chaffjs 0.19.0 → 0.21.0

package/src/detectors/email-letter.ts

@@ -0,0 +1,262 @@
+// A work email or letter's own parts: a greeting and a closing, a subject line, and an attachment the text mentions.
+// Pure. The words (お世話になっております, Dear, Best regards, 件名, attached) come from the language's email-greeting,
+// email-closing, email-subject-field, attachment-mention, attachment-not-mention and attachment-line lexicons. The
+// header and the signature are masked in doc.prose; a quoted reply is someone else's words and is left out.
+import { escapeRegExp } from "../orthography.ts";
+import { quoteAt } from "./structure-tree.ts";
+import type { Detector, Finding, ProseDocument, Span } from "../plugin.ts";
+
+/** A greeting comes before the body: past the subject, the addressee (株式会社… / 総務部 山田様) and the writer's name. */
+const GREETING_LINES = 5;
+/** A closing comes at the end, before the writer's name, title, company, address and contact lines, or an acceptance block. */
+const CLOSING_LINES = 10;
+
+/** One line of the writer's own text. breaks: a heading, the header or the signature stood between it and the line before. */
+export type BodyLine = { readonly start: number; readonly text: string; readonly breaks: boolean };
+
+const patternsOf = (doc: ProseDocument, id: string): string[] => (doc.lexicons[id] ?? []).map((entry) => entry.pattern);
+
+const LATIN_START = /^[A-Za-z]/u;
+const LETTER = /\p{L}/u;
+
+/**
+ * Whether a line holds one of the words. A Latin word opens the line (Dear, Best regards) and is followed by no letter,
+ * so "Hi" does not match "History"; a Japanese word may stand anywhere in it (いつもお世話になっております).
+ */
+export const lineHas = (line: string, words: readonly string[]): boolean => {
+  const trimmed = line.trim();
+  const folded = trimmed.toLowerCase();
+  return words.some((word) => {
+    if (!LATIN_START.test(word)) return trimmed.includes(word);
+    return folded.startsWith(word.toLowerCase()) && !LETTER.test(trimmed.charAt(word.length));
+  });
+};
+
+/**
+ * Whether each offset, asked in rising order, is inside one of the spans (sorted by start). One pass over the spans for
+ * the whole document: an archive with a quoted reply in every message stays linear.
+ */
+const insideCursor = (spans: readonly Span[]): ((offset: number) => boolean) => {
+  const sorted = spans.toSorted((left, right) => left.start - right.start);
+  const cursor = { next: 0, farthestEnd: -1 };
+  return (offset) => {
+    while (cursor.next < sorted.length && (sorted[cursor.next]?.start ?? Infinity) <= offset) {
+      cursor.farthestEnd = Math.max(cursor.farthestEnd, sorted[cursor.next]?.end ?? 0);
+      cursor.next += 1;
+    }
+    return offset < cursor.farthestEnd;
+  };
+};
+
+/**
+ * A line prose masks words at the head of: a header field whose value stays (Subject: …), not the writer's text. Emphasis
+ * marks are masked too, but they hold no letter.
+ */
+const isFieldLine = (sourceLine: string, proseLine: string): boolean => {
+  const sourceChars = [...sourceLine];
+  const kept = [...proseLine].findIndex((char, index) => char.trim() !== "" || sourceChars[index]?.trim() === "");
+  return LETTER.test(sourceChars.slice(0, kept === -1 ? sourceChars.length : kept).join(""));
+};
+
+type SourceLine = { readonly start: number; readonly text: string };
+
+const linesWithOffsets = (text: string): SourceLine[] => {
+  const lines: SourceLine[] = [];
+  text.split("\n").forEach((line) => lines.push({ start: (lines.at(-1)?.start ?? 0) + (lines.at(-1)?.text.length ?? -1) + 1, text: line }));
+  return lines;
+};
+
+/**
+ * The writer's own lines, in order: not blank once the header and signature are masked (prose), not a quoted reply, not a
+ * heading. A heading, or a line the source has and prose masks (a header, a signature), breaks the text into parts.
