chaffjs 0.17.0 → 0.18.0

package/src/commands/compare.ts

@@ -0,0 +1,80 @@
+import { loadAdapter } from "../adapter-load.ts";
+import { buildDocument, teamRules } from "../document.ts";
+import { readMarkdown } from "../markdown-read.ts";
+import { profileFor } from "../profile/for-file.ts";
+import { resolveGenre } from "../resolve-genre.ts";
+import { isMarkdownPath } from "../structure/markdown-path.ts";
+import { uiLanguageOf } from "../ui.ts";
+import { ATOM_KINDS, isAtomKind, type AtomKind } from "../compare/atom.ts";
+import { extractFacts } from "../compare/extract.ts";
+import { outcomeOf, type Allowed, type Compared } from "../compare/outcome.ts";
+import { renderCompact, renderFriendly, renderJson } from "../compare/render.ts";
+import { COMPARE_TEXT } from "../compare/text.ts";
+import { readSource, treeLanguage, type TreeContext } from "./tree.ts";
+
+/** Options whose value is the next argument. That value is not a file. */
+const VALUED: ReadonlySet<string> = new Set(["--language", "--genre", "--allow-dropped", "--allow-added"]);
+
+export const compareTargets = (argv: readonly string[]): string[] =>
+  argv.slice(1).filter((arg, index, all) => !arg.startsWith("--") && !VALUED.has(all[index - 1] ?? ""));
+
+/** Every value given to a repeatable option, split at commas: `--allow-dropped url,code --allow-dropped quote`. */
+const optionValues = (argv: readonly string[], name: string): string[] =>
+  argv.flatMap((arg, index) => (arg === name ? (argv[index + 1] ?? "").split(",") : [])).filter((value) => value !== "");
+
+type ParsedAllowed = { readonly allowed: Allowed } | { readonly unknown: string };
+
+/** The kinds allowed to be dropped and added, or the first word that is not a kind. */
+const parseAllowed = (argv: readonly string[]): ParsedAllowed => {
+  const dropped = optionValues(argv, "--allow-dropped");
+  const added = optionValues(argv, "--allow-added");
+  const unknown = [...dropped, ...added].find((kind) => !isAtomKind(kind));
+  if (unknown !== undefined) return { unknown };
+  const kinds = (values: readonly string[]): Set<AtomKind> => new Set(values.filter(isAtomKind));
+  return { allowed: { dropped: kinds(dropped), added: kinds(added) } };
+};
+
+type Read = Compared & { readonly language: string };
+
+/** One document's facts, read as lint reads it: its language, the team's names, and the parts of speech. */
+const readFacts = async (path: string, argv: readonly string[], context: TreeContext): Promise<Read | undefined> => {
+  const source = await readSource(path, context);
+  if (source === undefined) return undefined;
+  const language = treeLanguage(path, source, argv, context);
+  const adapter = await loadAdapter(language);
+  await adapter.prepare?.({ pos: true });
+  const genre = resolveGenre(path, source, context.config, context.flag(argv, "--genre")).genre;
+  const doc = buildDocument(path, source, adapter, teamRules(context.config), profileFor(context.config, path, source, language, genre));
+  const root = isMarkdownPath(path) ? readMarkdown(doc.source).root : undefined;
+  const extraction = extractFacts({ doc, root, structure: adapter.structure, names: context.config.names ?? [] });
+  return { path, extraction, language };
+};
+
+const renderFor = (argv: readonly string[]): typeof renderFriendly => {
+  if (argv.includes("--json")) return (outcome) => renderJson(outcome);
+  return argv.includes("--compact") ? renderCompact : renderFriendly;
+};
+
+/**
+ * Whether a rewrite kept every fact: numbers, dates, URLs, code, names, quotations, headings, references and footnotes.
+ * Compares; never rewrites. Ends with 1 when a fact was dropped or added, so a rewrite can be checked like a test.
+ */
+export const runCompare = async (targets: readonly string[], argv: readonly string[], context: TreeContext): Promise<number> => {
+  const host = COMPARE_TEXT[context.ui ?? "ja"];
+  const [beforePath, afterPath] = targets;
+  if (beforePath === undefined || afterPath === undefined || targets.length !== 2) {
+    console.error(host.usage);
+    return 1;
+  }
+  const parsed = parseAllowed(argv);
+  if ("unknown" in parsed) {
+    console.error(host.unknownKind(parsed.unknown, ATOM_KINDS.join(", ")));
+    return 1;
+  }
+  const before = await readFacts(beforePath, argv, context);
+  const after = before === undefined ? undefined : await readFacts(afterPath, argv, context);
+  if (before === undefined || after === undefined) return 1;
+  const outcome = outcomeOf(before, after, parsed.allowed);
+  console.log(renderFor(argv)(outcome, COMPARE_TEXT[uiLanguageOf(before.language)]));
+  return outcome.ok ? 0 : 1;
+};

