chaffjs 0.27.1 → 0.29.0

package/src/company-names.ts

@@ -0,0 +1,202 @@
+import { nameKey } from "./name-variants.ts";
+import { escapeRegExp } from "./orthography.ts";
+
+// 会社の名前を、文書の中で二通りに書いた所（株式会社アオバシステム と アオバシステム株式会社、Aoba Systems Inc. と Aoba Systems Ltd.）。
+// 会社の形の語（株式会社、Inc）は語彙表 company-form が言う。group が同じ語は同じ形（Inc と Incorporated、株式会社 と (株)）、
+// position は名前のどちら側に立つか（無ければどちらにも立つ）。名前の部分は、形の語のすぐ隣の、名前に使う字の連なり。
+
+export type CompanyForm = { readonly pattern: string; readonly group: string; readonly position?: "before" | "after" | undefined };
+
+/** surface は形の語も含めた書いたまま。base は名前の部分、form は形の組、position は形の語が名前のどちら側にあるか。 */
+export type CompanyMention = {
+  readonly surface: string;
+  readonly offset: number;
+  readonly base: string;
+  readonly form: string;
+  readonly position: "before" | "after";
+};
+
+/** 二つの書き方の違い。spelling は記号・幅・大小だけ、form は会社の形、position は形の語の側、near は名前の一字違い。 */
+type CompanyRelation = "spelling" | "form" | "position" | "near";
+
+type CompanyVariant = { readonly mention: CompanyMention; readonly usual: string; readonly kind: CompanyRelation };
+
+/** 名前の部分が、品詞解析が固有名詞と読んだ語に掛かるか。漢字だけの名前（清算、当該）は、ふつうの語と区別がつかない。 */
+export type IsProper = (start: number, end: number) => boolean;
+
+const LATIN = /\p{Script=Latin}/u;
+const LETTER = /\p{L}/u;
+/** 漢字・かなの名前に使う字。ひらがなは名前の外（の、は、まで）。 */
+const CJK_NAME = /[\p{Script=Katakana}\p{Script=Han}\p{Script=Latin}\p{N}ー・&＆]/u;
+const NOT_HAN_LETTER = /(?!\p{Script=Han})\p{L}/u;
+/** 英字の名前の一語。大文字か数字で始まる。 */
+const LATIN_WORD = /^[\p{Lu}\p{N}][\p{L}\p{N}&'’.-]*$/u;
+const AMPERSAND = "&";
+/** 形の語のすぐあとに空白一つで大文字の語が続けば、形の語は長い名前の一部（Limited Partnership、Corporation Tax）。Inc. のあとの文や行は続きでない。 */
+const CONTINUED = /^ \p{Lu}/u;
+
+const occurrencesOf = (source: string, pattern: string): number[] => [...source.matchAll(new RegExp(escapeRegExp(pattern), "gu"))].map((match) => match.index);
+
+/** 英字の語の続き。ハイフンでつないだ語（Limited-time）も一つの語。 */
+const WORD_CONTINUES = /[\p{L}\p{N}-]/u;
+
+/** 英字の形の語は、語の続き（Including の Inc、Limited-time の Limited）では形の語と読まない。 */
+const standsAsWord = (source: string, pattern: string, at: number): boolean =>
+  !LATIN.test(pattern) || (!LETTER.test(source.charAt(at - 1)) && !WORD_CONTINUES.test(source.charAt(at + pattern.length)));
+
+type Span = { readonly start: number; readonly end: number };
+
+/** 英字の名前: 形の語の前の、大文字で始まる語の続き。頭の、固有名詞と読めない語（文頭の Contact）は外す。 */
+const latinBaseBefore = (source: string, end: number, isProper: IsProper): Span | undefined => {
+  const before = source.slice(0, end).replace(/,?\s$/u, "");
+  const words = [...before.matchAll(/\S+/gu)].map((match): Span => ({ start: match.index, end: match.index + match[0].length }));
+  const isNameWord = (span: Span): boolean => {
+    const word = source.slice(span.start, span.end);
+    return LATIN_WORD.test(word) || word === AMPERSAND;
+  };
+  const run = words.slice(
+    words.findLastIndex((span, index) => !isNameWord(span) || before.slice(span.end, words[index + 1]?.start ?? before.length).includes("\n")) + 1,
+  );
+  const first = run.find((span) => isProper(span.start, span.end));
+  return first === undefined ? undefined : { start: first.start, end: before.length };
+};
+
+const CJK_BEFORE = new RegExp(`${CJK_NAME.source}+$`, "u");
+const CJK_AFTER = new RegExp(`^${CJK_NAME.source}+`, "u");
+const SPACES: ReadonlySet<string> = new Set([" ", "\u3000"]);
+
+/** 漢字・かなの名前: 形の語の隣の、名前に使う字の続き。形の語との間の空白一つは名前の外。 */
+const cjkBaseBefore = (source: string, end: number): Span | undefined => {
+  const stop = SPACES.has(source.charAt(end - 1)) ? end - 1 : end;
+  const match = CJK_BEFORE.exec(source.slice(0, stop));
+  return match === null ? undefined : { start: match.index, end: stop };
+};
+
+const cjkBaseAfter = (source: string, start: number): Span | undefined => {
+  const from = SPACES.has(source.charAt(start)) ? start + 1 : start;
+  const match = CJK_AFTER.exec(source.slice(from));
+  return match === null ? undefined : { start: from, end: from + match[0].length };
+};
+
+/** 名前の部分の字数。一字（法令の項目の印 イ、ロ）は名前と読まない。 */
+const MIN_BASE_LENGTH = 2;
+
+/** 名前と読めるか。二字以上で、記号で始まり終わらず、漢字だけなら固有名詞と読めるもの。 */
+const isNameBase = (source: string, span: Span, isProper: IsProper): boolean => {
+  const base = source.slice(span.start, span.end);
+  if ([...base].length < MIN_BASE_LENGTH || /^[・&＆]|[・&＆]$/u.test(base)) return false;
+  return NOT_HAN_LETTER.test(base) || isProper(span.start, span.end);
+};
+
+const mentionOf = (source: string, base: Span, form: Span, group: string): CompanyMention => {
+  const position = form.start < base.start ? "before" : "after";
+  const [start, end] = [Math.min(base.start, form.start), Math.max(base.end, form.end)];
