            cp === 0x0a ||
+            metrics.hasGlyph(cp, fontFamily, fontStyle) ||
+            // Colour emoji is drawn from a fallback font or an image, NOT from the text font - it is missing
+            // there by design, and removing it would delete the very thing `Document({ emoji })` exists for.
+            metrics.rendersAsEmoji?.(cp, fontFamily, fontStyle) === true;
+        if (!drawable) {
+            if (out === null)
+                out = text.slice(0, index);
+            const equivalent = EQUIVALENTS[cp];
+            if (equivalent !== undefined &&
+                metrics.hasGlyph(equivalent.codePointAt(0), fontFamily, fontStyle)) {
+                out += equivalent;
+            }
+            else {
+                dropped.push(cp);
+            }
+        }
+        else if (out !== null) {
+            out += char;
+        }
+        index += char.length;
+    }
+    return out === null ? { text, dropped } : { text: out, dropped };
+}
+/** `U+2713` - how a dropped code point is named to the caller. */
+export function formatCodePoint(cp) {
+    return "U+" + cp.toString(16).toUpperCase().padStart(4, "0");
+}

package/dist/text/line-breaker.d.ts

@@ -1,6 +1,7 @@
 import type { FontStyle } from "../utils/pdf-object-manager.js";
 import type { FontMetrics } from "../utils/font-metrics.js";
 import type { TextSegment } from "../elements/text-element.js";
+import { type WordSplitting } from "./word-splitting.js";
 /**
  * How far a JUSTIFIED line's spaces may be squeezed to keep one more word on it.
  *
@@ -22,6 +23,8 @@ export interface LineOptions {
     indent?: number;
     /** How far a justified line's spaces may be squeezed to keep one more word (0 = not justified). */
     shrink?: number;
+    /** How a word wider than its box is split - hyphenation, break-word, or neither (the default). */
+    splitting?: WordSplitting;
 }
 /** Default font for segments that don't override it. */
 export interface SegmentDefaults {
@@ -30,6 +33,8 @@ export interface SegmentDefaults {
     fontStyle: FontStyle;
     /** Extra space after every glyph, in points; a segment may override it. Default 0. */
     letterSpacing?: number;
+    /** Same knobs as the string path - see `LineOptions.splitting`. */
+    splitting?: WordSplitting;
 }
 /** What happens to text beyond `maxLines`: `"clip"` drops it, `"ellipsis"` ends the last kept line
  *  with an ellipsis. Mirrors Flutter's `TextOverflow`. */