+ */
+export const bodyLines = (source: string, prose: string, replyQuotes: readonly Span[]): BodyLine[] => {
+  const quoted = insideCursor(replyQuotes);
+  const lines: BodyLine[] = [];
+  const state = { broken: false };
+  linesWithOffsets(prose).forEach(({ start, text }) => {
+    const trimmed = text.trimStart();
+    const sourceLine = source.slice(start, start + text.length);
+    const masked = trimmed === "" ? sourceLine.trim() !== "" : isFieldLine(sourceLine, text);
+    const inQuote = quoted(start);
+    if (masked || trimmed.startsWith("#")) state.broken = true;
+    else if (trimmed !== "" && !trimmed.startsWith(">") && !inQuote) {
+      lines.push({ start, text, breaks: state.broken });
+      state.broken = false;
+    }
+  });
+  return lines;
+};
+
+/** The lines split where a heading, a header or a signature breaks them: a letter and its appendix, or one message and the next. */
+const partsOf = (lines: readonly BodyLine[]): BodyLine[][] => {
+  const parts: BodyLine[][] = [];
+  lines.forEach((line) => {
+    const current = parts.at(-1);
+    if (line.breaks || current === undefined) parts.push([line]);
+    else current.push(line);
+  });
+  return parts;
+};
+
+export type FrameSlip = { readonly variant: "greeting" | "closing"; readonly offset: number };
+
+/** A missing greeting (none in the first lines) and a missing closing (none in the last lines of any part). */
+export const frameSlips = (lines: readonly BodyLine[], greetings: readonly string[], closings: readonly string[]): FrameSlip[] => {
+  const first = lines[0];
+  const last = lines.at(-1);
+  if (first === undefined || last === undefined) return [];
+  const greeted = lines.slice(0, GREETING_LINES).some((line) => lineHas(line.text, greetings));
+  const closed = partsOf(lines).some((part) => part.slice(-CLOSING_LINES).some((line) => lineHas(line.text, closings)));
+  return [
+    ...(greeted ? [] : [{ variant: "greeting" as const, offset: first.start }]),
+    ...(closed ? [] : [{ variant: "closing" as const, offset: last.start }]),
+  ];
+};
+
+export const emailFrame: Detector = (doc): Finding[] =>
+  frameSlips(bodyLines(doc.source, doc.prose ?? doc.source, doc.replyQuotes ?? []), patternsOf(doc, "email-greeting"), patternsOf(doc, "email-closing")).map(
+    (slip) => ({
+      rule: "email-greeting-closing",
+      severity: "info",
+      line: 0,
+      column: 0,
+      quote: quoteAt(doc.source, slip.offset),
+      values: { offset: slip.offset },
+      variant: slip.variant,
+    }),
+  );
+
+export type SubjectLine = { readonly start: number; readonly subject: string };
+
+/** A field's name is one word of at most this many characters. */
+const MAX_FIELD_NAME = 20;
+const BLANK = /\s/u;
+
+/** "Subject: Lunch" as its name and value; undefined for a line that is no field. */
+export const fieldOf = (line: string): { readonly name: string; readonly value: string } | undefined => {
+  const colon = [...line].findIndex((char) => char === ":" || char === "：");
+  if (colon <= 0) return undefined;
+  const chars = [...line];
+  const name = chars.slice(0, colon).join("").trimEnd();
+  return name.length > MAX_FIELD_NAME || BLANK.test(name)
+    ? undefined
+    : {
+        name,
+        value: chars
+          .slice(colon + 1)
+          .join("")
+          .trim(),
+      };
+};
+
+/** The subject lines (Subject: …, 件名：…), outside a quoted reply, with reply and forward marks (Re:, Fwd:) taken off the value. */
+export const subjectLines = (source: string, fields: readonly string[], replyMarks: readonly string[], replyQuotes: readonly Span[]): SubjectLine[] => {
+  const names = new Set(fields.map((field) => field.toLowerCase()));
+  const quoted = insideCursor(replyQuotes);
+  return linesWithOffsets(source).flatMap(({ start, text }): SubjectLine[] => {
+    const field = fieldOf(text);
+    const inQuote = quoted(start);
+    return field === undefined || inQuote || !names.has(field.name.toLowerCase()) ? [] : [{ start, subject: withoutReplyMarks(field.value, replyMarks) }];