+  return { surface: source.slice(start, end), offset: start, base: source.slice(base.start, base.end), form: group, position };
+};
+
+const baseBefore = (source: string, at: number, latin: boolean, isProper: IsProper): Span | undefined =>
+  latin ? latinBaseBefore(source, at, isProper) : cjkBaseBefore(source, at);
+
+/** 形の語の一つの現れから、会社の名前の現れ。名前が前にあればそれを、無ければ後ろを取る。 */
+const mentionAt = (source: string, form: CompanyForm, at: number, isProper: IsProper): CompanyMention | undefined => {
+  const formSpan = { start: at, end: at + form.pattern.length };
+  const latin = LATIN.test(form.pattern);
+  if (latin && CONTINUED.test(source.slice(formSpan.end))) return undefined;
+  const before = form.position === "before" ? undefined : baseBefore(source, at, latin, isProper);
+  if (before !== undefined && isNameBase(source, before, isProper)) return mentionOf(source, before, formSpan, form.group);
+  const after = form.position === "after" || latin ? undefined : cjkBaseAfter(source, formSpan.end);
+  return after !== undefined && isNameBase(source, after, isProper) ? mentionOf(source, after, formSpan, form.group) : undefined;
+};
+
+/** 文書の中の会社の名前の現れ。長い形の語（Co., Ltd）を先に取り、その中の短い語（Ltd）は数えない。 */
+export const companyMentionsIn = (source: string, forms: readonly CompanyForm[], isProper: IsProper): CompanyMention[] => {
+  const taken: Span[] = [];
+  return forms
+    .toSorted((left, right) => right.pattern.length - left.pattern.length)
+    .flatMap((form) =>
+      occurrencesOf(source, form.pattern).flatMap((at) => {
+        const end = at + form.pattern.length;
+        if (!standsAsWord(source, form.pattern, at) || taken.some((span) => span.start < end && at < span.end)) return [];
+        taken.push({ start: at, end });
+        const mention = mentionAt(source, form, at, isProper);
+        return mention === undefined ? [] : [mention];
+      }),
+    )
+    .toSorted((left, right) => left.offset - right.offset);
+};
+
+const MIN_SLIP_LENGTH = 7;
+
+const spliced = (chars: readonly string[], at: number, count: number): string => chars.toSpliced(at, count).join("");
+const firstDifference = (left: readonly string[], right: readonly string[]): number => left.findIndex((char, index) => char !== right[index]);
+
+/** 一字の抜けなら、その字。 */
+const droppedChars = (longer: readonly string[], shorter: readonly string[]): string[] | undefined => {
+  if (longer.length !== shorter.length + 1) return undefined;
+  const at = firstDifference(longer, shorter);
+  return spliced(longer, at, 1) === shorter.join("") ? [longer[at] ?? ""] : undefined;
+};
+
+/** 一字の置き換えか、隣どうしの入れ替えなら、その字。 */
+const replacedChars = (left: readonly string[], right: readonly string[]): string[] | undefined => {
+  const at = firstDifference(left, right);
+  const changed = [left[at] ?? "", right[at] ?? ""];
+  if (left.length !== right.length || at < 0) return undefined;
+  if (spliced(left, at, 1) === spliced(right, at, 1)) return changed;
+  const swapped = left[at] === right[at + 1] && left[at + 1] === right[at] && spliced(left, at, 2) === spliced(right, at, 2);
+  return swapped ? changed : undefined;
+};
+
+/**
+ * 名前の部分が一字違いか（置き換え・抜け・隣どうしの入れ替え）。違う字が漢字なら見ない。漢字の一字違い（日本電気 と 日本電機）
+ * は別の会社のことがある。短い名前（Alpha と Alpho）も別の会社のことが多い。
+ */
+const isSlip = (left: string, right: string): boolean => {
+  const [leftChars, rightChars] = [[...left], [...right]];
+  if (Math.min(leftChars.length, rightChars.length) < MIN_SLIP_LENGTH) return false;
+  const changed = replacedChars(leftChars, rightChars) ?? droppedChars(leftChars, rightChars) ?? droppedChars(rightChars, leftChars);
+  return changed !== undefined && changed.every((char) => NOT_HAN_LETTER.test(char));
+};
+
+/** 名前の部分が同じ二つ。同じ形の短い書き方と長い書き方（株式会社 と (株)、Inc と Incorporated）は書き分けと見ない。 */
+const sameNameRelation = (left: CompanyMention, right: CompanyMention): CompanyRelation | undefined => {
+  if (left.form !== right.form) return "form";
+  if (left.position !== right.position) return "position";
+  return nameKey(left.surface) === nameKey(right.surface) ? "spelling" : undefined;
+};
+
+/**
+ * 二つの書き方が、同じ会社を二通りに書いたものか。名前の部分が同じ（記号・幅・大小を除いて）なら、形の組か形の語の側の違い、
+ * どちらも同じなら記号・幅・大小だけの違い。名前の部分が違うのは、形も側も同じで、漢字でない字の一字違いのときだけ（Aoba Systems と
+ * Aoba System）。名前の違う会社（親会社と子会社、別の取引先）は、名前の部分が一字より多く違うので同じと見ない。
+ */
+export const companyRelation = (left: CompanyMention, right: CompanyMention): CompanyRelation | undefined => {
+  if (left.surface === right.surface) return undefined;
+  const [leftKey, rightKey] = [nameKey(left.base), nameKey(right.base)];
+  if (leftKey === rightKey) return sameNameRelation(left, right);
