chaffjs 0.34.0 → 0.36.0

package/src/structure/side-step.ts

@@ -0,0 +1,21 @@
+import type { StructureNode } from "../plugin.ts";
+
+/** The label, then a gap of two or more spaces or a tab (the label column of a step list), then the step's own text. */
+const LABEL_COLUMN = /^(?<word>\S+)(?: {2}|\t)/u;
+
+/**
+ * A heading labelled with a word (side-step-label: "and", "or") where a step list puts a number: a step done alongside
+ * the one before it (GOV.UK's step-by-step pages). The word is matched as written, and must stand in a label column
+ * like a step's number does: "and more details" is an ordinary heading.
+ */
+export const sideStepReader = (labels: readonly string[]): ((node: StructureNode) => boolean) => {
+  const words = new Set(labels);
+  return (node) => {
+    const heading = node.attrs["heading"];
+    if (typeof heading !== "string") return false;
+    const label = LABEL_COLUMN.exec(heading);
+    const word = label?.groups?.["word"];
+    if (label === null || word === undefined) return false;
+    return words.has(word) && heading.slice(label[0].length).trim() !== "";
+  };
+};

package/src/structure/spanned-label-blank.ts

@@ -0,0 +1,39 @@
+// A row label that spans two columns in the table it was converted from (| Total score |  | 5 |): Markdown has no column
+// span, so the cell right after the label is left blank and the value follows. Pure; the caller passes the total words.
+
+/** A note in parentheses at the end of a label (Total score (count of all "yes" answers above)), dropped before reading it. */
+const TRAILING_NOTE = /[(（][^()（）]*[)）]$/u;
+const EMPHASIS = /[*_]/gu;
+const COLON_END = /[:：]$/u;
+const LATIN_LETTER = /\p{Script=Latin}/u;
+
+const labelText = (cell: string): string => cell.replace(EMPHASIS, "").trim().replace(TRAILING_NOTE, "").trim();
+
+/**
+ * Whether `word` stands at the start or the end of `label` as a word of its own: no Latin letter runs on from it. A word of
+ * one character (計) must be the whole label, since it starts or ends many words (計画, 設計).
+ */
+const startsOrEndsWith = (label: string, word: string): boolean => {
+  const lower = label.toLowerCase();
+  const target = word.toLowerCase();
+  if (Array.from(target).length === 1) return lower === target;
+  const after = lower.charAt(target.length);
+  const before = lower.charAt(lower.length - target.length - 1);
+  return (lower.startsWith(target) && !LATIN_LETTER.test(after)) || (lower.endsWith(target) && !LATIN_LETTER.test(before));
+};
+
+/** A label: it ends in a colon (Total:), or a total word starts or ends it (Total score, 得点合計). */
+export const isSpanningLabel = (cell: string, totalWords: readonly string[]): boolean => {
+  const label = labelText(cell);
+  if (label === "") return false;
+  return COLON_END.test(label) || totalWords.some((word) => word !== "" && startsOrEndsWith(label, word));
+};
+
+/**
+ * Whether the blank at `column` is the second half of a label that spans two columns: the cell before it holds such a label
+ * and the cell after it is filled. Cells are undefined when blank.
+ */
+export const isSpannedLabelGap = (cells: readonly (string | undefined)[], column: number, totalWords: readonly string[]): boolean => {
+  const label = cells[column - 1];
+  return cells[column] === undefined && cells[column + 1] !== undefined && label !== undefined && isSpanningLabel(label, totalWords);
+};

package/src/structure/universal.ts

@@ -51,10 +51,19 @@ export const dottedNumber = (line: string, context: NumberingContext, plainText
 export const isAmount = (dotted: DottedLine, countedAfter: StructurePatterns["countedAfter"]): boolean =>
   !dotted.closedByDot && countedAfter?.(dotted.number, dotted.rest) === true;
 