+  });
+};
+
+/** "Re: Fwd: Lunch" is "Lunch": the marks a mail program adds are not what the writer wrote. Read piece by piece between colons. */
+export const withoutReplyMarks = (subject: string, marks: readonly string[]): string => {
+  const folded = new Set(marks.map((mark) => mark.toLowerCase()));
+  const pieces = subject.split(/[:：]/u);
+  const kept = pieces.findIndex((piece, index) => index === pieces.length - 1 || !folded.has(piece.trim().toLowerCase()));
+  const dropped = pieces.slice(0, kept).reduce((length, piece) => length + piece.length + 1, 0);
+  return subject.slice(dropped).trim();
+};
+
+export const emailSubject: Detector = (doc, options): Finding[] =>
+  subjectLines(doc.source, patternsOf(doc, "email-subject-field"), patternsOf(doc, "email-reply-mark"), doc.replyQuotes ?? []).flatMap((line): Finding[] => {
+    const length = [...line.subject].length;
+    if (length > 0 && length <= options.limit) return [];
+    return [
+      {
+        rule: "email-subject-length",
+        severity: "warning",
+        line: 0,
+        column: 0,
+        quote: quoteAt(doc.source, line.start),
+        values: { count: length, limit: options.limit, offset: line.start },
+        ...(length === 0 ? { variant: "empty" } : {}),
+      },
+    ];
+  });
+
+/** Where the text first mentions an attachment (添付します, attached), outside the phrases that only look like one (attached to). */
+export const firstMention = (lines: readonly BodyLine[], mentions: readonly string[], notMentions: readonly string[]): number | undefined => {
+  const folded = (text: string): string => text.toLowerCase();
+  const isMention = (text: string): boolean => {
+    const cleared = notMentions.reduce((rest, phrase) => rest.split(folded(phrase)).join(" "), folded(text));
+    return mentions.some((word) => cleared.includes(folded(word)));
+  };
+  const line = lines.find((candidate) => isMention(candidate.text));
+  return line === undefined ? undefined : line.start + line.text.length - line.text.trimStart().length;
+};
+
+/** The marks around a label line: a heading's #, emphasis, 【】. */
+const LINE_MARKS = new Set("#*_【】 \t\u3000");
+
+const withoutMarks = (line: string): string => {
+  const chars = [...line];
+  const first = chars.findIndex((char) => !LINE_MARKS.has(char));
+  return first === -1 ? "" : chars.slice(first, chars.findLastIndex((char) => !LINE_MARKS.has(char)) + 1).join("");
+};
+
+/** A line that labels what is attached: the label and a colon (Attachment: quote.pdf), or the label alone (### Enclosures, 【添付資料】). */
+const labelsAttachment = (line: string, labels: readonly string[]): boolean => {
+  const bare = withoutMarks(line).toLowerCase();
+  return labels.some((label) => {
+    const folded = label.toLowerCase();
+    return bare === folded || (bare.startsWith(folded) && /^\s*[:：]/u.test(bare.slice(folded.length)));
+  });
+};
+
+/** An attachment shown: a line that labels it, a file name with a listed extension, or the archive's note. text is the writer's own (no quoted reply). */
+export const showsAttachment = (text: string, labels: readonly string[], extensions: readonly string[], notes: readonly string[]): boolean => {
+  const labelled = text.split("\n").some((line) => labelsAttachment(line, labels));
+  const fileName = new RegExp(`[\\w\\-ぁ-んァ-ヶ一-龠]\\.(?:${extensions.map(escapeRegExp).join("|")})\\b`, "iu");
+  return labelled || (extensions.length > 0 && fileName.test(text)) || notes.some((note) => text.includes(note));
+};
+
+/** The source without someone else's words: a quoted reply's spans blanked and its "> " lines dropped. An old message's attachment is not this one's. */
+export const ownText = (source: string, replyQuotes: readonly Span[]): string => {
+  const pieces: string[] = [];
+  const cursor = { at: 0 };
+  replyQuotes
+    .toSorted((left, right) => left.start - right.start)
+    .forEach((span) => {
+      const from = Math.max(span.start, cursor.at);