+  return left.form === right.form && left.position === right.position && isSlip(leftKey, rightKey) ? "near" : undefined;
+};
+
+type Tally = { readonly first: CompanyMention; readonly count: number };
+
+const talliesOf = (mentions: readonly CompanyMention[]): Tally[] => {
+  const tallies = new Map<string, Tally>();
+  mentions.forEach((mention) => {
+    const tally = tallies.get(mention.surface);
+    tallies.set(mention.surface, { first: tally?.first ?? mention, count: (tally?.count ?? 0) + 1 });
+  });
+  return [...tallies.values()];
+};
+
+/** 多いほうが先。同数なら先に書いたほう。 */
+const byUsage = (left: Tally, right: Tally): number => right.count - left.count || left.first.offset - right.first.offset;
+
+/** 同じ会社の、少ないほうの書き方。同じ会社と言える相手のうち一番多いものと比べる。書き方ごとに最初の現れを一つ。 */
+export const companyVariants = (mentions: readonly CompanyMention[]): CompanyVariant[] => {
+  const tallies = talliesOf(mentions);
+  return tallies.flatMap((tally) => {
+    const [usual] = tallies.filter((other) => other === tally || companyRelation(tally.first, other.first) !== undefined).toSorted(byUsage);
+    const kind = usual === undefined || usual === tally ? undefined : companyRelation(tally.first, usual.first);
+    return usual === undefined || kind === undefined ? [] : [{ mention: tally.first, usual: usual.first.surface, kind }];
+  });
+};

package/src/default-values.ts

@@ -0,0 +1,158 @@
+// One parameter's default stated two ways on a page: the signature says `width = 72`, the options table says 72, and the
+// prose says "`width` defaults to 80". Pure; reads the Markdown source. The words of a default column (Default, 既定値)
+// and of a default in prose (defaults to, の既定値は) come from the language's lexicons.
+import { codeFences } from "./detectors/code-fences.ts";
+
+/** Where a default is stated, from the most to the least authoritative: the code, a table, a sentence. */
+type Source = "code" | "table" | "prose";
+type Stated = { readonly name: string; readonly value: string; readonly offset: number; readonly source: Source };
+
+/** A default that disagrees with the one the code (or else a table) states. */
+export type DefaultSlip = { readonly name: string; readonly value: string; readonly offset: number; readonly expected: string };
+
+export type DefaultWords = { readonly headers: readonly string[]; readonly phrases: readonly string[] };
+
+type Line = { readonly text: string; readonly start: number };
+
+const RANK: Readonly<Record<Source, number>> = { code: 0, table: 1, prose: 2 };
+/** The values read: a code span, a quoted string, or a literal (a number, true, false, null). */
+const QUOTED: readonly RegExp[] = [/^`([^`]+)`/u, /^"([^"]*)"/u, /^'([^']*)'/u];
+const LITERAL = /^(?:-?\d+(?:\.\d+)?|true|false|True|False|null|none|None|undefined)(?!\w|\.\d)/u;
+const NAME_SPAN = /`([A-Za-z_$][\w$]*)`/gu;
+const IDENTIFIER = /^[A-Za-z_$][\w$]*/u;
+const SPREAD = /^(?:\.\.\.|\*{1,2})/u;
+/** How far after the name a default phrase may start: "`width` defaults to", "`width` の既定値は". */
+const PHRASE_REACH = 4;
+/** A declaration, not a call: a keyword first, or a return type after the list. A call's width=40 is an argument. */
+const KEYWORD_FIRST = /^[ \t]*(?:(?:export|async)[ \t]+)*(?:def|function\*?|fn|func)[ \t]/u;
+const RETURN_TYPE = /\)[ \t]*(?::|->)/u;
+const DELIMITER_ROW = /^[ \t|:-]+$/u;
+
+const linesOf = (source: string): Line[] => {
+  const starts = [0, ...[...source.matchAll(/\n/gu)].map((match) => match.index + 1)];
+  return starts.map((start, index) => ({ text: source.slice(start, (starts[index + 1] ?? source.length + 1) - 1), start }));
+};
+
+const valueAt = (text: string): string | undefined => {
+  const trimmed = text.trimStart();
+  const quoted = QUOTED.map((pattern) => pattern.exec(trimmed)?.[1]).find((value) => value !== undefined);
+  return quoted ?? LITERAL.exec(trimmed)?.[0];
+};
+
+/** The text between a line's first ( and the ) that closes it. */
+const parameterList = (text: string): { readonly list: string; readonly from: number } | undefined => {
+  const open = text.indexOf("(");
+  const state = { depth: 0, close: -1 };
+  Array.from(text.slice(open)).some((char, index) => {
+    if (char === "(") state.depth += 1;
+    if (char === ")") state.depth -= 1;
+    if (state.depth === 0) state.close = open + index;
+    return state.depth === 0;
+  });
+  return open === -1 || state.close === -1 ? undefined : { list: text.slice(open + 1, state.close), from: open + 1 };
+};
+
+/** width = 72, width: int = 72, *args: a parameter's name and its default value, if it has one. */
