postext-pdf 1.29.2 → 1.30.1

package/dist/canvasWords.js

@@ -0,0 +1,160 @@
+/**
+ * The words a browser's canvas shapes a string in (postext #770).
+ *
+ * Chrome does not hand a string to its shaper whole. It cuts it into words
+ * and shapes each alone, so that a word met again is answered from a cache
+ * (Blink's `CachingWordShapeIterator`): nothing a font does between two
+ * glyphs crosses from one word to the next, neither a kerned pair nor a
+ * ligature, and each word is given the script of its own letters. The
+ * layout was measured that way, and the canvas paints that way. A shaper
+ * handed the whole string does otherwise:
+ *
+ * - Source Serif 4 kerns a space against the `W` after it: `Wang Wei` in a
+ *   running head came out 0.06 em narrower than its measure;
+ * - Noto Sans JP kerns kana against kana (`える` 0.03 em, `われ` 0.05 em):
+ *   a line of 36 characters ended 1.4 pt short of its measure, where the
+ *   canvas kerns only a character and the mark after it (`が、`, `す。`);
+ * - Noto Sans SC draws the dash of `9704—2012` in its Chinese form, a full
+ *   em, when the word is shaped with the Latin words beside it, and 0.89
+ *   em wide alone, as measured;
+ * - Noto Serif JP draws `“` in its Japanese form beside a Han character
+ *   in one run, and in its default form as the word of its own it is on
+ *   the canvas;
+ * - Aref Ruqaa sets a space 0.146 em wide in Arabic text and 0.333 em in
+ *   any other: the space between two Arabic words is a word of its own on
+ *   the canvas, with no letter and so no script, and a title of three
+ *   words came out 0.37 em short.
+ *
+ * The cuts, as measured in Chrome 154 (widths of strings against the sum
+ * of their parts, in faces that kern across the cut or draw a mark
+ * otherwise by script):
+ *
+ * - a space (U+0020) and a tab are words of their own. No other space
+ *   cuts: a no-break space stays in its word;
+ * - a string with no character above U+00FF is cut there only;
+ * - a CJK character (a Han character, a kana, a CJK mark or symbol, a
+ *   full-width form, an enclosed number, an emoji) ends the word before
+ *   it and opens one that holds it, the marks and modifier letters after
+ *   it (`ー`, `々`, a voicing mark) and the CJK characters of no script
+ *   after it (`、`, `。`, `」`): `が、` is one word, `える` two. A word that
+ *   opens with such a mark (`「`, `ー`) takes the first character with a
+ *   script after it as well.
+ *
+ * The symbols counted as CJK were read off the canvas ({@link CJK_BASE}):
+ * the characters of no script after which `n` + character + `—` keeps the
+ * default dash in Noto Serif SC under `zh`, whose dash takes the Chinese
+ * form anywhere in a Latin word. The letters (Han, kana, Bopomofo, the
+ * full-width and half-width forms, Roman numerals) cannot be read that
+ * way and are Blink's own ranges; Hangul is not among them.
+ */
+/** The characters with no script, by block, that end a word (measured). */
+const SYMBOLS = '2020-2021 2030 203B-203C 2042 2047-2049 2051 2100 2103 2105 2109-210A 2113 2116 2121 213B 2150-2152 2156-215A 2189 2194-2195 2307 2312 231A-231B 2329 23BE-23CC 23CE 23E9-23EC 23F0 23F3 2423 2460-2492 249C-24FF 25A0-25A2 25AA-25AB 25B1-25B3 25B6-25B7 25BC-25BD 25C0-25C1 25C6-25C7 25C9 25CB-25CC 25CE-25D3 25E2-25E6 25EF 25FD-25FE 2600-2603 2605-2606 260E 2614-2617 261D 2620 2640 2642 2648-2653 2660-266F 2672-267D 267F 2693 2695-2696 26A0-26A1 26A7 26AA-26AB 26BD-26BE 26C4-26C5 26CE 26D3-26D4 26EA 26F2-26F3 26F5 26F9-26FA 26FD 2705 2708 270A-270D 2713 271A 2728 273F-2740 2744 274C 274E 2753-2757 2763-2764 2776-277F 2795-2797 27A1 27B0 27BF 2B1A-2B1C 2B50 2B55 2FF0-3004 3006 3008-3020 3030 3036-3037 303C-303F 3190-319F 31C0-31E5 31EF 3220-325F 327F-32CF 32FF 3358-33FF FE10-FE12 FE19 FE30-FE52 FE54-FE66 FE68-FE6B FF01-FF0C FF0E-FF1A FF1D FF1F-FF20 FF3B-FF3D FF3F FF5B-FF65 FFE0-FFE2 FFE4-FFE6 FFE8-FFEE';
+/** The letters and numbers: Å, Roman numerals, Han, kana, Bopomofo,
+ *  circled and squared kana, full-width Latin, half-width kana and Hangul,
+ *  the emoji and the Han of the upper planes. */
+const LETTERS = '212B 2160-216B 2170-217B 217F 2E80-2FDF 3007 3021-3029 3038-303A 3041-3096 309F 30A1-30FA 30FF 3105-312F 31A0-31BF 32D0-32FE 3300-3357 3400-4DBF 4E00-9FFF F900-FAFF FF21-FF3A FF41-FF5A FF66-FF6F FF71-FF9D FFA0-FFDC 1F000-1FAFF 20000-3FFFF';
+/** The marks and modifier letters of the CJK blocks: they open a CJK word
+ *  when a string starts with one, and never end a word. */
+const MODIFIERS = '3005 302A-302D 3031-3035 303B 3099-309E 30FC-30FE FF3E FF40 FF70 FF9E-FF9F FFE3';
+function ranges(list) {
+    const out = [];
+    for (const part of list.split(' ')) {
+        const [lo, hi = lo] = part.split('-');
+        out.push(parseInt(lo, 16), parseInt(hi, 16));
+    }
+    // By first code point, for the search.