package/dist/text/line-breaker.js

@@ -1,4 +1,5 @@
 import { runAdvance } from "./advance.js";
+import { splitLongWord } from "./word-splitting.js";
 /**
  * How far a JUSTIFIED line's spaces may be squeezed to keep one more word on it.
  *
@@ -31,6 +32,9 @@ const ELLIPSIS = "...";
  * hypothetical: it is what split a footer sized to its own content.
  */
 export function singleLineWidth(text, font, metrics, letterSpacing = 0, wordSpacing = 0) {
+    // A hard break makes this several lines; the natural width is the widest of them.
+    if (text.includes("\n"))
+        return Math.max(...text.split("\n").map((p) => singleLineWidth(p, font, metrics, letterSpacing, wordSpacing)));
     const space = runAdvance(metrics, " ", font, letterSpacing) + wordSpacing;
     return text
         .split(" ")
@@ -39,50 +43,78 @@ export function singleLineWidth(text, font, metrics, letterSpacing = 0, wordSpac
         : width + (space + runAdvance(metrics, word, font, letterSpacing)), 0);
 }
 export function wrapStringIntoLines(text, fontFamily, fontSize, fontStyle, maxWidth, metrics, maxLines, overflow, letterSpacing = 0, options = {}) {
-    const { wordSpacing = 0, indent = 0, shrink = 0 } = options;
+    const { wordSpacing = 0, indent = 0, shrink = 0, splitting = {} } = options;
     let currentLine = "";
     let currentWidth = 0;
     let gaps = 0; // spaces on the current line, which is what a justified line can squeeze
     const lines = [];
     const font = { fontFamily, fontSize, fontStyle };
-    const words = text.split(" ");
-    words.forEach((word) => {
-        // Word and space advances come from the one shared primitive (`advance.ts`), the same one
-        // `naturalWidth` uses - so a bounded and an unbounded layout of the same text agree bit for bit.
-        const wordWidth = runAdvance(metrics, word, font, letterSpacing);
-        const spaceWidth = runAdvance(metrics, " ", font, letterSpacing) + wordSpacing;
-        // The FIRST line has the indent taken out of its room; every later one gets the full box.
-        const room = lines.length === 0 ? maxWidth - indent : maxWidth;
-        // Break before a word that won't fit - counting the SPACE that would join it, which the old test
-        // forgot. It went unnoticed on the first line, where the very first word added a space too many
-        // and cancelled the error out; after a break `currentWidth` is the bare word, so every following
-        // test was short by one space and the line could overrun its box by that much.
-        //
-        // A single word wider than maxWidth still sits on its (empty) line and overflows, rather than
-        // pushing a phantom empty line before it (which would over-count the height by a line).
-        const candidate = currentWidth + (spaceWidth + wordWidth);
-        if (currentLine === "") {
-            currentLine = word;
-            currentWidth = wordWidth;
-            gaps = 0;
-            // How much a JUSTIFIED line may be squeezed to keep one more word: the spaces already on the
-            // line plus the one that would join it, each giving up at most `shrink` of its width. Zero for
-            // every other alignment, where a squeezed line would simply overflow instead.
-            // ONE expression, used for the test AND for the running total. Adding the same three numbers in
-            // a different order gives a different last bit, and a line that lands exactly on `maxWidth` is
-            // then broken by that bit alone - which is how a footer sized to its own text wrapped.
-        }
-        else if (candidate - (gaps + 1) * spaceWidth * shrink > room) {
+    // A `\n` is a HARD break: wrap each paragraph on its own, then cut. An EMPTY paragraph still has to
+    // produce a line, or a deliberate blank line between two blocks silently disappears.
+    text.split("\n").forEach((paragraph, paragraphIndex) => {
+        if (paragraphIndex > 0) {
             lines.push(currentLine.trim());
-            currentLine = word;
-            currentWidth = wordWidth;
+            currentLine = "";
+            currentWidth = 0;
             gaps = 0;
         }
-        else {
-            currentLine += " " + word;
-            currentWidth = candidate;
-            gaps += 1;
-        }
+        const words = paragraph.split(" ");
+        words.forEach((word) => {
+            // Word and space advances come from the one shared primitive (`advance.ts`), the same one
+            // `naturalWidth` uses - so a bounded and an unbounded layout of the same text agree bit for bit.
+            const wordWidth = runAdvance(metrics, word, font, letterSpacing);
+            const spaceWidth = runAdvance(metrics, " ", font, letterSpacing) + wordSpacing;
+            // The FIRST line has the indent taken out of its room; every later one gets the full box.
+            const room = lines.length === 0 ? maxWidth - indent : maxWidth;
+            // Break before a word that won't fit - counting the SPACE that would join it, which the old test
+            // forgot. It went unnoticed on the first line, where the very first word added a space too many
+            // and cancelled the error out; after a break `currentWidth` is the bare word, so every following
+            // test was short by one space and the line could overrun its box by that much.
+            //
+            // A single word wider than maxWidth still sits on its (empty) line and overflows, rather than
+            // pushing a phantom empty line before it (which would over-count the height by a line).
+            // A word too wide for the box, split into pieces that fit (hyphenation first, break-word as the
+            // floor). Every piece but the last closes a line; the last carries on as an ordinary word. Off by
+            // default, so a document that asks for neither reaches the untouched path below.
+            // The pieces land on lines that are NOT the first one whenever the current line is closed below,
+            // so they get the full box - `room` only loses the indent while we are genuinely still filling
+            // the indented first line.
+            const splitRoom = lines.length === 0 && currentLine === "" ? room : maxWidth;
+            const split = splitLongWord(word, splitRoom, font, metrics, letterSpacing, splitting);
+            if (split) {
+                if (currentLine !== "")
+                    lines.push(currentLine.trim());
+                for (const piece of split.slice(0, -1))
+                    lines.push(piece);
+                currentLine = split[split.length - 1];
+                currentWidth = runAdvance(metrics, currentLine, font, letterSpacing);
+                gaps = 0;
+                return;
+            }
+            const candidate = currentWidth + (spaceWidth + wordWidth);
+            if (currentLine === "") {
+                currentLine = word;
+                currentWidth = wordWidth;
+                gaps = 0;
+                // How much a JUSTIFIED line may be squeezed to keep one more word: the spaces already on the
+                // line plus the one that would join it, each giving up at most `shrink` of its width. Zero for
+                // every other alignment, where a squeezed line would simply overflow instead.
+                // ONE expression, used for the test AND for the running total. Adding the same three numbers in
+                // a different order gives a different last bit, and a line that lands exactly on `maxWidth` is
+                // then broken by that bit alone - which is how a footer sized to its own text wrapped.
+            }