+      pieces.push(source.slice(cursor.at, from), " ".repeat(Math.max(0, span.end - from)));
+      cursor.at = Math.max(cursor.at, span.end);
+    });
+  pieces.push(source.slice(cursor.at));
+  return pieces
+    .join("")
+    .split("\n")
+    .filter((line) => !line.trimStart().startsWith(">"))
+    .join("\n");
+};
+
+export const attachmentMissing: Detector = (doc): Finding[] => {
+  const lines = bodyLines(doc.source, doc.prose ?? doc.source, doc.replyQuotes ?? []);
+  const mention = firstMention(lines, patternsOf(doc, "attachment-mention"), patternsOf(doc, "attachment-not-mention"));
+  if (mention === undefined) return [];
+  const extensions = patternsOf(doc, "attachment-file-extension").map((extension) => extension.replace(/^\./u, ""));
+  if (showsAttachment(ownText(doc.source, doc.replyQuotes ?? []), patternsOf(doc, "attachment-line"), extensions, patternsOf(doc, "email-attachment-note")))
+    return [];
+  return [
+    {
+      rule: "attachment-not-attached",
+      severity: "warning",
+      line: 0,
+      column: 0,
+      quote: quoteAt(doc.source, mention),
+      values: { offset: mention },
+    },
+  ];
+};

package/src/detectors/emoji-heading.ts

@@ -0,0 +1,17 @@
+import type { Detector, Finding, MarkupHeading, ProseDocument } from "../plugin.ts";
+import { findingAt, writtenHeadings } from "./markup-finding.ts";
+
+/** A character drawn as an emoji: one that is by default, a pictograph asking for it (U+FE0F), or a keycap (1️⃣). ™, → and anything asking for text (U+FE0E) are text. */
+const EMOJI = new RegExp(String.raw`\p{Emoji_Presentation}(?!\u{FE0E})|\p{Extended_Pictographic}\u{FE0F}|[0-9#*]\u{FE0F}?\u{20E3}`, "u");
+
+export const hasEmoji = (text: string): boolean => EMOJI.test(text);
+
+/** The headings, in document order, whose words carry an emoji. */
+export const emojiHeadings = (doc: ProseDocument): MarkupHeading[] => writtenHeadings(doc).filter((heading) => hasEmoji(heading.text));
+
+/** A count, not a density: one decorated heading is a choice, a row of them is a template. */
+export const emojiHeading: Detector = (doc, options): Finding[] => {
+  const marked = emojiHeadings(doc);
+  if (marked.length === 0 || marked.length < options.limit) return [];
+  return marked.map((heading) => findingAt(doc, heading, { heading: heading.text, count: marked.length, limit: options.limit }));
+};

package/src/detectors/fact-consistency.ts

@@ -0,0 +1,54 @@
+import type { Detector, Finding, ProseDocument, StructureNode } from "../plugin.ts";
+import { nameSpans } from "../compare/proper-nouns.ts";
+import { factValues, type FactValue } from "../facts/fact-values.ts";
+import { labelledFacts, type AttributePhrase, type FactWords } from "../facts/labelled-facts.ts";
+import { tableFacts } from "../facts/table-facts.ts";
+import { scopedFacts, type ScopedFact } from "../facts/fact-scope.ts";
+import { scopeConflicts, summaryConflicts, type FactConflict } from "../facts/fact-conflicts.ts";
+import { quoteAt } from "./structure-tree.ts";
+
+const patternsOf = (doc: ProseDocument, id: string): string[] => (doc.lexicons[id] ?? []).map((entry) => entry.pattern);
+
+export const factWordsOf = (doc: ProseDocument): FactWords => ({
+  separators: patternsOf(doc, "fact-separator"),
+  valueEnds: patternsOf(doc, "fact-value-end"),
+  determiners: patternsOf(doc, "fact-label-drop"),
+  vague: patternsOf(doc, "fact-label-vague"),
+  attributes: (doc.lexicons["fact-attribute"] ?? []).map((entry): AttributePhrase => ({ pattern: entry.pattern, position: entry.position ?? "before" })),
+});
+
+const factsByDocument = new WeakMap<ProseDocument, readonly ScopedFact[]>();
+
+const readFacts = (doc: ProseDocument, tree: StructureNode): ScopedFact[] => {
+  const values = factValues(tree, doc.source, nameSpans(doc));
+  const facts = [...labelledFacts(doc.source, values, factWordsOf(doc)), ...tableFacts(doc.source, values)];
+  return scopedFacts(facts, tree, doc.source, patternsOf(doc, "summary-heading"));
+};
+
+/** 文書の名前付きの値。二つの rule が読むので、文書ごとに一度だけ読む。 */