package/src/commands/eval.ts

@@ -1,14 +1,14 @@
-import { plainSource } from "../plain-source.ts";
-import { readFile } from "node:fs/promises";
 import { loadAdapter } from "../adapter-load.ts";
 import { applyByPath } from "../config/by-path.ts";
 import { buildDocument, teamRules } from "../document.ts";
 import { guessLanguage } from "../detect.ts";
-import { collectTargets } from "../files.ts";
-import { loadRules } from "../rule-load.ts";
+import { collectTargets, readDocumentFile } from "../files.ts";
+import { rulesOf } from "../custom/load.ts";
 import { evaluate } from "../eval.ts";
 import { renderEval } from "../render/eval.ts";
-import { neededBy } from "../run.ts";
+import { neededBy, tokenFeaturesOf, wantsTags } from "../run.ts";
+import type { AdapterNeeds, RuleDefinition } from "../plugin.ts";
+import { optionLayersOf } from "../config/option-problems.ts";
 import type { Config } from "../config/load.ts";
 import { profileFor } from "../profile/for-file.ts";
 import { CLI_TEXT } from "../cli-text.ts";
@@ -28,6 +28,16 @@ const TEXT: Texts<{
   },
 };
 
+/**
+ * What the adapter must prepare. eval measures a rule whatever its level, so a rule named with --rule gets the tags and the
+ * token features it reads even when lint would not run it; without --rule, the tags are what lint would prepare.
+ */
+const evalNeeds = (rules: readonly RuleDefinition[], only: string | undefined, config: Config, genre: string, language: string): AdapterNeeds => {
+  const measured = rules.filter((rule) => only === undefined || rule.id === only);
+  const lint = neededBy(rules, config.rules, config.experimental, genre, language);
+  return { pos: lint.pos || (only !== undefined && measured.some(wantsTags)), features: tokenFeaturesOf(measured) };
+};
+
 export type Context = {
   readonly config: Config;
   readonly resolveGenre: (path: string, source: string, config: Config) => { genre: string; from: string };
@@ -51,11 +61,11 @@ export const runEval = async (targets: readonly string[], argv: readonly string[
   const only = flag(argv, "--rule");
   const docs = await Promise.all(
     paths.map(async (path) => {
-      const source = plainSource(await readFile(path, "utf8"));
+      const source = await readDocumentFile(path);
       const language = applyByPath(config.byPath, config.baseDir, path).language ?? config.language ?? guessLanguage(source).language;
       const adapter = await loadAdapter(language);
       const { genre } = resolveGenre(path, source, config);
-      await adapter.prepare?.(neededBy(loadRules(language), config.rules, config.experimental, genre, language));
+      await adapter.prepare?.(evalNeeds(rulesOf(language, config), only, config, genre, language));
       return { doc: buildDocument(path, source, adapter, teamRules(config), profileFor(config, path, source, language, genre)), language, genre };
     }),
   );
@@ -67,7 +77,7 @@ export const runEval = async (targets: readonly string[], argv: readonly string[
     console.error(TEXT[ui].mixed([...mixed].join(", ")));
     return 1;
   }
-  const rules = loadRules(language).filter((rule) => only === undefined || rule.id === only);
+  const rules = rulesOf(language, config).filter((rule) => only === undefined || rule.id === only);
   if (rules.length === 0) {
     console.error(TEXT[ui].unknownRule(only ?? CLI_TEXT[ui].unnamed));
     return 1;
@@ -80,6 +90,7 @@ export const runEval = async (targets: readonly string[], argv: readonly string[
         genre,
         language,
         config.limits,
+        optionLayersOf(config),
       ),
       docs.length,
       paths.length,

package/src/commands/feedback.ts

@@ -1,4 +1,4 @@
-import { plainSource } from "../plain-source.ts";
+import { readDocumentFile } from "../files.ts";
 import { existsSync, readFileSync, writeFileSync } from "node:fs";
 import { basename, join } from "node:path";
 import type { Finding, RuleDefinition } from "../plugin.ts";
@@ -86,15 +86,21 @@ const chosen = (findings: readonly Finding[], rule: string | undefined, line: nu
 const describe = (findings: readonly Finding[], none: string): string =>
   findings.length === 0 ? none : findings.map((finding) => `  ${String(finding.line)}  ${finding.rule}`).join("\n");
 