+            else if (candidate - (gaps + 1) * spaceWidth * shrink > room) {
+                lines.push(currentLine.trim());
+                currentLine = word;
+                currentWidth = wordWidth;
+                gaps = 0;
+            }
+            else {
+                currentLine += " " + word;
+                currentWidth = candidate;
+                gaps += 1;
+            }
+        });
     });
     if (currentLine)
         lines.push(currentLine.trim());
@@ -116,41 +148,80 @@ export function breakSegmentsIntoLines(segments, defaults, maxWidth, metrics, ma
         const letterSpacing = segment.letterSpacing ?? defaults.letterSpacing ?? 0;
         const font = { fontFamily: family, fontSize: size, fontStyle: style };
         const spaceWidth = runAdvance(metrics, " ", font, letterSpacing);
-        const words = segment.content.split(" ");
-        // Start this segment's piece empty; its content is filled word-by-word below. (Not
-        // the original content - otherwise a segment whose FIRST word overflows would carry
-        // its whole text into the line that just closed.)
-        lineSegments.push({ ...segment, fontFamily: family, content: "" });
-        combined = "";
-        words.forEach((word, wordIndex) => {
-            const wordWidth = runAdvance(metrics, word, font, letterSpacing);
-            // A space joins this word to what precedes it only INSIDE a segment; segments butt together
-            // with nothing between them, which is what `span("a") + span("b")` draws.
-            const joiner = wordIndex > 0 ? spaceWidth : 0;
-            // ONE expression for the test and for the running total, grouped exactly as `singleLineWidth`
-            // groups it - see the note there. A width built any other way can be a bit off the one the box
-            // was sized with, and the text wraps inside a box made to hold it.
-            const candidate = width + (joiner + wordWidth);
-            // Same guard as the string path: don't open a phantom empty line for an over-wide first word -
-            // place it (overflowing) on the current empty line instead.
-            if (candidate > maxWidth && width > 0) {
+        // A `\n` inside a span is a HARD break, exactly as in the string path: close the line where it
+        // sits, whatever else was already on it, and start the next one with this segment's style.
+        segment.content.split("\n").forEach((paragraph, paragraphIndex) => {
+            if (paragraphIndex > 0) {
                 lines.push({ segments: lineSegments, width });
-                width = wordWidth;
+                width = 0;
                 lineSegments = [];
-                combined = word;
-                lineSegments.push({ ...segment, content: combined });
             }
-            else {
-                combined += wordIndex === 0 ? word : " " + word;
-                width = candidate;
-                if (lineSegments.length === 0) {
-                    lineSegments.push({ ...segment, fontFamily: family, content: combined });
+            const words = paragraph.split(" ");
+            // Start this segment's piece empty; its content is filled word-by-word below. (Not
+            // the original content - otherwise a segment whose FIRST word overflows would carry
+            // its whole text into the line that just closed.)
+            lineSegments.push({ ...segment, fontFamily: family, content: "" });
+            combined = "";
+            words.forEach((word, wordIndex) => {
+                const wordWidth = runAdvance(metrics, word, font, letterSpacing);
+                // Same treatment as the string path: a word too wide for the box is split into pieces that fit.
+                // Each finished piece closes the line it sits on, keeping this segment's style.
+                const split = splitLongWord(word, maxWidth, font, metrics, letterSpacing, defaults.splitting ?? {});
+                if (split) {
+                    // Close whatever is already on the line FIRST. The pieces were measured against the full box,
+                    // not against what is left of this line - appending the first one put a line 80 wide into a
+                    // 50 box, and reported 50, which is exactly why it went unnoticed. The string path has always
+                    // pushed its current line before splitting; this is the same move.
+                    if (width > 0) {
+                        lines.push({ segments: lineSegments, width });
+                        lineSegments = [{ ...segment, fontFamily: family, content: "" }];
+                        combined = "";
+                    }
+                    for (const piece of split.slice(0, -1)) {
+                        lineSegments[lineSegments.length - 1].content = piece;
+                        // The width a line REPORTS is what alignment and line-metrics work from: measure it.
+                        lines.push({
+                            segments: lineSegments,
+                            width: runAdvance(metrics, piece, font, letterSpacing),
+                        });
+                        lineSegments = [{ ...segment, fontFamily: family, content: "" }];
+                    }
+                    combined = split[split.length - 1];
+                    width = runAdvance(metrics, combined, font, letterSpacing);
+                    lineSegments[lineSegments.length - 1].content = combined;
+                    return;
                 }
-                lineSegments[lineSegments.length - 1].content = combined;
-            }
+                // A space joins this word to what precedes it only INSIDE a segment; segments butt together
+                // with nothing between them, which is what `span("a") + span("b")` draws.
+                const joiner = wordIndex > 0 ? spaceWidth : 0;
+                // ONE expression for the test and for the running total, grouped exactly as `singleLineWidth`
+                // groups it - see the note there. A width built any other way can be a bit off the one the box
+                // was sized with, and the text wraps inside a box made to hold it.
+                const candidate = width + (joiner + wordWidth);
+                // Same guard as the string path: don't open a phantom empty line for an over-wide first word -
+                // place it (overflowing) on the current empty line instead.
+                if (candidate > maxWidth && width > 0) {
+                    lines.push({ segments: lineSegments, width });
+                    width = wordWidth;
+                    lineSegments = [];
+                    combined = word;
+                    lineSegments.push({ ...segment, content: combined });
+                }
+                else {
+                    combined += wordIndex === 0 ? word : " " + word;
+                    width = candidate;
+                    if (lineSegments.length === 0) {
+                        lineSegments.push({ ...segment, fontFamily: family, content: combined });
+                    }
+                    lineSegments[lineSegments.length - 1].content = combined;
+                }
+            });
         });
     });
-    if (lineSegments.length > 0) {
+    // Only if something actually landed on it. The string path ends with `if (currentLine)`, so a text
+    // ending in `\n` is ONE line there - the two must not disagree about the same text. An empty
+    // paragraph in the MIDDLE still becomes a line; that one is pushed at the break, not here.
+    if (lineSegments.some((seg) => seg.content !== "")) {
         lines.push({ segments: lineSegments, width });
     }
     // Open-end by default; cap only when maxLines is set.