+const factsOf = (doc: ProseDocument): readonly ScopedFact[] => {
+  const cached = factsByDocument.get(doc);
+  if (cached !== undefined) return cached;
+  const facts = doc.structure === undefined ? [] : readFacts(doc, doc.structure);
+  factsByDocument.set(doc, facts);
+  return facts;
+};
+
+const shown = (doc: ProseDocument, value: FactValue): string => doc.source.slice(value.start, value.end);
+
+const findingOf =
+  (rule: string, doc: ProseDocument) =>
+  (conflict: FactConflict): Finding => ({
+    rule,
+    severity: "warning",
+    line: 0,
+    column: 0,
+    quote: quoteAt(doc.source, conflict.fact.value.start),
+    values: { label: conflict.fact.label, value: shown(doc, conflict.fact.value), other: shown(doc, conflict.other), offset: conflict.fact.value.start },
+  });
+
+/** 同じ節で、同じ名前に二通りの値（締切：10月5日 と 締切：10月7日）。 */
+export const factConflict: Detector = (doc): Finding[] => scopeConflicts(factsOf(doc)).map(findingOf("fact-conflict", doc));
+
+/** 冒頭や要約の値が、本文の同じ名前の値と違う。 */
+export const summaryFactMismatch: Detector = (doc): Finding[] => summaryConflicts(factsOf(doc)).map(findingOf("summary-fact-mismatch", doc));

package/src/detectors/figure-order.ts

@@ -0,0 +1,81 @@
+// Figures and tables referred to out of numeric order, first referred to only a section after they appear, or never
+// referred to while their siblings are. Pure; the figure words come from the language's figure-label lexicon.
+import { figureMentions, type Mention as FigureMention } from "../figure-references.ts";
+import { labelWordsOf } from "./dangling-figure.ts";
+import { quoteAt } from "./structure-tree.ts";
+import type { Detector, Finding, ProseDocument } from "../plugin.ts";
+
+export type FigureOrderSlip =
+  | { readonly variant: "order"; readonly offset: number; readonly label: string; readonly after: string }
+  | { readonly variant: "late"; readonly offset: number; readonly label: string }
+  | { readonly variant: "unreferenced"; readonly offset: number; readonly label: string };
+
+/** The number as a tuple to compare (3 → [3], 3.2 → [3, 2]); undefined for letters and kanji numerals, which are not ordered here. */
+export const numberOf = (mention: FigureMention): readonly number[] | undefined => {
+  const number = mention.key.slice(mention.key.indexOf("\u0000") + 1).replace(/[-－]/gu, ".");
+  return /^\d+(?:\.\d+)*$/u.test(number) ? number.split(".").map(Number) : undefined;
+};
+
+const compare = (left: readonly number[], right: readonly number[]): number => {
+  const differs = left.findIndex((part, index) => part !== (right[index] ?? -1));
+  if (differs === -1) return left.length - right.length;
+  return (left[differs] ?? 0) - (right[differs] ?? -1);
+};
+
+/** The first mention of each figure, by its key, in document order. */
+const firstByKey = (mentions: readonly FigureMention[]): Map<string, FigureMention> =>
+  mentions.reduce((first, mention) => (first.has(mention.key) ? first : first.set(mention.key, mention)), new Map<string, FigureMention>());
+
+/** Each first reference whose number is lower than one of its kind first referred to earlier. */
+export const outOfOrder = (references: readonly FigureMention[]): FigureOrderSlip[] => {
+  const highest = new Map<string, { readonly number: readonly number[]; readonly written: string }>();
+  return [...firstByKey(references).values()].flatMap((mention): FigureOrderSlip[] => {
+    const number = numberOf(mention);
+    if (number === undefined) return [];
+    const before = highest.get(mention.kind);
+    if (before !== undefined && compare(number, before.number) < 0)
+      return [{ variant: "order", offset: mention.start, label: mention.written, after: before.written }];
+    highest.set(mention.kind, { number, written: mention.written });
+    return [];
+  });
+};
+
+/**
+ * Each captioned figure first referred to only after a heading that follows its caption, and each captioned figure never
+ * referred to, when the document refers to another of its kind (so it is the kind of document that refers to its figures).