+    const pairs = [];
+    for (let i = 0; i < out.length; i += 2)
+        pairs.push([out[i], out[i + 1]]);
+    pairs.sort((a, b) => a[0] - b[0]);
+    return pairs.flat();
+}
+const CJK_BASE = ranges(`${SYMBOLS} ${LETTERS}`);
+const CJK_MODIFIERS = ranges(MODIFIERS);
+function inRanges(table, cp) {
+    let lo = 0;
+    let hi = table.length / 2 - 1;
+    while (lo <= hi) {
+        const mid = (lo + hi) >> 1;
+        if (cp < table[mid * 2])
+            hi = mid - 1;
+        else if (cp > table[mid * 2 + 1])
+            lo = mid + 1;
+        else
+            return true;
+    }
+    return false;
+}
+/** No CJK character is below this. */
+const FIRST_CJK = 0x2020;
+/** A CJK character that ends the word before it. */
+function isCjkBase(cp) {
+    return cp >= FIRST_CJK && inRanges(CJK_BASE, cp);
+}
+/** A CJK character: one that opens a CJK word when a word starts with it. */
+function isCjk(cp) {
+    return cp >= FIRST_CJK && (inRanges(CJK_BASE, cp) || inRanges(CJK_MODIFIERS, cp));
+}
+function isDelimiter(cp) {
+    return cp === 0x20 || cp === 0x09;
+}
+/** A mark, a modifier letter or a modifier symbol, and the zero width
+ *  joiner: they stay in the word of the character before them. */
+const JOINS_RE = /^[\p{M}\p{Lm}\p{Sk}\u200D]$/u;
+/** A character with no script of its own. */
+const NO_SCRIPT_RE = /^[\p{Script=Common}\p{Script=Inherited}]$/u;
+/** A string a browser keeps in eight bits. */
+const LATIN_1_RE = /^[\u0000-\u00FF]*$/;
+/** Anything a string may be cut at: a space, a tab, a character from the
+ *  first CJK one on. */
+const MAY_CUT_RE = /[ \t\u2020-\uFFFF]/;
+/**
+ * `text` cut into the words the canvas shapes it in (see the module
+ * comment); the string itself, alone, when it is one word.
+ */
+export function canvasWords(text) {
+    if (text.length < 2 || !MAY_CUT_RE.test(text))
+        return [text];
+    const latin1 = LATIN_1_RE.test(text);
+    const words = [];
+    const n = text.length;
+    const width = (cp) => (cp > 0xFFFF ? 2 : 1);
+    for (let start = 0; start < n;) {
+        const first = text.codePointAt(start);
+        let end = start + width(first);
+        if (isDelimiter(first)) {
+            // A word of its own.
+        }
+        else if (latin1 || !isCjk(first)) {
+            while (end < n) {
+                const cp = text.codePointAt(end);
+                if (isDelimiter(cp) || (!latin1 && isCjkBase(cp)))
+                    break;
+                end += width(cp);
+            }
+        }
+        else {
+            let scripted = !NO_SCRIPT_RE.test(String.fromCodePoint(first));
+            while (end < n) {
+                const cp = text.codePointAt(end);
+                const ch = String.fromCodePoint(cp);
+                if (JOINS_RE.test(ch)) {
+                    end += width(cp);
+                    continue;
+                }
+                if (!isCjk(cp))
+                    break;
+                if (!NO_SCRIPT_RE.test(ch)) {
+                    if (scripted)
+                        break;
+                    scripted = true;
+                }
+                end += width(cp);
+            }
+        }
+        words.push(text.slice(start, end));
+        start = end;
+    }
+    return words.length === 1 ? [text] : words;
+}
+//# sourceMappingURL=canvasWords.js.map