+const parameterDefault = (part: string): { readonly name: string; readonly value: string } | undefined => {
+  const trimmed = part.trim().replace(SPREAD, "");
+  const name = IDENTIFIER.exec(trimmed)?.[0];
+  const equals = trimmed.search(/=(?!=)/u);
+  const value = equals === -1 ? undefined : valueAt(trimmed.slice(equals + 1));
+  return name === undefined || value === undefined ? undefined : { name, value };
+};
+
+const QUOTES = "\"'`";
+
+/** A parameter list split at its commas, leaving a comma inside quotes or brackets (sep = ", ") in its parameter. */
+const splitParameters = (list: string): string[] => {
+  const state = { quote: "", depth: 0, from: 0 };
+  const parts: string[] = [];
+  [...list].forEach((char, index) => {
+    if (state.quote !== "") {
+      if (char === state.quote) state.quote = "";
+      return;
+    }
+    if (QUOTES.includes(char)) state.quote = char;
+    if ("([{".includes(char)) state.depth += 1;
+    if (")]}".includes(char)) state.depth -= 1;
+    if (char !== "," || state.depth !== 0) return;
+    parts.push(list.slice(state.from, index));
+    state.from = index + 1;
+  });
+  return [...parts, list.slice(state.from)];
+};
+
+const isDeclaration = (text: string): boolean => KEYWORD_FIRST.test(text) || RETURN_TYPE.test(text);
+
+const inCodeLine = (line: Line, base: number): Stated[] => {
+  const params = isDeclaration(line.text) ? parameterList(line.text) : undefined;
+  if (params === undefined) return [];
+  return splitParameters(params.list).flatMap((part, index, parts) => {
+    const found = parameterDefault(part);
+    const from = params.from + parts.slice(0, index).reduce((length, earlier) => length + earlier.length + 1, 0);
+    return found === undefined ? [] : [{ ...found, offset: base + line.start + line.text.indexOf(found.name, from), source: "code" as const }];
+  });
+};
+
+const inCode = (source: string): Stated[] =>
+  codeFences(source).flatMap((fence) => linesOf(source.slice(fence.start, fence.end)).flatMap((line) => inCodeLine(line, fence.start)));
+
+const cellsOf = (text: string): string[] =>
+  text
+    .split("|")
+    .slice(1, -1)
+    .map((cell) => cell.trim());
+
+/** The rows of every table with a default column, as (first cell name, default cell). */
+const inTables = (lines: readonly Line[], headers: ReadonlySet<string>): Stated[] =>
+  lines.flatMap((line, index) => {
+    const column = cellsOf(line.text).findIndex((cell) => headers.has(cell.replace(/[*`]/gu, "").toLowerCase()));
+    const delimiter = lines[index + 1]?.text ?? "";
+    if (column <= 0 || !delimiter.includes("---") || !DELIMITER_ROW.test(delimiter)) return [];
+    const end = lines.findIndex((row, at) => at > index + 1 && !row.text.trimStart().startsWith("|"));
+    return lines.slice(index + 2, end === -1 ? lines.length : end).flatMap((row) => {
+      const cells = cellsOf(row.text);
+      const name = /^`?([A-Za-z_$][\w$]*)`?$/u.exec(cells[0] ?? "")?.[1];
+      const value = valueAt(cells[column] ?? "");
+      return name === undefined || value === undefined ? [] : [{ name, value, offset: row.start + row.text.indexOf(name), source: "table" as const }];
+    });
+  });
+
+/** "`width` defaults to 80", "`width` の既定値は 80 文字": a code name, a default phrase right after it, then the value. */
+const inProse = (lines: readonly Line[], phrases: readonly string[]): Stated[] =>
+  lines.flatMap((line) =>
+    [...line.text.matchAll(NAME_SPAN)].flatMap((match) => {
+      const after = line.text.slice(match.index + match[0].length);
+      const phrase = phrases.map((candidate) => ({ candidate, at: after.toLowerCase().indexOf(candidate) })).find(({ at }) => at !== -1 && at <= PHRASE_REACH);
+      const value = phrase === undefined ? undefined : valueAt(after.slice(phrase.at + phrase.candidate.length));
+      return value === undefined ? [] : [{ name: match[1] ?? "", value, offset: line.start + match.index + 1, source: "prose" as const }];
+    }),
+  );
+
+/** Python writes the same literals capitalised: True is true, None is null. */
+const LITERAL_SPELLINGS: Readonly<Record<string, string>> = { True: "true", False: "false", None: "null", none: "null" };
+
+const normalized = (value: string): string => LITERAL_SPELLINGS[value] ?? value;
+
+const sameValue = (left: string, right: string): boolean =>
+  normalized(left) === normalized(right) || (Number.isFinite(Number(left)) && Number(left) === Number(right) && left !== "" && right !== "");
+
+/**
+ * Every stated default that differs from the name's most authoritative statement (the code, else a table). Two
+ * sentences that disagree with no code or table to settle it are left alone.