-/** Pure: the reported rules' own lines of chaff.yaml's rules (a numeric limit wins over a level), or undefined when none is set. */
-export const settingsOf = (config: Pick<Config, "rules" | "limits">, ruleIds: readonly string[]): string | undefined => {
+/**
+ * Pure: the reported rules' own lines of chaff.yaml's rules (a numeric limit wins over a level), under the style chaff.yaml
+ * names, or undefined when none is set. A level the style decided is the style's, so only the style line reproduces it.
+ */
+export const settingsOf = (config: Pick<Config, "rules" | "limits" | "applied">, ruleIds: readonly string[]): string | undefined => {
+  const fromStyle = new Set(config.applied?.levelsFrom ?? []);
   const lines = [...new Set(ruleIds)].flatMap((id) => {
     const limit = config.limits[id];
     const level = config.rules[id];
     if (limit !== undefined) return [`  ${id}: ${String(limit)}`];
-    return level === undefined ? [] : [`  ${id}: ${level}`];
+    return level === undefined || fromStyle.has(id) ? [] : [`  ${id}: ${level}`];
   });
-  return lines.length === 0 ? undefined : ["rules:", ...lines].join("\n");
+  const style = config.applied === undefined ? [] : [`style: ${config.applied.style}`];
+  const rules = lines.length === 0 ? [] : ["rules:", ...lines];
+  return style.length + rules.length === 0 ? undefined : [...style, ...rules].join("\n");
 };
 
 /** Pure: a title on one line, no longer than GitHub shows. */
@@ -179,7 +185,7 @@ export const runFeedback = async (targets: readonly string[], argv: readonly str
     console.error(text.notFound(request.path));
     return 1;
   }
-  const source = plainSource(readFileSync(request.path, "utf8"));
+  const source = await readDocumentFile(request.path);
   const checked = await context.check(request.path);
   const picked = pick(request, checked, source.split("\n").length, context.ui);
   if ("error" in picked) {

package/src/commands/test.ts

@@ -1,11 +1,9 @@
-import { plainSource } from "../plain-source.ts";
-import { readFile } from "node:fs/promises";
 import { join } from "node:path";
 import { loadAdapter } from "../adapter-load.ts";
 import { applyByPath } from "../config/by-path.ts";
 import { buildDocument, teamRules } from "../document.ts";
 import { guessLanguage } from "../detect.ts";
-import { collectTargets } from "../files.ts";
+import { collectTargets, readDocumentFile } from "../files.ts";
 import { loadRules } from "../rule-load.ts";
 import { counted } from "../render/plural.ts";
 import { CHECKS_FILE, loadChecks, type UserCheck } from "../checks.ts";
@@ -87,7 +85,7 @@ type SemanticDocument = { readonly doc: ProseDocument; readonly rules: RuleDefin
 
 /** 意味を読む検査に渡す文書。絞り込みが品詞を使うなら、解析器を読み込んでから作る。dry-run も本番も同じ文書を見る。 */
 const semanticDocument = async (path: string, config: Config, resolveGenre: TestContext["resolveGenre"]): Promise<SemanticDocument> => {
-  const source = plainSource(await readFile(path, "utf8"));
+  const source = await readDocumentFile(path);
   const language = applyByPath(config.byPath, config.baseDir, path).language ?? config.language ?? guessLanguage(source).language;
   const adapter = await loadAdapter(language);
   const { genre } = resolveGenre(path, source, config);