package/dist/canvasWords.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"canvasWords.js","sourceRoot":"","sources":["../src/canvasWords.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAgDG;AAEH,2EAA2E;AAC3E,MAAM,OAAO,GAAG,+4BAA+4B,CAAC;AACh6B;;iDAEiD;AACjD,MAAM,OAAO,GAAG,gPAAgP,CAAC;AACjQ;2DAC2D;AAC3D,MAAM,SAAS,GAAG,iFAAiF,CAAC;AAEpG,SAAS,MAAM,CAAC,IAAY;IAC1B,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;QACnC,MAAM,CAAC,EAAE,EAAE,EAAE,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;QACtC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAG,EAAE,EAAE,CAAC,EAAE,QAAQ,CAAC,EAAG,EAAE,EAAE,CAAC,CAAC,CAAC;IACjD,CAAC;IACD,uCAAuC;IACvC,MAAM,KAAK,GAAuB,EAAE,CAAC;IACrC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,CAAC,MAAM,EAAE,CAAC,IAAI,CAAC;QAAE,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAE,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,CAAC,CAAC;IAC3E,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAClC,OAAO,KAAK,CAAC,IAAI,EAAE,CAAC;AACtB,CAAC;AAED,MAAM,QAAQ,GAAG,MAAM,CAAC,GAAG,OAAO,IAAI,OAAO,EAAE,CAAC,CAAC;AACjD,MAAM,aAAa,GAAG,MAAM,CAAC,SAAS,CAAC,CAAC;AAExC,SAAS,QAAQ,CAAC,KAAwB,EAAE,EAAU;IACpD,IAAI,EAAE,GAAG,CAAC,CAAC;IACX,IAAI,EAAE,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,GAAG,CAAC,CAAC;IAC9B,OAAO,EAAE,IAAI,EAAE,EAAE,CAAC;QAChB,MAAM,GAAG,GAAG,CAAC,EAAE,GAAG,EAAE,CAAC,IAAI,CAAC,CAAC;QAC3B,IAAI,EAAE,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,CAAE;YAAE,EAAE,GAAG,GAAG,GAAG,CAAC,CAAC;aAClC,IAAI,EAAE,GAAG,KAAK,CAAC,GAAG,GAAG,CAAC,GAAG,CAAC,CAAE;YAAE,EAAE,GAAG,GAAG,GAAG,CAAC,CAAC;;YAC3C,OAAO,IAAI,CAAC;IACnB,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,sCAAsC;AACtC,MAAM,SAAS,GAAG,MAAM,CAAC;AAEzB,oDAAoD;AACpD,SAAS,SAAS,CAAC,EAAU;IAC3B,OAAO,EAAE,IAAI,SAAS,IAAI,QAAQ,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC;AACnD,CAAC;AAED,6EAA6E;AAC7E,SAAS,KAAK,CAAC,EAAU;IACvB,OAAO,EAAE,IAAI,SAAS,IAAI,CAAC,QAAQ,CAAC,QAAQ,EAAE,EAAE,CAAC,IAAI,QAAQ,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,CAAC;AACpF,CAAC;AAED,SAAS,WAAW,CAAC,EAAU;IAC7B,OAAO,EAAE,KAAK,IAAI,IAAI,EAAE,KAAK,IAAI,CAAC;AACpC,CAAC;AAED;kEACkE;AAClE,MAAM,QAAQ,GAAG,8BAA8B,CAAC;AAChD,6CAA6C;AAC7C,MAAM,YAAY,GAAG,4CAA4C,CAAC;AAClE,8CAA8C;AAC9C,MAAM,UAAU,GAAG,oBAAoB,CAAC;AACxC;wBACwB;AACxB,MAAM,UAAU,GAAG,oBAAoB,CAAC;AAExC;;;GAGG;AACH,MAAM,UAAU,WAAW,CAAC,IAAY;IACtC,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,CAAC,IAAI,CAAC,CAAC;IAC7D,MAAM,MAAM,GAAG,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACrC,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;IACtB,MAAM,KAAK,GAAG,CAAC,EAAU,EAAU,EAAE,CAAC,CAAC,EAAE,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5D,KAAK,IAAI,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,GAAG,CAAC;QAC/B,MAAM,KAAK,GAAG,IAAI,CAAC,WAAW,CAAC,KAAK,CAAE,CAAC;QACvC,IAAI,GAAG,GAAG,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,CAAC;QAC/B,IAAI,WAAW,CAAC,KAAK,CAAC,EAAE,CAAC;YACvB,qBAAqB;QACvB,CAAC;aAAM,IAAI,MAAM,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,CAAC;YACnC,OAAO,GAAG,GAAG,CAAC,EAAE,CAAC;gBACf,MAAM,EAAE,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAE,CAAC;gBAClC,IAAI,WAAW,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,MAAM,IAAI,SAAS,CAAC,EAAE,CAAC,CAAC;oBAAE,MAAM;gBACzD,GAAG,IAAI,KAAK,CAAC,EAAE,CAAC,CAAC;YACnB,CAAC;QACH,CAAC;aAAM,CAAC;YACN,IAAI,QAAQ,GAAG,CAAC,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC,CAAC;YAC/D,OAAO,GAAG,GAAG,CAAC,EAAE,CAAC;gBACf,MAAM,EAAE,GAAG,IAAI,CAAC,WAAW,CAAC,GAAG,CAAE,CAAC;gBAClC,MAAM,EAAE,GAAG,MAAM,CAAC,aAAa,CAAC,EAAE,CAAC,CAAC;gBACpC,IAAI,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;oBACtB,GAAG,IAAI,KAAK,CAAC,EAAE,CAAC,CAAC;oBACjB,SAAS;gBACX,CAAC;gBACD,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;oBAAE,MAAM;gBACtB,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,CAAC;oBAC3B,IAAI,QAAQ;wBAAE,MAAM;oBACpB,QAAQ,GAAG,IAAI,CAAC;gBAClB,CAAC;gBACD,GAAG,IAAI,KAAK,CAAC,EAAE,CAAC,CAAC;YACnB,CAAC;QACH,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC,CAAC;QACnC,KAAK,GAAG,GAAG,CAAC;IACd,CAAC;IACD,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC;AAC7C,CAAC"}

package/dist/faceFiles.d.ts

@@ -12,30 +12,92 @@
  * text object, so a line of Han with a Latin word paints from three or four
  * slices and each slice keeps only the glyphs set in it.