package/dist/text/text-style.d.ts

@@ -2,6 +2,7 @@ import { Color } from "../common/color.js";
 import { FontStyle } from "../utils/pdf-object-manager.js";
 import { HorizontalAlignment } from "../elements/pdf-element.js";
 import type { Direction } from "./bidi.js";
+import type { Hyphenator } from "./word-splitting.js";
 /** CSS `text-transform`. `capitalize` upper-cases the first letter of every word, as CSS does. */
 export type TextTransform = "none" | "uppercase" | "lowercase" | "capitalize";
 /**
@@ -50,6 +51,14 @@ export interface ResolvedTextStyle {
     wordSpacing: number;
     /** CSS `text-indent`, in points: how far the FIRST line of a paragraph starts in. */
     textIndent: number;
+    /** CSS `overflow-wrap: break-word`: split a word that is wider than its box, anywhere, no hyphen.
+     *  The floor under `hyphenate` - it is what handles an e-mail, an IBAN or an invoice number, which
+     *  have no valid hyphenation point anywhere. Default off: a too-wide word overflows, as in CSS. */
+    breakWord: boolean;
+    /** Language-aware word splitting: given a word, return its parts (`["Rechts", "schutz", ...]`). A
+     *  hyphen is drawn at the break. Pluggable on purpose - no pattern data ships in this package, so
+     *  there is no licence question and nothing is added to a bundle that never asks for it. */
+    hyphenate?: Hyphenator;
 }
 /**
  * The root of the cascade: what a `Text` falls back to when neither it nor any ancestor sets a