+ */
+export const placementSlips = (
+  captions: readonly FigureMention[],
+  references: readonly FigureMention[],
+  headingStarts: readonly number[],
+  inTextKinds: ReadonlySet<string>,
+): FigureOrderSlip[] => {
+  const referredKinds = new Set(references.map((mention) => mention.kind));
+  const firstReference = firstByKey(references);
+  const firstCaptions = [...firstByKey(captions).values()].filter((mention) => inTextKinds.has(mention.kind));
+  return firstCaptions.flatMap((caption): FigureOrderSlip[] => {
+    const first = firstReference.get(caption.key);
+    if (first === undefined) return referredKinds.has(caption.kind) ? [{ variant: "unreferenced", offset: caption.start, label: caption.written }] : [];
+    const headingBetween = headingStarts.some((start) => caption.end < start && start < first.start);
+    return headingBetween ? [{ variant: "late", offset: first.start, label: first.written }] : [];
+  });
+};
+
+export const figureOrderSlips = (doc: ProseDocument): FigureOrderSlip[] => {
+  const { captions, references } = figureMentions(doc.source, doc.prose ?? doc.source, labelWordsOf(doc), doc.links);
+  const headingStarts = (doc.markup?.headings ?? []).map((heading) => heading.start);
+  const inTextKinds = new Set((doc.lexicons["figure-in-text"] ?? []).map((entry) => entry.pattern));
+  const slips = [...outOfOrder(references), ...placementSlips(captions, references, headingStarts, inTextKinds)];
+  return slips.toSorted((left, right) => left.offset - right.offset);
+};
+
+export const figureOrder: Detector = (doc): Finding[] =>
+  figureOrderSlips(doc).map((slip) => ({
+    rule: "figure-reference-order",
+    severity: "info",
+    line: 0,
+    column: 0,
+    quote: quoteAt(doc.source, slip.offset),
+    values: { label: slip.label, after: slip.variant === "order" ? slip.after : "", offset: slip.offset },
+    variant: slip.variant,
+  }));

package/src/detectors/heading-numbering.ts

@@ -0,0 +1,38 @@
+import type { Detector, Finding, LexiconEntry } from "../plugin.ts";
+import { findingAt, writtenHeadings } from "./markup-finding.ts";
+import { numberingMinorities, numberingStyleOf, UNNUMBERED, type NumberingMinority, type NumberLabel } from "../heading-numbering.ts";
+
+const labelsOf = (entries: readonly LexiconEntry[]): NumberLabel[] =>
+  entries.map((entry) => ({ word: entry.pattern, position: entry.position === "after" ? "after" : "before" }));
+
+/** 前後の見出しの語のあとに続いてよい区切り（「付録: 用語」「Appendix A」）。 */
+const AFTER_MATTER = /^[\s:：、]/u;
+
+/** 番号を付けない前後の見出し（目次・参考文献・付録 A）か。言葉全体が語彙表の語か、語彙表の語に区切りが続くとき。 */
+const isFrontOrBackMatter = (text: string, words: readonly string[]): boolean => {
+  const lowered = text.trim().toLowerCase();
+  return words.some((word) => lowered === word || (lowered.startsWith(word) && AFTER_MATTER.test(lowered.slice(word.length))));
+};
+
+const variantOf = (minority: NumberingMinority): string => {
+  if (minority.kind === "style") return "style";
+  return minority.majority === UNNUMBERED ? "numbered" : "unnumbered";
+};
+
+/**
+ * 兄弟の見出し（同じ親の下の同じ深さ）で、番号を付けたものと付けないもの、番号の書き方（1. と 1）と 第1章）が混ざっている所。
+ * 少ないほうを指す。どの語が番号を持つか（第1章、Chapter 1）は語彙表 heading-number-label、番号を付けない前後の見出しは unnumbered-heading が言う。
+ */
+export const headingNumbering: Detector = (doc): Finding[] => {
+  const headings = writtenHeadings(doc);
+  const labels = labelsOf(doc.lexicons["heading-number-label"] ?? []);
+  const matter = (doc.lexicons["unnumbered-heading"] ?? []).map((entry) => entry.pattern.toLowerCase());
+  const styles = headings.map((heading) => numberingStyleOf(heading.text, labels));