  *
+ * Which file sets a character two files hold (postext #770): a browser
+ * takes it from the `@font-face` rule declared last among those whose
+ * `unicode-range` holds it, and Fontsource's files overlap. A Japanese
+ * family repeats the ASCII letters in its `latin` file and in its numbered
+ * slices, and a pen declares the slices after `latin`: Chrome 154 sets the
+ * letters of `Shift_JIS` from slice 118 and the `_` from slice 113 of Noto
+ * Serif JP, 4252 thousandths of the em wide, where the `latin` file alone
+ * kerns `t_` and `_J` and sets it 4318 wide. A provider that hands each
+ * file with its range and its place in the declaration (`PdfFontFile`)
+ * gets each character from the file the browser took it from
+ * ({@link FaceFiles.declaredFile}), in a document shaped in its own
+ * language; one that hands bare files gets it from the first file with a
+ * glyph, as before, and so does a document that keeps the host's language.
+ *
  * Characters a face lacks in every file are recorded as they are shaped
  * ({@link noteMissingGlyphs}), for the `missingGlyph` warning.
  */
 import type { PDFFont } from 'pdf-lib';
+/** What a provider says of a file of a face (`PdfFontFile`): the
+ *  characters its `@font-face` rule answers for, and where the rule is
+ *  declared among the rules of the face. */
+export interface FileDeclaration {
+    unicodeRange?: string;
+    order?: number;
+}
 /** The files of one face, in the order their glyphs are looked up. */
 export declare class FaceFiles {
     readonly files: PDFFont[];
     private readonly faces;
     /** First file holding each code point, or null for none. */
     private readonly first;
-    constructor(files: readonly PDFFont[]);
-    /** Append a file (a slice fetched later); a file already listed is kept
-     *  where it is. */
-    add(file: PDFFont): void;
+    /** The range each file was given with; a file with none answers for
+     *  every character. */
+    private readonly ranges;
+    /** Where each file is declared: the number it was given with, else its
+     *  place in the list counted down, the first file being the one
+     *  declared last. */
+    private readonly orders;
+    /** The files from the one declared last to the one declared first. */
+    private lastDeclaredFirst;
+    /** The file a browser sets each code point in, or null for none. */
+    private readonly declared;
+    private hasRanges;
+    constructor(files: readonly PDFFont[], declarations?: ReadonlyMap<PDFFont, FileDeclaration>);
+    /** Append a file (a slice fetched later), with what the provider says
+     *  of it; a file already listed is kept where it is, as it was given. */
+    add(file: PDFFont, declaration?: FileDeclaration): void;
+    /** Whether the provider gave a file of the face with its range: the
+     *  face is then read as a browser reads its `@font-face` rules
+     *  ({@link declaredFile}). */
+    get ranged(): boolean;
     /** Whether `file` has a glyph for `cp`. */
     covers(file: PDFFont, cp: number): boolean;
+    /** Whether `file` answers for `cp` as a browser reads it: its range
+     *  holds the character and the file has a glyph for it. Chrome 154 skips
+     *  a file whose range claims a character it has no glyph for and takes
+     *  the next one down, and never draws a glyph a file has outside its
+     *  range. */
+    private answers;
     /** The first file with a glyph for `cp`, or null when none has one. */
     fileFor(cp: number): PDFFont | null;
+    /** The files a browser tries in turn: the one declared last first.
+     *  Files of one number keep the order they were given in. */
+    private byDeclaration;
+    /**
+     * The file a browser sets `cp` in: the one declared last among those
+     * whose range holds it and that have a glyph for it. Null when none
+     * does: the browser then leaves the family, for the next one of the
+     * `font-family` list or a font of the system.
+     */
+    declaredFile(cp: number): PDFFont | null;
+    /**
+     * The file a browser sets a letter and the marks on it in (`cps`, the
+     * letter first): the one declared last that answers for every one of
+     * them. With Source Serif 4 declared first and Noto Serif JP after it
+     * for `a` alone, Chrome sets `a` in the second (557 thousandths) and
+     * `a` + U+0308 whole in the first (541). Null when no file sets them
+     * all.
+     */
+    declaredFileOfAll(cps: readonly number[]): PDFFont | null;
 }
 /** Register the files of a face under its first file, the font the
  *  renderers are handed. A first file already registered (two faces the
  *  provider answered with the same files) keeps its list, which takes the
  *  new files too; the list in force is returned. */
-export declare function registerFaceFiles(primary: PDFFont, files: readonly PDFFont[]): FaceFiles;
+export declare function registerFaceFiles(primary: PDFFont, files: readonly PDFFont[], declarations?: ReadonlyMap<PDFFont, FileDeclaration>): FaceFiles;
 /** The files of the face whose first file is `font`, when it has several. */
 export declare function faceFilesOf(font: PDFFont): FaceFiles | undefined;
 /** A run of text and the file it is set in. */
@@ -46,6 +108,9 @@ export interface FileRun {
      *  and the text it was cut out of does (`besideLetterOf`, #764): the
      *  caller shapes the run inside `withBesideLetter`. */
     beside?: number;
+    /** The run is a word, or part of one, of a string with a character
+     *  above U+00FF ({@link wordRuns}): with no letter it has no script. */
+    wide?: true;
 }
 /**
  * `text` cut into runs by the file of `font`'s face that has a glyph for
@@ -85,8 +150,31 @@ export interface FileRun {
  * `خالة.»`, which the `latin` file draws), says the letter whose script
  * the browser shaped it in (`beside`, #764), in a document shaped in its
  * own language.