package/dist/text/text-style.js

@@ -54,6 +54,8 @@ export const DEFAULT_TEXT_STYLE = {
     textTransform: "none",
     wordSpacing: 0,
     textIndent: 0,
+    breakWord: false,
+    hyphenate: undefined,
 };
 /** Layers a partial override onto a complete style; an unset (undefined) field keeps the base. */
 export function mergeTextStyle(base, override) {
@@ -75,5 +77,7 @@ export function mergeTextStyle(base, override) {
         textTransform: override.textTransform ?? base.textTransform,
         wordSpacing: override.wordSpacing ?? base.wordSpacing,
         textIndent: override.textIndent ?? base.textIndent,
+        breakWord: override.breakWord ?? base.breakWord,
+        hyphenate: override.hyphenate ?? base.hyphenate,
     };
 }

package/dist/text/whitespace.d.ts

@@ -0,0 +1,4 @@
+import type { TextSegment } from "../elements/text-element.js";
+export declare function normalizeWhitespace(text: string): string;
+/** The same, for content that may already be split into styled spans. */
+export declare function normalizeContent(content: string | TextSegment[]): string | TextSegment[];

package/dist/text/whitespace.js

@@ -0,0 +1,43 @@
+/**
+ * A character with no glyph is still ENCODED and DRAWN: it comes out as the font's .notdef box, and
+ * veraPDF rejects the file for referencing .notdef (ISO 19005-3, 6.2.11.8). A `\n` in an invoice
+ * description was the first case we hit; zero-width characters are the same fault and arrive far more
+ * quietly - a paste out of Word or a web page carries them routinely.
+ *
+ * So nothing invisible survives into the breaker except `\n`, and that only because the breaker treats
+ * it as a hard break. The mapping follows a browser's `white-space: normal` apart from `\n`: CSS
+ * collapses that to a space, while a text COMPONENT breaks on it - Flutter's `Text` and react-pdf both
+ * do, and it is what anyone writing `"a\nb"` expects.
+ */
+/**
+ * Everything that MEANS a line break, matched as a WHOLE so a two-character ending never becomes two
+ * breaks. `\n\r` is in here beside `\r\n` because it is what a mistyped CRLF looks like, and a stray
+ * blank line is a worse answer than reading it as the one break it was meant to be. U+2028/U+2029 are
+ * Unicode's own line and paragraph separators.
+ */
+const LINE_BREAKS = /\r\n|\n\r|\r|\u2028|\u2029/g;
+/**
+ * Invisible and inert: a zero-width space and a byte-order mark carry no meaning we act on, so they are
+ * dropped rather than drawn. C0, C1 and DEL likewise.
+ *
+ * Deliberately NOT here: U+200C/U+200D (zero-width non-joiner and joiner) drive Arabic shaping and
+ * emoji sequences, and U+200E/U+200F (the bidi marks) drive `text/bidi.ts`. Dropping those would
+ * silently change how correct text is laid out - worse than the box they can still leave behind in a
+ * run that neither shapes nor reorders.
+ */
+// oxlint-disable-next-line no-control-regex -- matching control characters is the whole point here.
+const UNPRINTABLE = /[\u0000-\u0008\u000B\u000C\u000E-\u001F\u007F-\u009F\u200B\uFEFF]/g;
+export function normalizeWhitespace(text) {
+    return (text
+        // Every spelling of a break becomes `\n` first, so the breaker only ever sees one form.
+        .replace(LINE_BREAKS, "\n")
+        // A tab is collapsible whitespace; a PDF text operator has no tab stops to advance to.
+        .replace(/\t/g, " ")
+        .replace(UNPRINTABLE, ""));
+}
+/** The same, for content that may already be split into styled spans. */
+export function normalizeContent(content) {
+    if (typeof content === "string")
+        return normalizeWhitespace(content);
+    return content.map((seg) => ({ ...seg, content: normalizeWhitespace(seg.content) }));
+}