package/src/commands/tree.ts

@@ -1,5 +1,4 @@
-import { plainSource } from "../plain-source.ts";
-import { readFile } from "node:fs/promises";
+import { readDocumentFile } from "../files.ts";
 import { loadAdapter } from "../adapter-load.ts";
 import { applyByPath } from "../config/by-path.ts";
 import { profileFor } from "../profile/for-file.ts";
@@ -8,6 +7,7 @@ import { guessLanguage } from "../detect.ts";
 import { isMarkdownPath } from "../structure/markdown-path.ts";
 import { buildStructure } from "../structure/of.ts";
 import { toSexp } from "../structure/sexp.ts";
+import { foldPostOrder } from "../tree-walk.ts";
 import type { Config } from "../config/load.ts";
 import type { StructureNode } from "../plugin.ts";
 import type { Texts, UiLanguage } from "../ui.ts";
@@ -23,16 +23,19 @@ const TEXT: Texts<{
   readonly usage: string;
   readonly unreadable: (path: string, why: string) => string;
   readonly noStructure: (path: string, language: string) => string;
+  readonly tooDeepForJson: (path: string, depth: number) => string;
 }> = {
   ja: {
     usage: "使い方: chaff tree <file>... [--format sexp|json] [--language ja|en|…]",
     unreadable: (path, why) => `${path} を読めませんでした: ${why}`,
     noStructure: (path, language) => `${path}: 言語 ${language} のパッケージは文書の構造を読めません（structure がありません）`,
+    tooDeepForJson: (path, depth) => `${path}: 木が ${String(depth)} 段と深すぎて、字下げした JSON に書き出せません（--format sexp なら書けることがあります）`,
   },
   en: {
     usage: "usage: chaff tree <file>... [--format sexp|json] [--language ja|en|…]",
     unreadable: (path, why) => `Could not read ${path}: ${why}`,
     noStructure: (path, language) => `${path}: the ${language} package cannot read a document's structure (it has no structure)`,
+    tooDeepForJson: (path, depth) => `${path}: the tree is ${String(depth)} levels deep, too deep to write as indented JSON (--format sexp may still write it)`,
   },
 };
 
@@ -48,7 +51,7 @@ export const treeTargets = (argv: readonly string[]): string[] =>
 
 export const readSource = async (path: string, context: TreeContext): Promise<string | undefined> => {
   try {
-    return plainSource(await readFile(path, "utf8"));
+    return await readDocumentFile(path);
   } catch (err) {
     console.error(treeText(context).unreadable(path, err instanceof Error ? err.message : String(err)));
     return undefined;
@@ -83,13 +86,46 @@ export const readTree = async (path: string, argv: readonly string[], context: T
   // 日本語は形態素で数量と日付を読む。解析器が無ければ単位の表で読むので、木は作れる。
   await adapter.prepare?.({ pos: true });
   const profile = profileFor(context.config, path, source, language, resolveGenre(path, source, context.config, context.flag(argv, "--genre")).genre);
-  return { source, tree: buildStructure({ path, source, language, markdown: isMarkdownPath(path), profile }, adapter.structure) };
+  return { source, tree: buildStructure({ path, source, language, markdown: isMarkdownPath(path), profile, lexicons: adapter.lexicons }, adapter.structure) };
+};
+
+/**
+ * 字下げした JSON。書けないほど深い木（言語パッケージの番号が何千段も入れ子になる）なら undefined。
+ * JSON.stringify は 1 段ごとに呼び直すのでスタックが尽き、字下げの合計は深さの二乗で増えて文字列の上限を超える。どちらも RangeError になる。
+ */
+export const treeJson = (tree: StructureNode): string | undefined => {
+  try {
+    return JSON.stringify(tree, null, 2);
+  } catch (err) {
+    if (err instanceof RangeError) return undefined;
+    throw err;
+  }
+};
+
+/** 根を 1 段目と数えた、いちばん深い葉までの段の数。 */
+export const depthOf = (tree: StructureNode): number =>
+  foldPostOrder(
+    tree,
+    (node) => node.children,
+    (_node, depths: readonly number[]) => depths.reduce((deepest, depth) => Math.max(deepest, depth), 0) + 1,
+  );
+
+/** 出力する文字列。JSON に書けないほど深い木なら、その深さ。 */
+export const renderTree = (tree: StructureNode, format: string | undefined): { readonly text: string } | { readonly tooDeep: number } => {
+  if (format !== "json") return { text: toSexp(tree) };
+  const json = treeJson(tree);
+  return json === undefined ? { tooDeep: depthOf(tree) } : { text: json };
 };
 