+ */
+export const defaultSlips = (source: string, words: DefaultWords): DefaultSlip[] => {
+  const fences = codeFences(source);
+  const lines = linesOf(source).filter((line) => !fences.some((fence) => line.start >= fence.start && line.start < fence.end));
+  const stated = [...inCode(source), ...inTables(lines, new Set(words.headers)), ...inProse(lines, words.phrases)];
+  return stated.flatMap((statement) => {
+    const above = stated.filter((other) => other.name === statement.name && RANK[other.source] < RANK[statement.source]);
+    const top = Math.min(...above.map((other) => RANK[other.source]));
+    const authorities = above.filter((other) => RANK[other.source] === top);
+    // Two functions on one page may give one parameter name two defaults; then no statement settles the others.
+    const authority = authorities.every((other) => sameValue(other.value, authorities[0]?.value ?? "")) ? authorities[0] : undefined;
+    return authority === undefined || sameValue(authority.value, statement.value)
+      ? []
+      : [{ name: statement.name, value: statement.value, offset: statement.offset, expected: authority.value }];
+  });
+};

package/src/deprecated-use.ts

@@ -0,0 +1,122 @@
+// A name the document calls deprecated (`indent` is deprecated; | `indent` | 0 | 非推奨 |) still used in one of its code
+// examples ({ indent: 2 }, --indent, indent=2). Pure; reads the Markdown source. The words that mark a name deprecated
+// (deprecated, 非推奨) and the phrases that negate them come from the language's lexicons.
+import { codeFences } from "./detectors/code-fences.ts";
+
+/** A deprecated name used in code: where it is used, and where the document calls it deprecated. */
+export type DeprecatedUse = { readonly name: string; readonly offset: number; readonly marked: number };
+
+/** The lowercase words that mark a name deprecated, and the phrases that negate one ("not deprecated", 非推奨ではな). */
+export type DeprecationWords = { readonly words: readonly string[]; readonly negated: readonly string[] };
+
+type Line = { readonly text: string; readonly start: number };
+type Marked = { readonly name: string; readonly offset: number };
+type Span = { readonly start: number; readonly end: number };
+type Found = { readonly at: number; readonly end: number };
+
+/** A name in code marks: `indent`, or the flag `--indent`, which is the name indent. */
+const CODE_SPAN = /`-{0,2}([A-Za-z_$][\w$-]*)`/gu;
+const FIRST_CELL_NAME = /`-{0,2}([A-Za-z_$][\w$-]*)`|^\s*-{0,2}([A-Za-z_$][\w$-]*)\s*$/u;
+const TABLE_ROW = /^[ \t]{0,3}\|/u;
+/** An anchor before a field name in a table cell: <a name="schemaExample"></a>example. */
+const HTML_TAG = /<[^<>]*>/gu;
+/** Where one statement ends: a sentence or a clause after a semicolon ("`indent` is deprecated; use `prefix`"). */
+const STATEMENT_END = /[.;。；!?！？]/u;
+/** A block that declares an API (interface Options { indent: number }) documents the name, and does not use it. */
+const DECLARATION_BLOCK = /^[ \t]*(?:export[ \t]+)?(?:interface|class|type[ \t]+[\w$]+[ \t]*=)/mu;
+
+const linesOf = (source: string): Line[] => {
+  const starts = [0, ...[...source.matchAll(/\n/gu)].map((match) => match.index + 1)];
+  return starts.map((start, index) => ({ text: source.slice(start, (starts[index + 1] ?? source.length + 1) - 1), start }));
+};
+
+const insideOne = (spans: readonly Span[], offset: number): boolean => spans.some((span) => offset >= span.start && offset < span.end);
+
+const occurrences = (text: string, words: readonly string[]): Found[] => {
+  const lower = text.toLowerCase();
+  return words.flatMap((word) => {
+    const found: Found[] = [];
+    for (let at = lower.indexOf(word); at !== -1; at = lower.indexOf(word, at + word.length)) found.push({ at, end: at + word.length });
+    return found;
+  });
+};
+
+/** The deprecation words in a text, less those inside a negated phrase ("is not deprecated"). */
+const wordsIn = (text: string, words: DeprecationWords): Found[] => {
+  const negated = occurrences(text, words.negated);
+  return occurrences(text, words.words).filter((word) => !negated.some((phrase) => phrase.at <= word.at && word.end <= phrase.end));
+};
+
+/** Characters between a code span and the word: "`indent` を非推奨" is nearer than "非推奨にし、`prefix`". */
+const gap = (span: Found, word: Found): number => (span.end <= word.at ? word.at - span.end : span.at - word.end);
+