package/dist/text/word-splitting.d.ts

@@ -0,0 +1,42 @@
+import type { FontStyle } from "../utils/pdf-object-manager.js";
+import type { FontMetrics } from "../utils/font-metrics.js";
+/**
+ * What to do with a word that is wider than the box it has to fit in.
+ *
+ * Until now: nothing. It stayed on its line and drew straight over whatever was beside it - which on an
+ * invoice means a mandatory field (§14 UStG) painted across its own label. CSS overflows too, so this
+ * was never a bug; it was a missing feature, and it is the one that bites in German.
+ *
+ * Two layers, and they are NOT alternatives - CSS stacks them the same way:
+ *
+ *   1. HYPHENATION splits a word at a linguistically valid point and marks it with a hyphen:
+ *      `Rechtsschutzver-` / `sicherungsgesellschaften`. It needs language knowledge, which is why it is
+ *      PLUGGABLE rather than bundled - no pattern data in this package, no licence question, nothing
+ *      added to a bundle that never asks for it.
+ *   2. BREAK-WORD is the floor: split wherever the box ends, no hyphen. It is what handles the things
+ *      hyphenation cannot touch - an e-mail address, an IBAN, an invoice number. Those are single
+ *      "words" with no valid point anywhere, and they are exactly what an invoice is full of.
+ *
+ * With both on, hyphenation is tried first and break-word catches what it cannot split.
+ */
+/** Splits one word into its hyphenation points, e.g. `["Rechts", "schutz", "ver", "siche", ...]`. */
+export type Hyphenator = (word: string) => string[];
+export interface WordSplitting {
+    /** Language-aware splitting, with a hyphen at the break. Supply `hyphen`, `hyphenopoly`, your own. */
+    hyphenate?: Hyphenator;
+    /** Split anywhere as a last resort, without a hyphen. CSS `overflow-wrap: break-word`. */
+    breakWord?: boolean;
+}
+/**
+ * Splits `word` into pieces that each fit within `room`. Returns null when nothing applies - no
+ * splitting configured, or the word fits anyway - and the caller keeps its current behaviour, so a
+ * document that asks for neither is untouched.
+ *
+ * Every piece but the last is a finished line. The last is what carries on, so the caller keeps
+ * filling it as usual.
+ */
+export declare function splitLongWord(word: string, room: number, font: {
+    fontFamily: string;
+    fontSize: number;
+    fontStyle: FontStyle;
+}, metrics: FontMetrics, letterSpacing: number, options: WordSplitting): string[] | null;