-/** 言語を問わない通し番号。後ろが単位なら数量、本文の文の続きなら文の中の数なので、番号にしない。 */
+/** 年まで書いた日付の値（2022-12-08）。年の無い「2 May」は、番号 2 の題「May」とも読める。 */
+const WITH_YEAR = /^\d{4}-/u;
+
+/** 番号が年まで書いた日付の頭（「8 December 2022」の 8）か。日付を言語の dates が読む。年表の見出しの日は番号の並びではない。 */
+export const startsDate = (text: string, dates: StructurePatterns["dates"]): boolean => {
+  const at = text.length - text.trimStart().length;
+  return (dates?.(text) ?? []).some((date) => date.start <= at && at < date.end && WITH_YEAR.test(String(date.attrs["value"] ?? "")));
+};
+
+/** 言語を問わない通し番号。後ろが単位なら数量、日付なら日、本文の文の続きなら文の中の数なので、番号にしない。 */
 export const universalNumber = (patterns: StructurePatterns, text: string, context: NumberingContext, plainText: boolean): NumberedLine | undefined => {
   const dotted = dottedNumber(text, context, plainText);
-  if (dotted === undefined || isAmount(dotted, patterns.countedAfter)) return undefined;
+  if (dotted === undefined || isAmount(dotted, patterns.countedAfter) || startsDate(text, patterns.dates)) return undefined;
   const numbered = { number: dotted.number, rest: dotted.rest, isHeading: context.isHeading };
   return numberInSentence(numbered, patterns.continuesSentence) ? undefined : dotted;
 };

package/src/structure/unspaced-heading.ts

@@ -0,0 +1,27 @@
+import type { Heading } from "../document.ts";
+import type { NumberingContext, StructurePatterns } from "../plugin.ts";
+import type { Line } from "./lines.ts";
+
+/** Hashes with no space before the words: 「##第一四条 （存続条項）」. CommonMark reads the line as a paragraph. */
+const UNSPACED = /^ {0,3}(?<hashes>#{1,6})(?<words>[^#\s].*)$/u;
+
+const ON_HEADING: NumberingContext = { open: [], isHeading: true };
+
+/**
+ * Lines (by number) written as a heading without the space after the hashes, kept only when the language reads the words as an article or
+ * a chapter. The writer's neighbours are headings of the same shape, so reading this one as a paragraph puts a gap in the
+ * article numbers that is not in the document. 「#タグ」 and 「#1 の件」 stay text.
+ */
+export const unspacedHeadings = (lines: readonly Line[], headed: ReadonlyMap<number, Heading>, patterns: StructurePatterns): [number, Heading][] =>
+  lines.flatMap((line) => {
+    if (headed.has(line.number)) return [];
+    const groups = UNSPACED.exec(line.text)?.groups;
+    const hashes = groups?.["hashes"];
+    const words = groups?.["words"]?.trim();
+    if (hashes === undefined || words === undefined) return [];
+    const kind = patterns.numbered(words, ON_HEADING)?.kind;
+    if (kind !== "article" && kind !== "chapter") return [];
+    const heading: Heading = { depth: hashes.length, text: words, start: line.start, end: line.start + line.text.length };
+    const entry: [number, Heading] = [line.number, heading];
+    return [entry];
+  });