 const printTree = async (path: string, argv: readonly string[], context: TreeContext): Promise<boolean> => {
   const read = await readTree(path, argv, context);
   if (read === undefined) return false;
-  console.log(context.flag(argv, "--format") === "json" ? JSON.stringify(read.tree, null, 2) : toSexp(read.tree));
+  const rendered = renderTree(read.tree, context.flag(argv, "--format"));
+  if ("tooDeep" in rendered) {
+    console.error(treeText(context).tooDeepForJson(path, rendered.tooDeep));
+    return false;
+  }
+  console.log(rendered.text);
   return true;
 };
 

package/src/compare/atom.ts

@@ -0,0 +1,33 @@
+/** The kinds of fact `chaff compare` reads, in the order it reports them. */
+export const ATOM_KINDS = ["number", "date", "time", "url", "code", "name", "quote", "heading", "reference", "footnote"] as const;
+
+export type AtomKind = (typeof ATOM_KINDS)[number];
+
+export const isAtomKind = (value: string): value is AtomKind => ATOM_KINDS.some((kind) => kind === value);
+
+/**
+ * One fact as a document states it. Two atoms are the same fact when kind and key are equal; `text` is how it was
+ * written (1,000円 and 1000円 share a key). Position is only for showing where it is: comparing ignores it.
+ */
+export type Atom = {
+  readonly kind: AtomKind;
+  readonly key: string;
+  readonly text: string;
+  readonly line: number;
+  /** A number written with no unit the reader knows. It may stand for the same number written with a unit. */
+  readonly unitless?: boolean;
+  /** For a number: its value alone, without the unit. */
+  readonly value?: string;
+};
+
+/** A kind this document could not be fully read for, and why. Zero facts of it then does not mean none were there. */
+export type Unread = { readonly kind: AtomKind; readonly reason: UnreadReason };
+
+export type UnreadReason = "no-structure" | "no-dates" | "no-pos" | "plain-text";
+
+export type Extraction = {
+  readonly atoms: readonly Atom[];
+  readonly unread: readonly Unread[];
+  /** The prose spelled as name keys are, so the other document can ask whether a name is written here at all. */
+  readonly nameText: string;
+};