+/** A table row's first-cell name, and whether its later cells call it deprecated. */
+const rowOf = (line: Line, words: DeprecationWords): (Marked & { readonly deprecated: boolean }) | undefined => {
+  const cells = line.text.split("|");
+  const first = cells.findIndex((cell) => cell.trim() !== "");
+  if (first === -1) return undefined;
+  const name = FIRST_CELL_NAME.exec((cells[first] ?? "").replace(HTML_TAG, ""));
+  const written = name?.[1] ?? name?.[2];
+  const deprecated = wordsIn(cells.slice(first + 1).join("|"), words).length > 0;
+  return written === undefined ? undefined : { name: written, offset: line.start + line.text.indexOf(written), deprecated };
+};
+
+const markedInRow = (line: Line, words: DeprecationWords): Marked[] => {
+  const row = rowOf(line, words);
+  return row?.deprecated === true ? [{ name: row.name, offset: row.offset }] : [];
+};
+
+/**
+ * Names a table also lists without calling them deprecated: one page may describe two fields of one name, of which only
+ * one is deprecated (an example field of one object, and of another), and an example cannot tell which it uses.
+ */
+const listedAlive = (lines: readonly Line[], words: DeprecationWords): Set<string> =>
+  new Set(
+    lines.flatMap((line) => (line.text.includes("|") ? [rowOf(line, words)] : [])).flatMap((row) => (row === undefined || row.deprecated ? [] : [row.name])),
+  );
+
+/** In a statement, the code name nearest to each deprecation word: "`indent` is deprecated", "Deprecated: `indent`". */
+const markedInProse = (line: Line, words: DeprecationWords): Marked[] =>
+  wordsIn(line.text, words).flatMap((word) => {
+    const { at } = word;
+    const from = Math.max(...[...line.text.slice(0, at).matchAll(new RegExp(STATEMENT_END, "gu"))].map((match) => match.index + 1), 0);
+    const after = line.text.slice(at).search(STATEMENT_END);
+    const to = after === -1 ? line.text.length : at + after;
+    const spans = [...line.text.slice(from, to).matchAll(CODE_SPAN)].map((match) => ({
+      name: match[1] ?? "",
+      at: from + match.index,
+      end: from + match.index + match[0].length,
+    }));
+    const nearest = spans.toSorted((left, right) => gap(left, word) - gap(right, word))[0];
+    return nearest === undefined ? [] : [{ name: nearest.name, offset: line.start + line.text.indexOf(nearest.name, nearest.at) }];
+  });
+
+const escape = (text: string): string => text.replace(/[.*+?^${}()|[\]\\]/gu, "\\$&");
+
+/** A name used as an option in code: a key ({ indent: 2 }, "indent": 2), an argument (indent=2) or a flag (--indent). */
+const usePattern = (name: string): RegExp =>
+  new RegExp(`(?:(?<![\\w$.-])["']?${escape(name)}["']?[ \\t]*(?::(?!:)|=(?!=))|(?<![\\w-])--${escape(name)}(?![\\w-]))`, "u");
+
+const usesIn = (line: Line, marks: ReadonlyMap<string, number>): DeprecatedUse[] =>
+  [...marks].flatMap(([name, marked]) => {
+    const match = usePattern(name).exec(line.text);
+    return match === null ? [] : [{ name, offset: line.start + match.index + match[0].indexOf(name), marked }];
+  });
+
+/**
+ * The code lines that use a name the document calls deprecated. A code line that says so itself (`indent: 2 // deprecated`)
+ * shows the old use on purpose and is left out, and so is a block that declares an API rather than using it.
+ */
+export const deprecatedUses = (source: string, words: DeprecationWords): DeprecatedUse[] => {
+  if (words.words.length === 0) return [];
+  const fences = codeFences(source);
+  const lines = linesOf(source);
+  const prose = lines.filter((line) => !insideOne(fences, line.start));
+  const alive = listedAlive(prose, words);
+  const marked = prose
+    .flatMap((line) => (TABLE_ROW.test(line.text) ? markedInRow(line, words) : markedInProse(line, words)))
+    .filter((mark) => !alive.has(mark.name));
+  const firstMark = new Map(marked.toReversed().map((mark) => [mark.name, mark.offset]));
+  const examples = fences.filter((fence) => !DECLARATION_BLOCK.test(source.slice(fence.start, fence.end)));
+  return lines.filter((line) => insideOne(examples, line.start) && wordsIn(line.text, words).length === 0).flatMap((line) => usesIn(line, firstMark));
+};

package/src/derived/number-word-counts.ts

@@ -0,0 +1,28 @@
+import type { Span } from "../plugin.ts";
+import { escapeRegExp } from "../orthography.ts";
+
+// A count written as a word before its unit ("twelve years", "Ten years"), which the tree does not read as a quantity.