package/src/structure/versioned-label.ts

@@ -0,0 +1,52 @@
+/**
+ * A label naming the version it is about: the version after the label and a word (End of support for 5.0.0:), or before
+ * it with a word (3.0.0 のサポート終了日：). The words come from the language's lexicon; a version is digits joined by dots,
+ * with or without a leading v. Pure.
+ */
+
+/** The words that join a version to a label: after for the label, word, version order; before for version, word, label. */
+export type VersionWords = { readonly after: readonly string[]; readonly before: readonly string[] };
+
+/** A list marker, a table cell's bar and emphasis before the label. A numbered marker needs its space: 3.0.0 is a version, not item 3. */
+const LINE_LEAD = /^[ \t]*(?:(?:[-*+]|\d{1,3}[.)])[ \t]+)?(?:\|[ \t]*)?[*_]*/u;
+const VERSION = /^v?\d+(?:\.\d+)+/iu;
+/** After the label's version, only the label's end: a colon or a table cell's bar. */
+const VERSION_END = /^[ \t]*[:：|]/u;
+const SPACES = /^[ \t\u3000]*/u;
+
+const spacesAt = (text: string): number => SPACES.exec(text)?.[0].length ?? 0;
+
+const versionAt = (text: string): number => VERSION.exec(text)?.[0].length ?? 0;
+
+/** After the label: one word, spaces on both sides, a version and the label's end. Returns the label's end on. */
+const afterVersion = (rest: string, word: string): string | undefined => {
+  const gap = spacesAt(rest);
+  if (gap === 0 || !rest.slice(gap).toLowerCase().startsWith(`${word.toLowerCase()} `)) return undefined;
+  const from = gap + word.length + spacesAt(rest.slice(gap + word.length));
+  const version = versionAt(rest.slice(from));
+  const tail = rest.slice(from + version);
+  return version > 0 && VERSION_END.test(tail) ? tail : undefined;
+};
+
+/** Before the label: a version, the word and the label, spaces allowed between. Returns the text from the label on. */
+const beforeVersion = (lead: string, word: string, label: string): string | undefined => {
+  const version = versionAt(lead);
+  if (version === 0) return undefined;
+  const atWord = version + spacesAt(lead.slice(version));
+  if (!lead.slice(atWord).startsWith(word)) return undefined;
+  const atLabel = atWord + word.length + spacesAt(lead.slice(atWord + word.length));
+  return lead.slice(atLabel).toLowerCase().startsWith(label.toLowerCase()) ? lead.slice(atLabel) : undefined;
+};
+
+/**
+ * The line read as the bare label: "End of support for 5.0.0: …" as "End of support: …", "3.0.0 のサポート終了日：…" as
+ * "サポート終了日：…". undefined when the line does not start with the label carrying a version that way.
+ */
+export const withoutLabelVersion = (text: string, label: string, words: VersionWords): string | undefined => {
+  const lead = text.replace(LINE_LEAD, "");
+  const after = lead.toLowerCase().startsWith(label.toLowerCase())
+    ? words.after.map((word) => afterVersion(lead.slice(label.length), word)).find((tail) => tail !== undefined)
+    : undefined;
+  if (after !== undefined) return `${lead.slice(0, label.length)}${after}`;
+  return words.before.map((word) => beforeVersion(lead, word, label)).find((rest) => rest !== undefined);
+};