+  const read = headings.map((heading, at) => ({ depth: heading.depth, style: styles[at] ?? UNNUMBERED, skipped: isFrontOrBackMatter(heading.text, matter) }));
+  return numberingMinorities(read).flatMap((minority) => {
+    const heading = headings[minority.at];
+    if (heading === undefined) return [];
+    const values = { heading: heading.text.trim(), count: minority.count, example: headings[minority.example]?.text.trim() ?? "" };
+    return [findingAt(doc, heading, values, variantOf(minority))];
+  });
+};

package/src/detectors/impossible-date.ts

@@ -0,0 +1,141 @@
+import type { Detector, Finding, ProseDocument } from "../plugin.ts";
+import { escapeRegExp } from "../orthography.ts";
+import { quoteAround } from "./quote-around.ts";
+
+// 暦に無い日付（2月30日、4月31日、2023-02-29）。月の長さとうるう年は暦で決まるので、書き損じは機械で言える。
+// 月の名前は語彙表 month-name（初めの十二が一月から順に）、年月日の単位は date-time-unit（大きい順の初めの三つ）、元号は calendar-era が言う。
+
+/** 日付を読むのに要る語。months は一月から順の月の名前と略した名前、units は年・月・日の単位、eras は元号。 */
+export type CalendarWords = { readonly months: readonly string[]; readonly units: readonly string[]; readonly eras: readonly string[] };
+
+/** 暦に無い日付 1 つ。reason は、月が無い（month）、その月にその日が無い（day）、うるう年でない年の 2 月 29 日（leap）。 */
+export type ImpossibleDate = {
+  readonly offset: number;
+  readonly written: string;
+  readonly reason: "month" | "day" | "leap";
+  readonly month: number;
+  readonly days: number;
+};
+
+const DIGIT = "[0-9０-９]";
+const NOT_AFTER_NUMBER = "(?<![\\w/.\\-０-９])";
+/** 日付の後ろに来ない字。ただし日時（2023-02-29T10:00）の T は日付の終わり。 */
+const NOT_BEFORE_NUMBER = "(?:(?=T\\d)|(?![\\w/／\\-０-９]|\\.\\w))";
+const MONTHS_IN_YEAR = 12;
+const LONGEST_MONTH = 31;
+const FEBRUARY = 2;
+const LEAP_DAY = 29;
+const FIRST_CALENDAR_YEAR = 1000;
+const DAYS_IN_MONTH: readonly number[] = [31, 29, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31];
+const FULLWIDTH_OFFSET = 0xfee0;
+
+const halfWidth = (text: string): string => text.replace(/[０-９]/gu, (char) => String.fromCharCode(char.charCodeAt(0) - FULLWIDTH_OFFSET));
+
+const isLeapYear = (year: number): boolean => (year % 4 === 0 && year % 100 !== 0) || year % 400 === 0;
+
+/** 書き方ごとの読み方。strict は数の組が日付だと単位や名前で分かる形で、月が 1〜12 にない・日が 31 を超えるのも言う。 */
+type Shape = { readonly pattern: RegExp; readonly strict: boolean };
+
+/** 年の数。元号の年や 4 桁でない年は、うるう年を決められないので undefined。 */
+const yearOf = (groups: Readonly<Record<string, string | undefined>>): number | undefined => {
+  const written = groups["y"];
+  if (written === undefined || groups["era"] !== undefined) return undefined;
+  const year = Number(halfWidth(written));
+  return year >= FIRST_CALENDAR_YEAR ? year : undefined;
+};
+
+/** 一月から順の月の名前（初めの十二）で、略した名前（Sept、Feb.）が何月か。 */
+const monthNamed = (months: readonly string[], written: string): number => {
+  const name = written.replace(/\.$/u, "");
+  return months.slice(0, MONTHS_IN_YEAR).findIndex((full) => full.startsWith(name)) + 1;
+};
+
+type Groups = Readonly<Record<string, string | undefined>>;
+
+const monthOf = (groups: Groups, months: readonly string[]): number =>
+  groups["name"] === undefined ? Number(halfWidth(groups["m"] ?? "")) : monthNamed(months, groups["name"]);
+
+/** その月の日数。2 月は、年が分かってうるう年でなければ 28 日、分からなければ 29 日。 */
+const daysIn = (month: number, year: number | undefined): number =>
+  month === FEBRUARY && year !== undefined && !isLeapYear(year) ? LEAP_DAY - 1 : (DAYS_IN_MONTH[month - 1] ?? LONGEST_MONTH);
+
+const problemOf = (month: number, day: number, year: number | undefined, strict: boolean): ImpossibleDate["reason"] | undefined => {