+ *
+ * A face given with the ranges of its files is cut as a browser cuts it
+ * ({@link declaredRuns}, #770) in a document shaped in its own language,
+ * and none of the rules above that guess at the browser's file apply.
  */
 export declare function fileRuns(font: PDFFont, text: string, rtl?: boolean): FileRun[];
+/**
+ * `text`, a string the canvas was handed whole (a line of a running head,
+ * a word of the body), cut into the runs it is shaped in: its words
+ * (`canvasWords.ts`, #770), each cut by file ({@link fileRuns}). The
+ * canvas shapes every word alone, so no kerned pair and no ligature
+ * crosses a space or stands between two CJK characters, and a word takes
+ * the script of its own letters: one with no letter, a space among them,
+ * has none in a string not kept in eight bits (`wide`).
+ *
+ * {@link fileRuns} itself, the string shaped whole as up to postext-pdf
+ * 1.29, for a string of one word, in a document that keeps the host's
+ * language, and with a script in force (#725: a word with no letter the
+ * layout measured in the script of its context).
+ */
+export declare function wordRuns(font: PDFFont, text: string, rtl?: boolean): FileRun[];
+/** Run `fn`, the shaping of `run` ({@link wordRuns}), with what the run
+ *  says of its script in force: the letter it was cut from beside, and
+ *  whether it is a word of a string not kept in eight bits. */
+export declare function inRun<T>(run: FileRun, fn: () => T): T;
 /** `text`'s width at `size` (points), each run measured in its file. */
 export declare function widthOfTextAtSize(font: PDFFont, text: string, size: number): number;
 /** Record the characters of `text` that `font` has no glyph for (it draws

package/dist/faceFiles.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"faceFiles.d.ts","sourceRoot":"","sources":["../src/faceFiles.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AACH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAsBvC,sEAAsE;AACtE,qBAAa,SAAS;IACpB,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,CAAC;IAC1B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAgD;IACtE,4DAA4D;IAC5D,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAqC;gBAE/C,KAAK,EAAE,SAAS,OAAO,EAAE;IAKrC;uBACmB;IACnB,GAAG,CAAC,IAAI,EAAE,OAAO,GAAG,IAAI;IAQxB,2CAA2C;IAC3C,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,GAAG,OAAO;IAI1C,uEAAuE;IACvE,OAAO,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI;CAapC;AAID;;;oDAGoD;AACpD,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,SAAS,OAAO,EAAE,GAAG,SAAS,CASxF;AAED,6EAA6E;AAC7E,wBAAgB,WAAW,CAAC,IAAI,EAAE,OAAO,GAAG,SAAS,GAAG,SAAS,CAGhE;AAED,+CAA+C;AAC/C,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb;;2DAEuD;IACvD,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAsCG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,UAAQ,GAAG,OAAO,EAAE,CAmE5E;AAgCD,wEAAwE;AACxE,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CAMnF;AAQD;;;6EAG6E;AAC7E,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI,CAanE;AAED;yEACyE;AACzE,wBAAgB,eAAe,CAAC,IAAI,EAAE,OAAO,GAAG,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,SAAS,CAEtF;AAED;4BAC4B;AAC5B,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAE9C"}
+{"version":3,"file":"faceFiles.d.ts","sourceRoot":"","sources":["../src/faceFiles.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA8BG;AACH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAwBvC;;4CAE4C;AAC5C,MAAM,WAAW,eAAe;IAC9B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,sEAAsE;AACtE,qBAAa,SAAS;IACpB,QAAQ,CAAC,KAAK,EAAE,OAAO,EAAE,CAAC;IAC1B,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAgD;IACtE,4DAA4D;IAC5D,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAqC;IAC3D;2BACuB;IACvB,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAuC;IAC9D;;yBAEqB;IACrB,OAAO,CAAC,QAAQ,CAAC,MAAM,CAA8B;IACrD,sEAAsE;IACtE,OAAO,CAAC,iBAAiB,CAAwB;IACjD,oEAAoE;IACpE,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAqC;IAC9D,OAAO,CAAC,SAAS,CAAS;gBAEd,KAAK,EAAE,SAAS,OAAO,EAAE,EAAE,YAAY,CAAC,EAAE,WAAW,CAAC,OAAO,EAAE,eAAe,CAAC;IAK3F;6EACyE;IACzE,GAAG,CAAC,IAAI,EAAE,OAAO,EAAE,WAAW,CAAC,EAAE,eAAe,GAAG,IAAI;IAiBvD;;kCAE8B;IAC9B,IAAI,MAAM,IAAI,OAAO,CAEpB;IAED,2CAA2C;IAC3C,MAAM,CAAC,IAAI,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,GAAG,OAAO;IAI1C;;;;iBAIa;IACb,OAAO,CAAC,OAAO;IAKf,uEAAuE;IACvE,OAAO,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI;IAcnC;iEAC6D;IAC7D,OAAO,CAAC,aAAa;IAQrB;;;;;OAKG;IACH,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI;IAQxC;;;;;;;OAOG;IACH,iBAAiB,CAAC,GAAG,EAAE,SAAS,MAAM,EAAE,GAAG,OAAO,GAAG,IAAI;CAG1D;AAID;;;oDAGoD;AACpD,wBAAgB,iBAAiB,CAAC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,SAAS,OAAO,EAAE,EAAE,YAAY,CAAC,EAAE,WAAW,CAAC,OAAO,EAAE,eAAe,CAAC,GAAG,SAAS,CAS9I;AAED,6EAA6E;AAC7E,wBAAgB,WAAW,CAAC,IAAI,EAAE,OAAO,GAAG,SAAS,GAAG,SAAS,CAGhE;AAED,+CAA+C;AAC/C,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