package/src/structure/written-headings.ts

@@ -0,0 +1,137 @@
+// 平文の契約の見出し。Markdown の見出しを使わない契約（.txt）は、行の頭や文の切れ目に「9. Indemnification.」
+// 「Section 2.3   Appraisal Shares.」と番号と題を書く。木は本文の番号と見分けられないのでこれを番地にしないが、
+// 参照の行き先としては読める。目次の行（「Section 2.3 Appraisal Shares 4」）は題が点で閉じないので見出しにしない。
+
+/** 書かれた見出し。address は参照の target と同じ形（「6.1」）。start は見出しの位置。 */
+export type WrittenHeading = { readonly address: string; readonly start: number };
+
+/** 参照の位置と行き先。木の reference の節点から取る。 */
+export type ReferenceSpan = { readonly start: number; readonly end: number; readonly target: string };
+
+/** 題の語の数の上限。見出しの題は短い。 */
+const MAX_TITLE_WORDS = 15;
+/** 題の語: 文字で始まり、文字・アポストロフィ・アンパサンド・ハイフンが続き、読点やセミコロンで終わってもよい。数字は入らない。 */
+const TITLE_WORD = /^\p{L}[\p{L}’'&-]*[,;]?$/u;
+const CAPITAL = /^\p{Lu}/u;
+
+/**
+ * Title Case の題か（「Effect on Capital Stock of the Company and Merger Sub」）。最初の語が大文字で始まり、語の半分より多くが
+ * 大文字で始まり、数字を含まない。文（「The Company shall comply」）は小文字の語が多いので当たらない。Pure.
+ */
+export const isTitle = (title: string): boolean => {
+  const words = title.trim().split(/\s+/u);
+  const [first] = words;
+  if (first === undefined || first === "" || words.length > MAX_TITLE_WORDS || !words.every((word) => TITLE_WORD.test(word))) return false;
+  return CAPITAL.test(first) && words.filter((word) => CAPITAL.test(word)).length * 2 > words.length;
+};
+
+/** 点で閉じた題（「Appraisal Shares. As used」の「Appraisal Shares」）。点の後ろは空白か行末。無ければ undefined。 */
+const MAX_TITLE_LENGTH = 120;
+const titleClosedByDot = (text: string): string | undefined => {
+  const dot = text.slice(0, MAX_TITLE_LENGTH).search(/\.(?:\s|$)/u);
+  return dot <= 0 ? undefined : text.slice(0, dot);
+};
+
+/** 前の語が番号（「Section 8.07. Section 16 Matters」の 8.07.）なら、後ろはその見出しの題で、見出しではない。 */
+const NUMBER_BEFORE = /\d[.:]?$/u;
+const SENTENCE_END = /[.;:]$/u;
+const isSpace = (char: string | undefined): boolean => char === " " || char === "\t";
+
+/** 見出しが始まれる位置: 行の頭、文の終わり（. ; :）の後ろの空白、2 つ以上の空白の後ろ。前が番号なら始まらない。 */
+const startsHeading = (source: string, at: number): boolean => {
+  const before = source.slice(Math.max(source.lastIndexOf("\n", at - 1) + 1, at - MAX_TITLE_LENGTH), at);
+  let spaces = 0;
+  while (isSpace(before[before.length - 1 - spaces])) spaces += 1;
+  if (spaces === before.length) return true;
+  const word =
+    before
+      .slice(0, before.length - spaces)
+      .split(/[ \t]/u)
+      .at(-1) ?? "";
+  if (NUMBER_BEFORE.test(word)) return false;
+  return spaces >= 2 || (spaces >= 1 && SENTENCE_END.test(word));
+};
+
+/** 参照のすぐ後ろが題か（「Section 4.12. Compliance with Laws.」「Section 1.1      Closing. The Parties」）。 */
+const NUMBER_CLOSE = /^\.?[ \t]+/u;
+/** 番号と題のあいだの空白の長さの上限（データセットの契約は十数個の空白を置く）。 */
+const MAX_NUMBER_CLOSE = 24;
+const titledAfter = (source: string, end: number): boolean => {
+  const rest = source.slice(end, end + MAX_TITLE_LENGTH + MAX_NUMBER_CLOSE);
+  const close = NUMBER_CLOSE.exec(rest)?.[0];
+  const title = close === undefined ? undefined : titleClosedByDot(rest.slice(close.length));
+  return title !== undefined && isTitle(title);
+};
+
+/** 参照の形で書かれた見出し（「… Effect. Section 4.12. Compliance with Laws.」）。 */
+const headingsAtReferences = (references: readonly ReferenceSpan[], source: string): WrittenHeading[] =>
+  references
+    .filter((reference) => startsHeading(source, reference.start) && titledAfter(source, reference.end))
+    .map((reference) => ({ address: reference.target, start: reference.start }));