package/dist/text/word-splitting.js

@@ -0,0 +1,72 @@
+import { runAdvance } from "./advance.js";
+/** The hyphen actually drawn at a hyphenation break. U+002D, in every font we can encode. */
+const HYPHEN = "-";
+/**
+ * Splits `word` into pieces that each fit within `room`. Returns null when nothing applies - no
+ * splitting configured, or the word fits anyway - and the caller keeps its current behaviour, so a
+ * document that asks for neither is untouched.
+ *
+ * Every piece but the last is a finished line. The last is what carries on, so the caller keeps
+ * filling it as usual.
+ */
+export function splitLongWord(word, room, font, metrics, letterSpacing, options) {
+    const { hyphenate, breakWord } = options;
+    if (!hyphenate && !breakWord)
+        return null;
+    if (room <= 0)
+        return null;
+    if (runAdvance(metrics, word, font, letterSpacing) <= room)
+        return null;
+    const width = (s) => runAdvance(metrics, s, font, letterSpacing);
+    const pieces = [];
+    let rest = word;
+    // Bounded by construction: every pass either emits a piece and shortens `rest`, or gives up.
+    while (width(rest) > room) {
+        // The layering: a linguistically valid split if one is offered and reaches far enough, otherwise
+        // the blunt one - which is what an e-mail address or an IBAN needs, having no valid point at all.
+        const head = (hyphenate ? longestHyphenatedPrefix(rest, room, width, hyphenate) : null) ??
+            (breakWord ? longestPrefix(rest, room, width) : null);
+        // Nothing fits - not even one character, or hyphenation found nothing and break-word is off.
+        // Leave the remainder whole and let it overflow, exactly as before.
+        if (head === null)
+            break;
+        pieces.push(head.text);
+        rest = head.rest;
+    }
+    if (pieces.length === 0)
+        return null;
+    return [...pieces, rest];
+}
+/** The longest prefix that fits, plus a hyphen - split only where the hyphenator allows. */
+function longestHyphenatedPrefix(word, room, width, hyphenate) {
+    const parts = hyphenate(word);
+    if (parts.length < 2)
+        return null;
+    let taken = "";
+    let best = null;
+    for (let i = 0; i < parts.length - 1; i++) {
+        taken += parts[i];
+        // The hyphen has to fit too, or the line it lands on overflows by exactly that hyphen.
+        if (width(taken + HYPHEN) > room)
+            break;
+        best = { text: taken + HYPHEN, rest: parts.slice(i + 1).join("") };
+    }
+    return best;
+}
+/** The longest prefix that fits, split anywhere. Code points, so an astral character stays whole. */
+function longestPrefix(word, room, width) {
+    const chars = [...word];
+    let taken = "";
+    let count = 0;
+    for (const char of chars) {
+        if (width(taken + char) > room)
+            break;
+        taken += char;
+        count += 1;
+    }
+    // Not even one character fits: the box is narrower than a single glyph. Splitting cannot help, and
+    // looping on it would never terminate.
+    if (count === 0 || count === chars.length)
+        return null;
+    return { text: taken, rest: chars.slice(count).join("") };
+}

package/dist/utils/font-metrics.d.ts

@@ -25,6 +25,12 @@ export interface FontMetrics {
     getKernPairs(text: string, fontFamily: string, fontStyle: FontStyle): number[];
     /** Whether this family can draw the code point at all - what picks a face out of a fallback stack. */
     hasGlyph(codePoint: number, fontFamily: string, fontStyle: FontStyle): boolean;
+    /** Whether the document draws this code point as colour emoji rather than from the text font - it
+     *  is missing from that font on purpose, so glyph coverage must not touch it. */
+    rendersAsEmoji?(codePoint: number, fontFamily: string, fontStyle: FontStyle): boolean;
+    /** Told when a code point had to be removed because no font could draw it. Optional: a metrics
+     *  object used only for measuring (a test double) has nothing to report to. */
+    reportMissingGlyph?(codePoints: number[]): void;
     /** How many glyphs the run will DRAW, when a ligature made that differ from its code-point count.
      *  Absent or `undefined` means "the same", which is every Latin run. */
     shapedGlyphCount?(text: string, fontFamily?: string, fontStyle?: FontStyle): number | undefined;

package/dist/utils/pdf-object-manager.d.ts

@@ -65,6 +65,16 @@ export declare class PDFObjectManager implements FontMetrics {
     addObject(content: string): number;
     replaceObject(objectNumber: number, content: string): void;
     setCompress(on: boolean): void;
+    /**
+     * Code points no font in the document could draw, so they were removed rather than emitted as
+     * `.notdef` (which PDF/A forbids). Collected here because this object is the one thing threaded
+     * through BOTH passes; the caller gets them back with the bytes, so an application can say which
+     * character went missing instead of a server log nobody reads.
+     */
+    private readonly missingGlyphs;
+    reportMissingGlyph(codePoints: number[]): void;
+    /** The dropped code points, in the order first seen. Empty for a document that hit none. */
+    getMissingGlyphs(): number[];
     setOverflowPolicy(policy: OverflowPolicy): void;
     getOverflowPolicy(): OverflowPolicy;
     private encToken;
@@ -142,6 +152,13 @@ export declare class PDFObjectManager implements FontMetrics {
      * Used to pick a family out of a fallback stack, per code point.
      */
     hasGlyph(codePoint: number, fontFamily: string, fontStyle?: FontStyle): boolean;
+    /**
+     * Whether the document draws this code point through the colour-emoji path instead of the text
+     * font - a fallback emoji FONT, or an image source (`Document({ emoji })`). Such a code point is
+     * missing from the text font ON PURPOSE, so glyph coverage must leave it alone. Mirrors exactly
+     * what `getCharWidth` measures it as, or a dropped emoji would have been measured and not drawn.
+     */
+    rendersAsEmoji(codePoint: number, fontFamily: string, fontStyle: FontStyle): boolean;
     private colorRenders;
     getCustomFontResource(name: string, style?: FontStyle): FontIndexes | undefined;
     /** How many glyphs a run draws - its code-point count unless a ligature merged some. `letterSpacing`