,OAAO,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb;;2DAEuD;IACvD,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB;4EACwE;IACxE,IAAI,CAAC,EAAE,IAAI,CAAC;CACb;AAED;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GA0CG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,UAAQ,GAAG,OAAO,EAAE,CA2D5E;AA4ID;;;;;;;;;;;;;GAaG;AACH,wBAAgB,QAAQ,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,UAAQ,GAAG,OAAO,EAAE,CAa5E;AAED;;+DAE+D;AAC/D,wBAAgB,KAAK,CAAC,CAAC,EAAE,GAAG,EAAE,OAAO,EAAE,EAAE,EAAE,MAAM,CAAC,GAAG,CAAC,CAErD;AAgCD,wEAAwE;AACxE,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,GAAG,MAAM,CAMnF;AAQD;;;6EAG6E;AAC7E,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,GAAG,IAAI,CAanE;AAED;yEACyE;AACzE,wBAAgB,eAAe,CAAC,IAAI,EAAE,OAAO,GAAG,WAAW,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,SAAS,CAEtF;AAED;4BAC4B;AAC5B,wBAAgB,UAAU,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAE9C"}

package/dist/faceFiles.js

@@ -1,6 +1,8 @@
 import { mirroredCodePoint } from 'postext';
 import { isDefaultIgnorable, isFallbackHandled, substituteGlyph } from './pdf-backend/fallbackSpaces';
-import { besideLetterOf, shapingScriptLetter } from './shapingLanguage';
+import { besideLetterOf, shapesAsCanvas, shapingScript, shapingScriptLetter, withBesideLetter, withinWideText } from './shapingLanguage';
+import { canvasWords } from './canvasWords';
+import { parseUnicodeRange, rangesHold } from './unicodeRange';
 function coverageFace(font) {
     const face = font.embedder?.font;
     return face && typeof face.hasGlyphForCodePoint === 'function' ? face : undefined;
@@ -17,27 +19,63 @@ export class FaceFiles {
     faces = new Map();
     /** First file holding each code point, or null for none. */
     first = new Map();
-    constructor(files) {
+    /** The range each file was given with; a file with none answers for
+     *  every character. */
+    ranges = new Map();
+    /** Where each file is declared: the number it was given with, else its
+     *  place in the list counted down, the first file being the one
+     *  declared last. */
+    orders = new Map();
+    /** The files from the one declared last to the one declared first. */
+    lastDeclaredFirst;
+    /** The file a browser sets each code point in, or null for none. */
+    declared = new Map();
+    hasRanges = false;
+    constructor(files, declarations) {
         this.files = [];
         for (const file of files)
-            this.add(file);
+            this.add(file, declarations?.get(file));
     }
-    /** Append a file (a slice fetched later); a file already listed is kept
-     *  where it is. */
-    add(file) {
+    /** Append a file (a slice fetched later), with what the provider says
+     *  of it; a file already listed is kept where it is, as it was given. */
+    add(file, declaration) {
         if (this.faces.has(file))
             return;
+        this.orders.set(file, declaration?.order ?? -this.files.length);
         this.files.push(file);
         this.faces.set(file, coverageFace(file));
-        // A code point no file had may be in this one.
+        const ranges = declaration?.unicodeRange === undefined ? undefined : parseUnicodeRange(declaration.unicodeRange);
+        if (ranges) {
+            this.ranges.set(file, ranges);
+            this.hasRanges = true;
+        }
+        // A code point no file had may be in this one, and a file declared
+        // later than the one that set a character takes it.
         for (const [cp, hit] of this.first)
             if (hit === null)
                 this.first.delete(cp);
+        this.declared.clear();
+        this.lastDeclaredFirst = undefined;
+    }
+    /** Whether the provider gave a file of the face with its range: the
+     *  face is then read as a browser reads its `@font-face` rules
+     *  ({@link declaredFile}). */
+    get ranged() {
+        return this.hasRanges;
     }
     /** Whether `file` has a glyph for `cp`. */
     covers(file, cp) {
         return this.faces.get(file)?.hasGlyphForCodePoint(cp) ?? false;
     }
+    /** Whether `file` answers for `cp` as a browser reads it: its range
+     *  holds the character and the file has a glyph for it. Chrome 154 skips
+     *  a file whose range claims a character it has no glyph for and takes
+     *  the next one down, and never draws a glyph a file has outside its
+     *  range. */
+    answers(file, cp) {
+        const ranges = this.ranges.get(file);
+        return (ranges === undefined || rangesHold(ranges, cp)) && this.covers(file, cp);
+    }
     /** The first file with a glyph for `cp`, or null when none has one. */
     fileFor(cp) {
         const hit = this.first.get(cp);
@@ -53,20 +91,54 @@ export class FaceFiles {
         this.first.set(cp, found);
         return found;
     }
+    /** The files a browser tries in turn: the one declared last first.