+
+/** 行の頭の番号（「9. Indemnification.」「4 Intellectual Property. 4.1 Each」「6.1 eGain warrants」「2 Quality and Food Safety」）。 */
+const LINE_NUMBER = /^[ \t]{0,3}(?<number>\d{1,3}(?:\.\d{1,3}){0,5})\.?[ \t]+(?<rest>\S.*)$/u;
+/** 語のどこかに大文字があるか（「eGain」も名前）。 */
+const HAS_CAPITAL = /\p{Lu}/u;
+
+/**
+ * 行の頭の番号が見出しか。題が点で閉じ、同じ行に本文が続けば見出し（「9. Indemnification. Each party …」）。題だけの行
+ * （「1. Introduction」）は箇条書きと見分けられないので読まない。点を含む番号（6.1）は、最初の語に大文字があり、題より長い
+ * 段落が続けば段落の番号（「6.1 eGain warrants …」）。本文の折り返しの「2.5 percent」は小文字、目次の「3.17 Reserves 33」は短いので当たらない。
+ */
+const isLineHeading = (number: string, rest: string): boolean => {
+  const title = titleClosedByDot(rest);
+  if (title !== undefined && rest.slice(title.length + 1).trim() !== "" && isTitle(title)) return true;
+  const words = rest.split(/\s+/u);
+  return number.includes(".") && words.length > MAX_TITLE_WORDS && HAS_CAPITAL.test(words[0] ?? "");
+};
+
+const headingsAtLineStarts = (source: string): WrittenHeading[] => {
+  const found: WrittenHeading[] = [];
+  let start = 0;
+  source.split("\n").forEach((line) => {
+    const groups = LINE_NUMBER.exec(line)?.groups;
+    const number = groups?.["number"];
+    if (number !== undefined && isLineHeading(number, groups?.["rest"] ?? "")) found.push({ address: number, start });
+    start += line.length + 1;
+  });
+  return found;
+};
+
+/** 文書に平文で書かれた見出しを、文書の順に。Pure. */
+export const writtenHeadings = (references: readonly ReferenceSpan[], source: string): WrittenHeading[] =>
+  [...headingsAtReferences(references, source), ...headingsAtLineStarts(source)].toSorted((left, right) => left.start - right.start);
+
+/** 括弧の小見出し（「(i)」「(b)」「(3)」）。参照の一部（「Section 2.02(b)」「(a)(i)」）は、前が文字・数字・括弧なので読まない。 */
+const SUBDIVISION_LABEL = /(?<![\p{L}\p{N})\]])\((?<label>[a-zA-Z]{1,4}|\d{1,3})\)/gu;
+
+/** 本文に書かれた括弧の小見出しの札。Pure. */
+export const writtenSubdivisions = (body: string): ReadonlySet<string> =>
+  new Set([...body.matchAll(SUBDIVISION_LABEL)].flatMap((match) => (match.groups?.["label"] === undefined ? [] : [match.groups["label"]])));
+
+/** 参照の札（「Section 2.1(a)(i)」の a と i）。 */
+const LABEL_IN_REFERENCE = /\((?<label>[a-zA-Z]{1,4}|\d{1,3})\)/gu;
+
+/**
+ * 参照の番地「2.1.a.i」を、親「2.1」と、参照が括弧で書いた札「a」「i」に分ける。札を括弧で書いていない参照（Section 2.1）や、
+ * 番地の末尾が書いた札と合わなければ undefined。Pure.
+ */
+export const parentAndLabels = (target: string, written: string): { readonly parent: string; readonly labels: readonly string[] } | undefined => {
+  const labels = [...written.matchAll(LABEL_IN_REFERENCE)].flatMap((match) => (match.groups?.["label"] === undefined ? [] : [match.groups["label"]]));
+  const parts = target.split(".");
+  const cut = parts.length - labels.length;
+  if (labels.length === 0 || cut <= 0 || parts.slice(cut).join(".") !== labels.join(".")) return undefined;
+  return { parent: parts.slice(0, cut).join("."), labels };
+};
+
+/** 書かれた見出しごとの範囲（次の書かれた見出しまで）。同じ番地が二度あれば、どちらの範囲もその番地のもの。Pure. */
+export const headingRegions = (headings: readonly WrittenHeading[], length: number): ReadonlyMap<string, readonly (readonly [number, number])[]> => {
+  const regions = new Map<string, (readonly [number, number])[]>();
+  headings.forEach((heading, index) => {
+    const end = headings[index + 1]?.start ?? length;
+    const found = regions.get(heading.address) ?? [];
+    found.push([heading.start, end]);
+    regions.set(heading.address, found);
+  });
+  return regions;
+};