package/src/compare/clock-time.ts

@@ -0,0 +1,86 @@
+import type { Span } from "../plugin.ts";
+import { overlapsAny, spanIndex } from "./spans.ts";
+
+/** A time of day and where it is written. key is "HH:MM" on a 24-hour clock, with ":SS" when seconds are given. */
+export type ClockTime = Span & { readonly key: string };
+
+const DIGIT = "[0-9０-９]";
+const COLON = "[:：]";
+
+/** 10:30, 15:30:05, 3:30 p.m. */
+const COLON_TIME = new RegExp(
+  `(?<![0-9０-９])(${DIGIT}{1,2})${COLON}(${DIGIT}{2})(?:${COLON}(${DIGIT}{2}))?(?:\\s?([ap])\\.?m\\.?(?![a-z]))?(?![0-9０-９])`,
+  "giu",
+);
+
+/** 3 p.m., 11am. */
+const MERIDIEM_TIME = new RegExp(`(?<![0-9０-９:：])(${DIGIT}{1,2})\\s?([ap])\\.?m\\.?(?![a-z])`, "giu");
+
+/** 午後3時30分, 10時半, 9時. */
+const JAPANESE_TIME = new RegExp(`(午前|午後)?(${DIGIT}{1,2})時(?:(${DIGIT}{1,2})分|(半))?`, "gu");
+
+const LAST_HOUR = 24;
+const MINUTES_PER_HOUR = 60;
+const HALF_HOUR = 30;
+const NOON = 12;
+const PADDED = 2;
+
+type Reading = { readonly hour: number; readonly minute: number; readonly second?: number | undefined; readonly afternoon?: boolean | undefined };
+
+const hourOf = (hour: number, afternoon: boolean | undefined): number => {
+  if (afternoon === undefined) return hour;
+  if (afternoon) return hour < NOON ? hour + NOON : hour;
+  return hour === NOON ? 0 : hour;
+};
+
+const padded = (value: number): string => String(value).padStart(PADDED, "0");
+
+/** Whether the hour is one on its clock: up to 12 with a.m. or p.m., else up to 23, and 24 only as 24:00. */
+const isHour = ({ hour, minute, second, afternoon }: Reading): boolean => {
+  if (afternoon !== undefined) return hour <= NOON;
+  return hour < LAST_HOUR || (hour === LAST_HOUR && minute === 0 && (second ?? 0) === 0);
+};
+
+/** The key, or undefined when the numbers are not a time of day (24:30, 13 p.m., 3:75). */
+const keyOf = (reading: Reading): string | undefined => {
+  const { hour, minute, second } = reading;
+  if (!isHour(reading) || minute >= MINUTES_PER_HOUR || (second ?? 0) >= MINUTES_PER_HOUR) return undefined;
+  const parts = [hourOf(hour, reading.afternoon), minute, ...(second === undefined ? [] : [second])];
+  return parts.map(padded).join(":");
+};
+
+const numberOf = (digits: string | undefined): number | undefined => (digits === undefined ? undefined : Number(digits.normalize("NFKC")));
+
+const meridiem = (letter: string | undefined): boolean | undefined => (letter === undefined ? undefined : letter.toLowerCase() === "p");
+
+const japaneseMeridiem = (word: string | undefined): boolean | undefined => (word === undefined ? undefined : word === "午後");
+
+type Reader = { readonly pattern: RegExp; readonly read: (match: RegExpMatchArray) => Reading };
+
+const READERS: readonly Reader[] = [
+  {
+    pattern: JAPANESE_TIME,
+    read: (match) => ({
+      hour: numberOf(match[2]) ?? 0,
+      minute: match[4] === undefined ? (numberOf(match[3]) ?? 0) : HALF_HOUR,
+      afternoon: japaneseMeridiem(match[1]),
+    }),
+  },
+  {
+    pattern: COLON_TIME,
+    read: (match) => ({ hour: numberOf(match[1]) ?? 0, minute: numberOf(match[2]) ?? 0, second: numberOf(match[3]), afternoon: meridiem(match[4]) }),
+  },
+  { pattern: MERIDIEM_TIME, read: (match) => ({ hour: numberOf(match[1]) ?? 0, minute: 0, afternoon: meridiem(match[2]) }) },
+];
+
+/** Times of day in text, in the order written. Where two readings overlap, the earlier reader in READERS wins. */
+export const clockTimes = (text: string): ClockTime[] =>
+  READERS.reduce<ClockTime[]>((found, reader) => {
+    const taken = spanIndex(found);
+    const fresh = [...text.matchAll(reader.pattern)].flatMap((match) => {
+      const key = keyOf(reader.read(match));
+      const span = { start: match.index, end: match.index + match[0].length };
+      return key === undefined || overlapsAny(taken, span) ? [] : [{ ...span, key }];
+    });
+    return [...found, ...fresh];
+  }, []).toSorted((left, right) => left.start - right.start);