+     *  Files of one number keep the order they were given in. */
+    byDeclaration() {
+        if (!this.lastDeclaredFirst) {
+            const at = new Map(this.files.map((file, i) => [file, i]));
+            this.lastDeclaredFirst = [...this.files].sort((a, b) => this.orders.get(b) - this.orders.get(a) || at.get(a) - at.get(b));
+        }
+        return this.lastDeclaredFirst;
+    }
+    /**
+     * The file a browser sets `cp` in: the one declared last among those
+     * whose range holds it and that have a glyph for it. Null when none
+     * does: the browser then leaves the family, for the next one of the
+     * `font-family` list or a font of the system.
+     */
+    declaredFile(cp) {
+        const hit = this.declared.get(cp);
+        if (hit !== undefined)
+            return hit;
+        const found = this.byDeclaration().find((file) => this.answers(file, cp)) ?? null;
+        this.declared.set(cp, found);
+        return found;
+    }
+    /**
+     * The file a browser sets a letter and the marks on it in (`cps`, the
+     * letter first): the one declared last that answers for every one of
+     * them. With Source Serif 4 declared first and Noto Serif JP after it
+     * for `a` alone, Chrome sets `a` in the second (557 thousandths) and
+     * `a` + U+0308 whole in the first (541). Null when no file sets them
+     * all.
+     */
+    declaredFileOfAll(cps) {
+        return this.byDeclaration().find((file) => cps.every((cp) => this.answers(file, cp))) ?? null;
+    }
 }
 const faceFiles = new WeakMap();
 /** Register the files of a face under its first file, the font the
  *  renderers are handed. A first file already registered (two faces the
  *  provider answered with the same files) keeps its list, which takes the
  *  new files too; the list in force is returned. */
-export function registerFaceFiles(primary, files) {
+export function registerFaceFiles(primary, files, declarations) {
     const known = faceFiles.get(primary);
     if (known) {
         for (const file of files)
-            known.add(file);
+            known.add(file, declarations?.get(file));
         return known;
     }
-    const list = new FaceFiles([primary, ...files]);
+    const list = new FaceFiles([primary, ...files], declarations);
     faceFiles.set(primary, list);
     return list;
 }
@@ -113,11 +185,17 @@ export function faceFilesOf(font) {
  * `خالة.»`, which the `latin` file draws), says the letter whose script
  * the browser shaped it in (`beside`, #764), in a document shaped in its
  * own language.
+ *
+ * A face given with the ranges of its files is cut as a browser cuts it
+ * ({@link declaredRuns}, #770) in a document shaped in its own language,
+ * and none of the rules above that guess at the browser's file apply.
  */
 export function fileRuns(font, text, rtl = false) {
     const list = faceFilesOf(font);
     if (!list)
         return [{ font, text }];
+    if (setsAsDeclared(list))
+        return besideLetters(declaredRuns(list, font, text, rtl), text, font);
     const letter = shapingScriptLetter();
     if (letter !== undefined) {
         const file = list.fileFor(letter);
@@ -177,6 +255,11 @@ export function fileRuns(font, text, rtl = false) {
         else
             runs.push({ font, text: pending });
     }
+    return besideLetters(runs, text, font);
+}
+/** `runs`, the runs `text` was cut into, each saying the letter it was
+ *  cut from beside when it has none; one run in `font` for no run. */
+function besideLetters(runs, text, font) {
     if (runs.length > 1) {
         // The runs follow each other through the text.
         let at = 0;
@@ -189,6 +272,169 @@ export function fileRuns(font, text, rtl = false) {
     }
     return runs.length > 0 ? runs : [{ font, text }];
 }
+/**
+ * Whether the text being painted takes each character from the file a
+ * browser takes it from: a face whose provider gave its files with their
+ * ranges, in a document shaped in its own language.
+ *
+ * The page chose the file of a character the same way whatever language
+ * it shaped the text in, but a document that keeps the host's language is
+ * shaped string by string, each run in the script it shows, as it always
+ * was, and prints the same bytes. Cut by the browser's files, its strings
+ * fall into more runs with no letter, which that shaping gives no script:
+ * in Noto Serif SC the space is another slice's than the letters and the
+ * digits, so `GB/T 15834` became `GB/T`, a space and `15834`, and the
+ * digits alone took the proportional forms of the default script, 0.80 pt
+ * short of the tabular ones the page measured in a Latin string (15.75
+ * pt, the Cookbook's Chinese official document). `renderToPdf` drops what
+ * the files say when no document of the render is shaped in its own
+ * language (`withBareFiles`); this is for a page of one that is not, in a
+ * render that has one.
+ */
+function setsAsDeclared(list) {
+    return list.ranged && shapesAsCanvas();
+}
+/** A combining mark: it is set in the file of the letter it stands on. */
+const MARK_RE = /^\p{M}$/u;
+/**
+ * `text` cut into runs by the file a browser sets each character in, for
+ * a face given with the ranges of its files (postext #770). As measured
+ * in Chrome 154 (two files that draw the same letters at other widths,
+ * declared as one family, on a canvas in a page and in a worker):
+ *
+ * - a character is set in the file declared last among those whose range
+ *   holds it: of two `FontFace` objects, the one added to the font set
+ *   last, whichever was made or loaded first; of two `@font-face` rules,
+ *   the later one in the document; and a `FontFace` added by script
+ *   rather than any rule of a style sheet, whichever came first;
+ * - a file with no range answers for every character;
+ * - a file whose range claims a character it has no glyph for is passed
+ *   over for the next one down, and a glyph a file has outside its range
+ *   is never drawn;
+ * - nothing stays in a file because the character before it was set
+ *   there: `Aa` with the capitals in one file is `A` from it and `a` from
+ *   the other, 1269 thousandths, where either file alone gives 1253 or
+ *   1274. A space goes to the file whose range holds it, though every
+ *   file has one;
+ * - a letter and the marks on it are set in one file, the one declared
+ *   last that answers for them all. (Chrome also composes the pair where
+ *   a file answers for the composed character: `o` + U+0304 is the `ō` of
+ *   `latin-ext`. The PDF's shapers compose nothing, so here the pair goes
+ *   to a file that has both.)