package/src/structure/written-list-path.ts

@@ -0,0 +1,46 @@
+// 箇条書きで書いた小見出し。「- (b) Composition: …」の下に字下げして「- i. one representative …」と書いた文書は、
+// 「Section 8(b)(i)」の (b) と i を本文の括弧ではなく行の頭の札で書く。札の入れ子は字下げで読む。
+
+/** 行の頭の札。indent は札の前の字下げの幅（タブは 4）。 */
+export type ListLabel = { readonly label: string; readonly indent: number };
+
+/** 字下げ、箇条書きの記号（-、*、+）、札、空白、本文。 */
+const LIST_LINE = /^(?<indent>[ \t]*)(?:[-*+][ \t]+)?(?<mark>\S+)[ \t]+\S/u;
+/** 札「(b)」「i.」「2)」。 */
+const MARK = /^\(?(?<label>[a-zA-Z]{1,4}|\d{1,3})[.)]$/u;
+
+/** 札の中身。開き括弧で始まる札は閉じ括弧で終わる（「(b.」は札ではない）。 */
+const labelOf = (mark: string): string | undefined => {
+  const label = MARK.exec(mark)?.groups?.["label"];
+  return mark.startsWith("(") && !mark.endsWith(")") ? undefined : label;
+};
+const TAB_WIDTH = 4;
+
+const widthOf = (indent: string): number => [...indent].reduce((width, char) => width + (char === "\t" ? TAB_WIDTH : 1), 0);
+
+/** 本文の行の頭に書かれた札を、文書の順に。Pure. */
+export const listLabels = (body: string): ListLabel[] =>
+  body.split("\n").flatMap((line) => {
+    const groups = LIST_LINE.exec(line)?.groups;
+    const label = labelOf(groups?.["mark"] ?? "");
+    return label === undefined ? [] : [{ label, indent: widthOf(groups?.["indent"] ?? "") }];
+  });
+
+/** 札 index の下（次に同じか浅い字下げの札が来るまで）の終わり。 */
+const scopeEnd = (items: readonly ListLabel[], index: number, end: number): number => {
+  const depth = items[index]?.indent ?? 0;
+  const next = items.findIndex((item, at) => at > index && at < end && item.indent <= depth);
+  return next === -1 ? end : next;
+};
+
+/** items[from, end) の一番浅い字下げの札（直下の札）から、labels を入れ子の順にたどれるか。深い札を飛ばして親にしない。 */
+const pathFrom = (items: readonly ListLabel[], labels: readonly string[], from: number, end: number): boolean => {
+  const [first, ...rest] = labels;
+  if (first === undefined) return true;
+  const scope = items.slice(from, end);
+  const top = Math.min(...scope.map((item) => item.indent));
+  return scope.some((item, at) => item.indent === top && item.label === first && pathFrom(items, rest, from + at + 1, scopeEnd(items, from + at, end)));
+};
+
+/** 札「b」「iv」が、行の頭の札として「(b)」の下に「iv.」と入れ子で書かれているか。札が無ければ false。Pure. */
+export const hasListPath = (items: readonly ListLabel[], labels: readonly string[]): boolean => labels.length > 0 && pathFrom(items, labels, 0, items.length);

package/src/structure/written-resolve.ts

@@ -0,0 +1,64 @@
+import type { StructureNode } from "../plugin.ts";
+import { headingRegions, parentAndLabels, writtenHeadings, writtenSubdivisions, type WrittenHeading } from "./written-headings.ts";
+import { hasListPath, listLabels } from "./written-list-path.ts";
+
+// 木が番地にしない平文の見出しと、本文の中に書いた括弧の小見出し（「6.1 … that (i) the Software …」）を、参照の行き先として引く。
+
+type Region = readonly [number, number];
+
+/** addresses: 平文で書かれた見出しの番地。subdivision: 参照の札が、親の範囲の本文に書かれているか。 */
+export type WrittenReading = { readonly addresses: readonly string[]; readonly subdivision: (reference: StructureNode) => boolean };
+
+const targetOf = (node: StructureNode): string => String(node.attrs["target"] ?? "");
+
+/**
+ * 括弧の小見出しを読む範囲。木の条は、項や号を一つも読まなかったときだけ、その条の始まりから次の平文の見出しまで。
+ * 項や号を読んでいれば、無い札は本当に無い。
+ */
+const treeRegions = (nodes: readonly StructureNode[], headings: readonly WrittenHeading[]): Map<string, Region[]> => {
+  const regions = new Map<string, Region[]>();
+  // 条も見出しも文書の順なので、次の見出しは前へ進めるだけで見つかる。
+  let next = 0;
+  nodes
+    .filter((node) => node.kind === "article" && !node.children.some((child) => child.kind === "item"))
+    .toSorted((left, right) => left.span.start - right.span.start)
+    .forEach((node) => {
+      while ((headings[next]?.start ?? Infinity) <= node.span.start) next += 1;
+      const end = Math.min(headings[next]?.start ?? node.span.end, node.span.end);
+      const found = regions.get(node.address) ?? [];
+      found.push([node.span.start, end]);
+      regions.set(node.address, found);
+    });
+  return regions;
+};
+
+/** 範囲の本文を read で読んだ結果。範囲ごとに、要るときに一度だけ読む。 */
+const perRegion = <T>(source: string, read: (body: string) => T): ((region: Region) => T) => {
+  const cache = new Map<string, T>();
+  return ([start, end]) => {
+    const key = `${String(start)}:${String(end)}`;
+    const found = cache.get(key) ?? read(source.slice(start, end));
+    cache.set(key, found);
+    return found;
+  };
+};
+
+/**
+ * 平文の見出しの番地と、参照（Section 6.1(i)）の親（6.1）が木の条か平文の見出しとしてあり、その範囲の本文に札「(i)」が
+ * 書かれているか（札を行の頭に入れ子で書いた箇条書きも読む）を引く関数。Pure.
+ */
+export const writtenReading = (nodes: readonly StructureNode[], source: string): WrittenReading => {
+  const references = nodes.filter((node) => node.kind === "reference").map((node) => ({ start: node.span.start, end: node.span.end, target: targetOf(node) }));
+  const headings = writtenHeadings(references, source);
+  const written = headingRegions(headings, source.length);
+  const tree = treeRegions(nodes, headings);
+  const labelsIn = perRegion(source, writtenSubdivisions);
+  const listIn = perRegion(source, listLabels);
+  const regionsOf = (address: string): Region[] => [...(tree.get(address) ?? []), ...(written.get(address) ?? [])];
+  const subdivision = (reference: StructureNode): boolean => {
+    const wanted = parentAndLabels(targetOf(reference), String(reference.attrs["label"] ?? ""));
+    if (wanted === undefined) return false;
+    return regionsOf(wanted.parent).some((region) => wanted.labels.every((label) => labelsIn(region).has(label)) || hasListPath(listIn(region), wanted.labels));
+  };
+  return { addresses: headings.map((heading) => heading.address), subdivision };
+};