package/src/compare/extract.ts

@@ -0,0 +1,78 @@
+import { lineStarts, placeOf } from "../position.ts";
+import type { MarkdownNode } from "../markdown-node.ts";
+import type { ProseDocument, Span, StructurePatterns } from "../plugin.ts";
+import type { Atom, Extraction, Unread } from "./atom.ts";
+import { clockTimes } from "./clock-time.ts";
+import { factTextOf, type FactText } from "./fact-text.ts";
+import { codeAtoms, headingAtoms, urlAtoms } from "./markup-atoms.ts";
+import { nameKey, properNouns, readsProperNouns } from "./proper-nouns.ts";
+import { spanIndex } from "./spans.ts";
+import { atomsOf, bareNumbers, footnotes, numericDates, quotations, teamNames, type TextInput } from "./text-atoms.ts";
+import { treeFacts, type TreeFacts } from "./tree-atoms.ts";
+
+export type ExtractInput = {
+  readonly doc: ProseDocument;
+  /** The Markdown tree, or undefined for a plain-text document. */
+  readonly root: MarkdownNode | undefined;
+  /** The language package's structure reader: numbers with units, dates and references come from it. */
+  readonly structure: StructurePatterns | undefined;
+  /** chaff.yaml's `names:`. */
+  readonly names: readonly string[];
+};
+
+const NO_TREE_FACTS: TreeFacts = { atoms: [], taken: [] };
+
+const STRUCTURE_KINDS = ["number", "date", "reference"] as const;
+
+/** The kinds this document could not be fully read for. Each says why, so "nothing dropped" is never "nothing read". */
+const unreadOf = (input: ExtractInput): Unread[] => [
+  ...(input.structure === undefined ? STRUCTURE_KINDS.map((kind): Unread => ({ kind, reason: "no-structure" })) : []),
+  ...(input.structure !== undefined && input.structure.dates === undefined ? [{ kind: "date", reason: "no-dates" } as const] : []),
+  ...(readsProperNouns(input.doc) ? [] : [{ kind: "name", reason: "no-pos" } as const]),
+  ...(input.root === undefined ? [{ kind: "code", reason: "plain-text" } as const] : []),
+];
+
+type Readers = { readonly facts: FactText; readonly text: TextInput; readonly lineOf: (offset: number) => number };
+
+/** Facts read from the prose itself. Each reader skips what an earlier one took, so one figure is one fact. */
+const proseFacts = (input: ExtractInput, readers: Readers, tree: TreeFacts, times: readonly Span[]): Atom[] => {
+  const { facts, text, lineOf } = readers;
+  const marks = footnotes(facts.codeless);
+  const dates = numericDates(facts.text, spanIndex([...tree.taken, ...times]));
+  const names = teamNames(facts.text, input.names);
+  const taken = [...tree.taken, ...times, ...marks, ...dates];
+  return [
+    ...atomsOf(dates, "date", text),
+    ...atomsOf(marks, "footnote", text),
+    ...atomsOf(names, "name", text),
+    ...bareNumbers(text, spanIndex([...taken, ...facts.listMarkers])),
+    ...quotations(text),
+    // "April" in a date and "Section" in a reference are read with them, not again as names.
+    ...properNouns({ doc: input.doc, taken: spanIndex([...taken, ...names, ...facts.blanked]), lineOf }),
+  ];
+};
+
+const structureFacts = (input: ExtractInput, facts: FactText, times: readonly Span[], lineOf: (offset: number) => number): TreeFacts => {
+  const tree = input.doc.structure;
+  return tree === undefined ? NO_TREE_FACTS : treeFacts({ tree, source: input.doc.source, blanked: facts.blanked, times: spanIndex(times), lineOf });
+};
+
+/** Every fact atom of one document, and the kinds it could not be fully read for. */
+export const extractFacts = (input: ExtractInput): Extraction => {
+  const source = input.doc.source;
+  const starts = lineStarts(source);
+  const lineOf = (offset: number): number => placeOf(starts, offset).line;
+  const facts = factTextOf(source, input.root);
+  const text = { text: facts.text, source, lineOf };
+  const times = clockTimes(facts.text);
+  const tree = structureFacts(input, facts, times, lineOf);
+  const atoms = [
+    ...tree.atoms,
+    ...atomsOf(times, "time", text),
+    ...proseFacts(input, { facts, text, lineOf }, tree, times),
+    ...urlAtoms(input.doc.markup, facts.bareUrls, lineOf),
+    ...(input.root === undefined ? [] : codeAtoms(input.root, lineOf)),
+    ...headingAtoms(input.doc.markup, lineOf),
+  ];
+  return { atoms, unread: unreadOf(input), nameText: nameKey(facts.text) };
+};