+// Pure: the number words come in order from one (the lexicon count-number, the position is the number).
+
+export type WordCount = Span & { readonly amount: number };
+
+const alternation = (words: readonly string[]): string =>
+  words
+    .filter((word) => word !== "")
+    .map(escapeRegExp)
+    .join("|");
+
+/**
+ * Each number word standing as a word of its own, followed by one of the units after a space. A word joined to letters,
+ * digits or a dash on either side is the tail of a larger number (二十年's 十, "twenty-one years", "fortyten").
+ */
+export const numberWordCounts = (source: string, numberWords: readonly string[], units: readonly string[]): WordCount[] => {
+  const [words, unitWords] = [alternation(numberWords), alternation(units)];
+  if (words === "" || unitWords === "") return [];
+  const pattern = new RegExp(`(?<![\\p{L}\\p{N}\\p{Pd}])(${words})\\s+(?:${unitWords})(?![\\p{L}\\p{N}])`, "giu");
+  const lowered = numberWords.map((word) => word.toLowerCase());
+  return [...source.matchAll(pattern)].flatMap((match) => {
+    const position = lowered.indexOf((match[1] ?? "").toLowerCase());
+    return position < 0 ? [] : [{ start: match.index, end: match.index + match[0].length, amount: position + 1 }];
+  });
+};

package/src/derived/stays.ts

@@ -0,0 +1,145 @@
+import type { Span } from "../plugin.ts";
+import { bySentence } from "./sentence-buckets.ts";
+import { calendarDateOf, dateOf, shifted } from "./date-arithmetic.ts";
+import type { DatedValue } from "./durations.ts";
+import { cellsOf, tablesOf, type Cell } from "../facts/table-facts.ts";
+import { linesOf } from "../structure/lines.ts";
+import { withoutEdgeMarks } from "../facts/trim-marks.ts";
+
+/**
+ * 泊数。チェックインからチェックアウトまでの泊数は、二つの日付の差の日数（10月12日から10月14日までは2泊）。
+ * 一つの文に日付がちょうど二つと泊数がちょうど一つあるときと、チェックイン・チェックアウト・泊数の列がそろった表の行で比べる。
+ * 「2泊3日」のように泊数と日数を並べたものは、日数が泊数より一つ多いかを見る。Pure.
+ */
+export type Count = Span & { readonly amount: number };
+
+export type NightsMismatch = { readonly start: DatedValue; readonly end: DatedValue; readonly nights: Span; readonly expected: number };
+
+export type NightsDaysMismatch = { readonly nights: Count; readonly days: Count; readonly expected: number };
+
+export type StayColumnWords = {
+  readonly checkIn: readonly string[];
+  readonly checkOut: readonly string[];
+  readonly nights: readonly string[];
+  /** The words of a nights unit (泊, nights) a nights cell may write after its number. */
+  readonly units: readonly string[];
+};
+
+const DAY_MS = 86_400_000;
+
+type Between = { start: DatedValue; end: DatedValue; nights: number };
+
+const daysFrom = (from: Date, to: Date): number => Math.round((to.getTime() - from.getTime()) / DAY_MS);
+
+/** 月日だけの二つの日付は、近いほうの向き（12月30日と1月2日は年をまたいで3泊、10月14日と10月12日は2泊）。 */
+const nearerWay = (first: DatedValue, second: DatedValue, firstDay: Date, secondDay: Date): Between => {
+  const forward = daysFrom(firstDay, secondDay);
+  const ahead = forward < 0 ? daysFrom(firstDay, shifted(secondDay, 1, "year")) : forward;
+  const back = -forward < 0 ? daysFrom(secondDay, shifted(firstDay, 1, "year")) : -forward;
+  return back < ahead ? { start: second, end: first, nights: back } : { start: first, end: second, nights: ahead };
+};
+
+/** 二つの日付の間の泊数。年まで書いたら早いほうから。年の有る日付と無い日付の組は読まない。 */
+const nightsBetween = (first: DatedValue, second: DatedValue): Between | undefined => {
+  const [firstDate, secondDate] = [calendarDateOf(first.value), calendarDateOf(second.value)];
+  if (firstDate === undefined || secondDate === undefined || (firstDate.year === undefined) !== (secondDate.year === undefined)) return undefined;
+  const [firstDay, secondDay] = [dateOf(firstDate), dateOf(secondDate)];
+  if (firstDate.year === undefined) return nearerWay(first, second, firstDay, secondDay);
+  const forward = daysFrom(firstDay, secondDay);
+  return forward < 0 ? { start: second, end: first, nights: -forward } : { start: first, end: second, nights: forward };
+};
+
+const mismatchOf = (first: DatedValue, second: DatedValue, nights: Count): NightsMismatch | undefined => {
+  const between = nightsBetween(first, second);
+  if (between === undefined || between.nights <= 0 || between.nights === nights.amount) return undefined;
+  return { start: between.start, end: between.end, nights, expected: between.nights };
+};
+
+/** 一つの文の中の、二つの日付と泊数。 */