package/src/title-line.ts

@@ -0,0 +1,52 @@
+// A title written on a line of its own, as a plain-text heading (Exhibit A to Merger Agreement, ANNEX 1): a few words with
+// no full stop, and a blank line or the document's edge on both sides. Pure.
+
+/**
+ * How a reader tells a title from a sentence. dangling-figure-reference reads a caption of any labelled thing; duplicate-
+ * definition reads where an attached instrument starts. Their notions differ, so each says how.
+ */
+export type TitleNotion = {
+  /** A title ending in one of these is a sentence or runs on. duplicate-definition also refuses a trailing `: ; ,`. */
+  readonly sentenceEnd: RegExp;
+  /** Every word longer than three letters starts with a capital. dangling-figure-reference only: Figure 3 shows the trend. */
+  readonly capitalised: boolean;
+  /**
+   * "any-whitespace": the document's edge may be reached across any whitespace (a line of U+3000 or NBSP before the
+   * first line). "blank-line": only lines of spaces and tabs count. dangling-figure-reference takes the first.
+   */
+  readonly edge: "any-whitespace" | "blank-line";
+};
+
+/** A title runs at most this many words after its label and number. */
+const TITLE_MAX_WORDS = 6;
+/** A word this long or shorter may stay in lower case in a title (to, of, the, and). */
+const TITLE_SMALL_WORD = 3;
+const BLANK_LINE = /\n[ \t\r]*\n[ \t\r]*$/u;
+const ENDS_LINE = /^[ \t\r]*\n/u;
+const BLANK = /^[ \t\r]*$/u;
+const LOWER_START = /^\p{Ll}/u;
+
+const isShortTitle = (rest: string, notion: TitleNotion): boolean => {
+  const title = rest.trim();
+  const words = title.split(/\s+/u).filter((word) => word !== "");
+  return (
+    words.length <= TITLE_MAX_WORDS &&
+    !notion.sentenceEnd.test(title) &&
+    (!notion.capitalised || words.every((word) => word.length <= TITLE_SMALL_WORD || !LOWER_START.test(word)))
+  );
+};
+
+const isAloneBefore = (before: string, notion: TitleNotion): boolean =>
+  before === "" || BLANK_LINE.test(`\n${before}`) || (notion.edge === "any-whitespace" && before.trim() === "");
+
+const isAloneAfter = (after: string, isLastLine: boolean, notion: TitleNotion): boolean =>
+  isLastLine || ENDS_LINE.test(after) || BLANK.test(after) || (notion.edge === "any-whitespace" && after.trim() === "");
+
+/**
+ * The line from lineStart to lineEnd (-1: the last line) is a title: rest, the words after its label and number, are a
+ * short title, and a blank line or the document's edge is on both sides.
+ */
+export const isTitleLine = (source: string, lineStart: number, lineEnd: number, rest: string, notion: TitleNotion): boolean =>
+  isShortTitle(rest, notion) &&
+  isAloneBefore(source.slice(0, lineStart), notion) &&
+  isAloneAfter(lineEnd === -1 ? "" : source.slice(lineEnd + 1), lineEnd === -1, notion);