+  if (month < 1 || month > MONTHS_IN_YEAR) return strict ? "month" : undefined;
+  if (day < 1 || day > LONGEST_MONTH) return strict ? "day" : undefined;
+  if (day <= daysIn(month, year)) return undefined;
+  return month === FEBRUARY && day === LEAP_DAY ? "leap" : "day";
+};
+
+const shapesOf = (words: CalendarWords): Shape[] => {
+  const [yearUnit, monthUnit, dayUnit] = words.units.map(escapeRegExp);
+  const names = words.months.map(escapeRegExp).join("|");
+  const eras = words.eras.map(escapeRegExp).join("|");
+  const era = eras === "" ? "" : `(?:(?<era>${eras})\\s?)?`;
+  const unitShape =
+    yearUnit === undefined || monthUnit === undefined || dayUnit === undefined
+      ? []
+      : [
+          {
+            pattern: new RegExp(
+              `(?<!${DIGIT})${era}(?:(?<y>${DIGIT}{1,4}|元)${yearUnit}\\s?)?(?<m>${DIGIT}{1,2})${monthUnit}\\s?(?<d>${DIGIT}{1,2})${dayUnit}`,
+              "gu",
+            ),
+            strict: true,
+          },
+        ];
+  const nameShapes =
+    names === ""
+      ? []
+      : [
+          { pattern: new RegExp(`\\b(?<name>${names})\\.?\\s(?<d>\\d{1,2})(?:st|nd|rd|th)?(?![\\d:])(?:,?\\s(?<y>\\d{4})\\b)?`, "gu"), strict: false },
+          { pattern: new RegExp(`(?<![\\d.,])\\b(?<d>\\d{1,2})(?:st|nd|rd|th)?\\s(?<name>${names})\\b\\.?(?:,?\\s(?<y>\\d{4})\\b)?`, "gu"), strict: false },
+        ];
+  return [
+    { pattern: new RegExp(`${NOT_AFTER_NUMBER}(?<y>\\d{4})-(?<m>\\d{1,2})-(?<d>\\d{1,2})${NOT_BEFORE_NUMBER}`, "gu"), strict: false },
+    { pattern: new RegExp(`${NOT_AFTER_NUMBER}(?<y>\\d{4})[/／](?<m>\\d{1,2})[/／](?<d>\\d{1,2})${NOT_BEFORE_NUMBER}`, "gu"), strict: false },
+    ...unitShape,
+    ...nameShapes,
+  ];
+};
+
+const impossibleIn = (text: string, shape: Shape, months: readonly string[]): ImpossibleDate[] =>
+  [...text.matchAll(shape.pattern)].flatMap((match) => {
+    const groups = match.groups ?? {};
+    const month = monthOf(groups, months);
+    const year = yearOf(groups);
+    const reason = problemOf(month, Number(halfWidth(groups["d"] ?? "")), year, shape.strict);
+    if (reason === undefined) return [];
+    return [{ offset: match.index, written: match[0].trim(), reason, month, days: daysIn(month, year) }];
+  });
+
+/** 文字列の中の、暦に無い日付。同じ所に二つの形が当たれば、先に見つけたほうだけ。 */
+export const impossibleDates = (text: string, words: CalendarWords): ImpossibleDate[] => {
+  const byOffset = new Map<number, ImpossibleDate>();
+  shapesOf(words)
+    .flatMap((shape) => impossibleIn(text, shape, words.months))
+    .forEach((found) => {
+      if (!byOffset.has(found.offset)) byOffset.set(found.offset, found);
+    });
+  return [...byOffset.values()].toSorted((left, right) => left.offset - right.offset);
+};
+
+const patternsOf = (doc: ProseDocument, id: string): string[] => (doc.lexicons[id] ?? []).map((entry) => entry.pattern);
+
+export const calendarWordsOf = (doc: ProseDocument): CalendarWords => ({
+  months: patternsOf(doc, "month-name"),
+  units: patternsOf(doc, "date-time-unit").slice(0, 3),
+  eras: patternsOf(doc, "calendar-era"),
+});
+
+export const impossibleDate: Detector = (doc): Finding[] => {
+  const text = doc.prose ?? doc.source;
+  return impossibleDates(text, calendarWordsOf(doc)).map((found) => ({
+    rule: "",
+    severity: "warning",
+    line: 0,
+    column: 0,
+    quote: quoteAround(text, found.offset, found.offset + found.written.length),
+    values: { written: found.written, month: found.month, days: found.days, offset: found.offset },
+    variant: found.reason,
+  }));
+};