+ *
+ * So no character follows the one before it here, an opening bracket does
+ * not go with what it opens and a word shaped in a script is not looked up
+ * by a letter of that script: the browser did none of that. What stays is
+ * what the shapers need: a zero width joiner or non-joiner with the letter
+ * it joins, an invisible character in the run it stands in.
+ *
+ * A character no file answers for was drawn by the browser in another
+ * family; here it is set as in a face given without ranges: in the run
+ * before it when that file has a glyph, else in the first file that has
+ * one (its mirror's, in a right-to-left run), and reported as missing
+ * when none has.
+ */
+function declaredRuns(list, font, text, rtl) {
+    const runs = [];
+    let current;
+    // Joining controls waiting for the letter after them.
+    let pending = '';
+    let previous = 0;
+    // The file of the letter whose marks follow, up to `wholeEnd`.
+    let whole = null;
+    let wholeEnd = 0;
+    const chars = [...text];
+    for (let at = 0; at < chars.length; at++) {
+        const ch = chars[at];
+        const cp = ch.codePointAt(0);
+        if (isJoiningControl(cp) && (!current || !joinsOnto(previous))) {
+            pending += ch;
+            continue;
+        }
+        let file;
+        if (at < wholeEnd) {
+            file = whole;
+        }
+        else if (current && isDefaultIgnorable(cp)) {
+            file = current.font;
+        }
+        else if (MARK_RE.test(ch)) {
+            // A mark whose letter no file sets with it: beside its letter when
+            // that file has a glyph for it.
+            file = current && list.covers(current.font, cp) ? current.font : list.declaredFile(cp);
+        }
+        else {
+            let end = at + 1;
+            while (end < chars.length && MARK_RE.test(chars[end]))
+                end++;
+            if (end > at + 1) {
+                whole = list.declaredFileOfAll(chars.slice(at, end).map((c) => c.codePointAt(0)));
+                wholeEnd = whole ? end : 0;
+            }
+            file = (at < wholeEnd ? whole : null) ?? list.declaredFile(cp);
+        }
+        if (!file) {
+            const mirror = rtl ? mirroredCodePoint(cp) : cp;
+            const shown = mirror !== cp && list.fileFor(mirror) ? mirror : cp;
+            file = current && (list.covers(current.font, shown) || isDefaultIgnorable(cp)) ? current.font : list.fileFor(shown) ?? current?.font ?? font;
+        }
+        if (current && current.font === file) {
+            current.text += pending + ch;
+        }
+        else {
+            current = { font: file, text: pending + ch };
+            runs.push(current);
+        }
+        pending = '';
+        previous = cp;
+    }
+    if (pending) {
+        if (current)
+            current.text += pending;
+        else
+            runs.push({ font, text: pending });
+    }
+    return runs;
+}
+/** A string a browser keeps in eight bits: no character above U+00FF. */
+const LATIN_1_RE = /^[\u0000-\u00FF]*$/;
+/**
+ * `text`, a string the canvas was handed whole (a line of a running head,
+ * a word of the body), cut into the runs it is shaped in: its words
+ * (`canvasWords.ts`, #770), each cut by file ({@link fileRuns}). The
+ * canvas shapes every word alone, so no kerned pair and no ligature
+ * crosses a space or stands between two CJK characters, and a word takes
+ * the script of its own letters: one with no letter, a space among them,
+ * has none in a string not kept in eight bits (`wide`).
+ *
+ * {@link fileRuns} itself, the string shaped whole as up to postext-pdf
+ * 1.29, for a string of one word, in a document that keeps the host's
+ * language, and with a script in force (#725: a word with no letter the
+ * layout measured in the script of its context).
+ */
+export function wordRuns(font, text, rtl = false) {
+    if (!shapesAsCanvas() || shapingScript() !== undefined)
+        return fileRuns(font, text, rtl);
+    const words = canvasWords(text);
+    if (words.length === 1)
+        return fileRuns(font, text, rtl);
+    const wide = !LATIN_1_RE.test(text);
+    const runs = [];
+    for (const word of words) {
+        for (const run of fileRuns(font, word, rtl)) {
+            if (wide)
+                run.wide = true;
+            runs.push(run);
+        }
+    }
+    return runs;
+}
+/** Run `fn`, the shaping of `run` ({@link wordRuns}), with what the run
+ *  says of its script in force: the letter it was cut from beside, and
+ *  whether it is a word of a string not kept in eight bits. */
+export function inRun(run, fn) {
+    return run.wide ? withinWideText(true, () => withBesideLetter(run.beside, fn)) : withBesideLetter(run.beside, fn);
+}
 /** Opening brackets and initial quotation marks (`Ps`, `Pi`). */
 const OPENING_RE = /^[\p{Ps}\p{Pi}]$/u;
 /** The file the first character after a row of opening marks (from `from`)