package/dist/utils/pdf-object-manager.js

@@ -189,6 +189,15 @@ export class PDFObjectManager {
         // Interactive form fields (AcroForm). Empty unless a form field was placed; then finalize() emits the
         // catalog /AcroForm dict and each field's widget lands in its page /Annots.
         this._acroform = new AcroFormCollector();
+        // Overflow policy for unbreakable content taller than a page region (default "error"); the renderer
+        // seeds it into the layout-context root so packChildren can act on it.
+        /**
+         * Code points no font in the document could draw, so they were removed rather than emitted as
+         * `.notdef` (which PDF/A forbids). Collected here because this object is the one thing threaded
+         * through BOTH passes; the caller gets them back with the bytes, so an application can say which
+         * character went missing instead of a server log nobody reads.
+         */
+        this.missingGlyphs = new Set();
         // Whether to kern. Document-level: measuring (via `runAdvance`, which reads this off the metrics)
         // and drawing (the backend emits a `TJ`) both consult it, so they can never disagree. OFF while the
         // embedded-font path (kern/GPOS) is unbuilt - a standard-14-only kerning would set the 14 system
@@ -216,8 +225,14 @@ export class PDFObjectManager {
     setCompress(on) {
         this.compress = on;
     }
-    // Overflow policy for unbreakable content taller than a page region (default "error"); the renderer
-    // seeds it into the layout-context root so packChildren can act on it.
+    reportMissingGlyph(codePoints) {
+        for (const cp of codePoints)
+            this.missingGlyphs.add(cp);
+    }
+    /** The dropped code points, in the order first seen. Empty for a document that hit none. */
+    getMissingGlyphs() {
+        return [...this.missingGlyphs];
+    }
     setOverflowPolicy(policy) {
         this.overflowPolicy = policy;
     }
@@ -694,6 +709,23 @@ endstream`;
             return ttf.getGlyphIndex(codePoint) !== 0;
         return isWindows1252(codePoint);
     }
+    /**
+     * Whether the document draws this code point through the colour-emoji path instead of the text
+     * font - a fallback emoji FONT, or an image source (`Document({ emoji })`). Such a code point is
+     * missing from the text font ON PURPOSE, so glyph coverage must leave it alone. Mirrors exactly
+     * what `getCharWidth` measures it as, or a dropped emoji would have been measured and not drawn.
+     */
+    rendersAsEmoji(codePoint, fontFamily, fontStyle) {
+        const ttf = this.getCustomFont(fontFamily, fontStyle);
+        if (ttf && this.colorRenders(ttf, codePoint))
+            return true;
+        if (this.emojiFontName && this.emojiFontName !== fontFamily) {
+            const emoji = this.getColorFont(this.emojiFontName, fontStyle);
+            if (emoji && this.colorRenders(emoji, codePoint))
+                return true;
+        }
+        return this.emojiImage !== undefined && isEmojiCodePoint(codePoint);
+    }
     colorRenders(ttf, codePoint) {
         return ttf.getColorGlyph(ttf.getGlyphIndex(codePoint)) !== null;
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jasy/pdf",
-  "version": "1.0.0-alpha.10",
+  "version": "1.0.0-alpha.12",
   "description": "Declarative, component-based PDF generation in pure TypeScript - Flutter-style components, real AFM font metrics, a hand-rolled PDF writer. No headless browser, no Java.",
   "keywords": [
     "declarative",
@@ -48,6 +48,7 @@
     "@jimp/types": "^1.6.0",
     "@types/node": "^25.9.3",
     "@vitest/coverage-v8": "^2.1.1",
+    "hyphen": "^1.14.1",
     "ncp": "^2.0.0",
     "oxfmt": "^0.55.0",
     "oxlint": "^1.70.0",