package/src/compare/fact-text.ts

@@ -0,0 +1,68 @@
+import { bareUrls, type BareUrl } from "../bare-url.ts";
+import { linkChrome } from "../document.ts";
+import { opaqueSpans } from "../markdown-read.ts";
+import { spanOf, type MarkdownNode } from "../markdown-node.ts";
+import { maskSpans } from "../mask.ts";
+import { eachPreOrder } from "../tree-walk.ts";
+import type { Span } from "../plugin.ts";
+import { overlapsAny, spanIndex, type SpanIndex } from "./spans.ts";
+
+/**
+ * The document as the facts in prose are read from it: the same length as the source, with what is not prose
+ * blanked. Code, HTML and front matter are compared as code or not at all; a link's target and a bare URL are
+ * compared as URLs. Blanking them keeps the digits of a URL from being read again as numbers.
+ */
+export type FactText = {
+  /** The source with code and URLs blanked. Offsets are the source's. */
+  readonly text: string;
+  /** The source with only code blanked: footnote marks are link syntax, so they are read before links are blanked. */
+  readonly codeless: string;
+  /** Where the blanked parts are. A fact the structure reader found there is not a fact of the prose. */
+  readonly blanked: SpanIndex;
+  readonly bareUrls: readonly BareUrl[];
+  /** Ordered-list markers (`1.`): they number the list, they do not state a number. */
+  readonly listMarkers: readonly Span[];
+};
+
+/** Nodes whose whole span is not read as prose: a link reference definition's URL is read as a URL, an image is not text. */
+const WHOLE = new Set(["definition", "image", "imageReference"]);
+const LINKS = new Set(["link", "linkReference"]);
+
+/** `1.` or `1)` at the start of an ordered list item. */
+const ORDERED_MARKER = /^\s*\d{1,9}[.)]/u;
+
+const listMarker = (node: MarkdownNode, source: string): Span[] => {
+  const span = spanOf(node);
+  if (node.type !== "listItem" || span === undefined) return [];
+  const marker = ORDERED_MARKER.exec(source.slice(span.start, span.end));
+  return marker === null ? [] : [{ start: span.start, end: span.start + marker[0].length }];
+};
+
+type MarkdownParts = { readonly chrome: Span[]; readonly linkSpans: Span[]; readonly listMarkers: Span[] };
+
+const markdownParts = (root: MarkdownNode, source: string): MarkdownParts => {
+  const parts: MarkdownParts = { chrome: [], linkSpans: [], listMarkers: [] };
+  eachPreOrder(root, (node) => {
+    const span = spanOf(node);
+    if (span === undefined) return;
+    if (WHOLE.has(node.type)) parts.chrome.push(span);
+    if (LINKS.has(node.type)) {
+      parts.chrome.push(...linkChrome(node));
+      parts.linkSpans.push(span);
+    }
+    parts.listMarkers.push(...listMarker(node, source));
+  });
+  return parts;
+};
+
+/** root is the Markdown tree, or undefined for a plain-text document, where nothing is code and URLs are bare. */
+export const factTextOf = (source: string, root: MarkdownNode | undefined): FactText => {
+  const code = root === undefined ? [] : opaqueSpans(root);
+  const codeless = maskSpans(source, code);
+  const parts = root === undefined ? { chrome: [], linkSpans: [], listMarkers: [] } : markdownParts(root, source);
+  const insideLinks = spanIndex(parts.linkSpans);
+  // An autolink's text is its URL: it is read once, as the link.
+  const bare = bareUrls(codeless).filter((url) => !overlapsAny(insideLinks, url));
+  const blanked = [...code, ...parts.chrome, ...bare];
+  return { text: maskSpans(source, blanked), codeless, blanked: spanIndex(blanked), bareUrls: bare, listMarkers: parts.listMarkers };
+};

package/src/compare/markup-atoms.ts

@@ -0,0 +1,29 @@
+import { spanOf, type MarkdownNode } from "../markdown-node.ts";
+import { eachPreOrder } from "../tree-walk.ts";
+import type { Markup } from "../plugin.ts";
+import type { BareUrl } from "../bare-url.ts";
+import type { Atom } from "./atom.ts";
+
+const CODE = new Set(["code", "inlineCode"]);
+
+/** Inline code and code blocks, compared as written: the key is the code itself, without its backticks or fences. */
+export const codeAtoms = (root: MarkdownNode, lineOf: (offset: number) => number): Atom[] => {
+  const found: Atom[] = [];
+  eachPreOrder(root, (node) => {
+    const span = spanOf(node);
+    if (!CODE.has(node.type) || span === undefined) return;
+    const code = node.value ?? "";
+    found.push({ kind: "code", key: code, text: code, line: lineOf(span.start) });
+  });
+  return found;
+};
+
+/** Where links point (a Markdown link, an autolink, a used reference definition) and URLs written bare. */
+export const urlAtoms = (markup: Markup | undefined, bare: readonly BareUrl[], lineOf: (offset: number) => number): Atom[] => [
+  ...(markup?.links ?? []).map((link) => ({ kind: "url" as const, key: link.destination, text: link.destination, line: lineOf(link.start) })),
+  ...bare.map((url) => ({ kind: "url" as const, key: url.url, text: url.url, line: lineOf(url.start) })),
+];
+
+/** Markdown headings as structure: the key is the level, so a reworded heading is the same heading in another form. */
+export const headingAtoms = (markup: Markup | undefined, lineOf: (offset: number) => number): Atom[] =>
+  (markup?.headings ?? []).map((heading) => ({ kind: "heading", key: "#".repeat(heading.depth), text: heading.text, line: lineOf(heading.start) }));