+export const stayNightsMismatches = (sentences: readonly Span[], dates: readonly DatedValue[], nights: readonly Count[]): NightsMismatch[] => {
+  const datesIn = bySentence(sentences, dates);
+  return [...bySentence(sentences, nights).entries()].flatMap(([index, inNights]) => {
+    const inDates = datesIn.get(index) ?? [];
+    const [first, second] = inDates;
+    const [count] = inNights;
+    if (inDates.length !== 2 || inNights.length !== 1 || first === undefined || second === undefined || count === undefined) return [];
+    const mismatch = mismatchOf(first, second, count);
+    return mismatch === undefined ? [] : [mismatch];
+  });
+};
+
+const normalizedGap = (text: string): string => text.normalize("NFKC").trim().toLowerCase().replace(/\s+/gu, " ");
+
+const joined = (source: string, left: Span, right: Span, joiners: readonly string[]): boolean => {
+  if (right.start < left.end) return false;
+  const gap = normalizedGap(source.slice(left.end, right.start));
+  return gap === "" || joiners.some((joiner) => normalizedGap(joiner) === gap);
+};
+
+/** 泊数と日数を並べたもの（2泊3日、2 days, 1 night）。泊数の直前か直後に、つなぎの語だけを挟んで日数がある。 */
+export const nightsDaysPairs = (
+  source: string,
+  nights: readonly Count[],
+  days: readonly Count[],
+  joiners: readonly string[],
+): { nights: Count; days: Count }[] =>
+  nights.flatMap((count) => {
+    const paired = days.find((day) => joined(source, count, day, joiners) || joined(source, day, count, joiners));
+    return paired === undefined ? [] : [{ nights: count, days: paired }];
+  });
+
+/** 日数が泊数より一つ多くない組。 */
+export const nightsDaysMismatches = (pairs: readonly { nights: Count; days: Count }[]): NightsDaysMismatch[] =>
+  pairs.flatMap((pair) => (pair.days.amount === pair.nights.amount + 1 ? [] : [{ ...pair, expected: pair.nights.amount + 1 }]));
+
+const keyOf = (text: string): string => withoutEdgeMarks(text).normalize("NFKC").toLowerCase().replace(/\s+/gu, " ");
+
+const columnOf = (header: readonly Cell[], names: readonly string[]): number => {
+  const wanted = new Set(names.map(keyOf));
+  return header.findIndex((cell) => wanted.has(keyOf(cell.text)));
+};
+
+const LEADING_DIGITS = /^\d+/u;
+
+/** 泊数の升の数（「2」「2泊」「2 nights」）。それ以外を書いた升は読まない。 */
+const nightsInCell = (cell: Cell, units: readonly string[]): Count | undefined => {
+  const key = keyOf(cell.text);
+  const digits = LEADING_DIGITS.exec(key)?.[0];
+  if (digits === undefined) return undefined;
+  const rest = key.slice(digits.length).trim();
+  if (rest !== "" && !units.some((unit) => keyOf(unit) === rest)) return undefined;
+  const offset = cell.text.length - cell.text.trimStart().length;
+  return { start: cell.start + offset, end: cell.end, amount: Number(digits) };
+};
+
+const onlyDateIn = (cell: Cell | undefined, dates: readonly DatedValue[]): DatedValue | undefined => {
+  if (cell === undefined) return undefined;
+  const inside = dates.filter((date) => date.start >= cell.start && date.end <= cell.end);
+  return inside.length === 1 ? inside[0] : undefined;
+};
+
+type StayColumns = { readonly checkIn: number; readonly checkOut: number; readonly nights: number };
+
+const rowMismatch = (cells: readonly Cell[], columns: StayColumns, dates: readonly DatedValue[], units: readonly string[]): NightsMismatch | undefined => {
+  const checkIn = onlyDateIn(cells[columns.checkIn], dates);
+  const checkOut = onlyDateIn(cells[columns.checkOut], dates);
+  const nightsCell = cells[columns.nights];
+  const nights = nightsCell === undefined ? undefined : nightsInCell(nightsCell, units);
+  return checkIn === undefined || checkOut === undefined || nights === undefined ? undefined : mismatchOf(checkIn, checkOut, nights);
+};
+
+/** チェックイン・チェックアウト・泊数の列がある表の、行ごとの泊数。 */
+export const tableNightsMismatches = (source: string, dates: readonly DatedValue[], words: StayColumnWords): NightsMismatch[] =>
+  tablesOf(linesOf(source)).flatMap(({ header, rows }) => {
+    const headerCells = cellsOf(header);
+    const columns = {
+      checkIn: columnOf(headerCells, words.checkIn),
+      checkOut: columnOf(headerCells, words.checkOut),
+      nights: columnOf(headerCells, words.nights),
+    };
+    if (Object.values(columns).some((index) => index < 0)) return [];
+    return rows.flatMap((row) => {
+      const mismatch = rowMismatch(cellsOf(row), columns, dates, words.units);
+      return mismatch === undefined ? [] : [mismatch];
+    });
+  });