chaffjs 0.16.0 → 0.17.0

package/src/stray-space.ts

@@ -0,0 +1,81 @@
+// 日本語の語と語のあいだの空白。どの境目を数えるかを品詞で決める。
+import type { Sentence, Span, Token } from "./plugin.ts";
+import { isJapanese } from "./orthography.ts";
+import { isWithinAny, quotedIn } from "./quoted-span.ts";
+
+/** between-phrases は助詞の後ろ（文節の切れ目）、inside-phrase は文節を始められない語の前（助詞・助動詞・名詞に続く「する」）。 */
+export type JointKind = "between-phrases" | "inside-phrase";
+
+/** offset は前の語の終わり（文書の座標）。before / after は境目の両側の語。 */
+export type Joint = { readonly offset: number; readonly kind: JointKind; readonly spaced: boolean; readonly before: string; readonly after: string };
+
+const SPACES = /^[ \u3000]+$/u;
+const PARTICLE = new Set(["ADP", "SCONJ"]);
+// 文節の頭に立てない語。「確認 しました」の「する」は名詞の後ろのときだけ（「し始める」の「し」は動詞の頭に立つ）。
+const BOUND = new Set(["ADP", "SCONJ", "PART", "AUX"]);
+const LIGHT_VERB = "する";
+
+/**
+ * 空白の後ろの「が」を、解析器は文頭の接続詞（「が、」）と読む。一字の接続詞（が・と・や・か）は助詞として扱う。
+ * 二字以上（また・なお）は文節を始める接続詞。
+ */
+const isBound = (token: Token): boolean => BOUND.has(token.pos) || (token.pos === "CCONJ" && [...token.surface].length === 1);
+
+const kindOf = (left: Token, right: Token): JointKind | undefined => {
+  if (isBound(right) || (right.lemma === LIGHT_VERB && left.pos === "NOUN")) return "inside-phrase";
+  // 名詞どうし（担当 田中、赤坂 文弥、フルデスクトップ モード）は名札・名前・複合語の書き方で、数えない。
+  return PARTICLE.has(left.pos) ? "between-phrases" : undefined;
+};
+
+/**
+ * 行（か文）の頭の語か。助詞は前の語に付くので、行頭の「助詞」（「ヘ」）は項目の記号を読み違えたもので、後ろの空白は記号と本文の区切り。
+ * 行頭の内容語（折り返した「確認 しました」の「確認」）は語句の頭なので、ここでは外さない。
+ */
+const isLineHead = (text: string, at: number): boolean => text.slice(text.lastIndexOf("\n", at - 1) + 1, at).trim() === "";
+
+const isWord = (token: Token): boolean => token.surface.trim() !== "";
+
+type Pair = { readonly left: Token; readonly right: Token };
+
+const pairsOf = (tokens: readonly Token[]): Pair[] =>
+  tokens.filter(isWord).flatMap((right, index, words) => {
+    const left = words[index - 1];
+    return left === undefined ? [] : [{ left, right }];
+  });
+
+// 強調とリンクの括弧。読み手には見えない。
+const CHROME = /[*_~[\]]/gu;
+
+/**
+ * 空けたか詰めたか。間が何も無いか強調・リンクの括弧だけなら詰めた、書き手が書いた空白（半角・全角）だけなら空けた。
+ * 括弧の隣の空白（「**…こと** とする」「手順は [README]」）は、日本語の隣で強調が効かない Markdown のための空白なので数えない。
+ * 改行（Markdown の改行や、行の折り返し）や、見える字を覆った空白（コード）をまたぐ語も、境目に数えない。
+ */
+const spacing = (sentence: Sentence, written: string, pair: Pair): boolean | undefined => {
+  const [from, to] = [pair.left.span.end - sentence.span.start, pair.right.span.start - sentence.span.start];
+  const gap = written.slice(from, to);
+  if (gap.replace(CHROME, "") === "") return false;
+  return SPACES.test(gap) ? true : undefined;
+};
+
+const jointOf = (sentence: Sentence, written: string, quoted: readonly Span[], pair: Pair): Joint | undefined => {
+  const { left, right } = pair;
+  if (!isJapanese(left.surface.at(-1)) || !isJapanese(right.surface.at(0))) return undefined;
+  const spaced = spacing(sentence, written, pair);
+  if (spaced === undefined || isWithinAny(quoted, { start: left.span.start, end: right.span.end })) return undefined;
+  const kind = kindOf(left, right);
+  if (kind === undefined || (kind === "between-phrases" && isLineHead(sentence.text, left.span.start - sentence.span.start))) return undefined;
+  return { offset: left.span.end, kind, spaced, before: left.surface, after: right.surface };
+};
+
+/**
+ * 文の中の、日本語の語と語の境目のうち、空けるか詰めるかを数えるもの。written は本文でないものを覆う前の、文と同じ範囲。
+ * 英字・数字との境目は latin-spacing が見るので入れない。
+ */
+export const phraseJoints = (sentence: Sentence, written: string): Joint[] => {
+  const quoted = quotedIn(sentence);
+  return pairsOf(sentence.tokens ?? []).flatMap((pair) => {
+    const joint = jointOf(sentence, written, quoted, pair);
+    return joint === undefined ? [] : [joint];
+  });
+};

package/src/structure/relative-resolve.ts

@@ -10,7 +10,13 @@ import type { StructureNode } from "../plugin.ts";
 /** range は「第四十三条から第五十五条まで 削除」のように、一行で並びをまとめたもの。番地は最初のものしか持たない。 */
 type Place = { readonly address: string; readonly level: number; readonly range?: boolean };
 
-type Last = { readonly target: string; readonly document: string | undefined };
+/**
+ * 参照が指した文書: 他の文書の名前と、番号を名指した単位の語。
+ * 単位の語は、見出しに「1 目的」と番号を振った指針の「法第16条」が、条で数える別の文書の条だと決めるのに使う（unit-word.ts）。
+ */
+type Cited = { readonly document: string | undefined; readonly unitWord: string | undefined };
+
+type Last = Cited & { readonly target: string };
 
 const NUMBERED = new Set(["chapter", "article", "item"]);
 
@@ -100,21 +106,21 @@ const continued = (level: number, node: StructureNode, context: Context): Last |
   if (origin === undefined || context.articleLevel === undefined || level < context.articleLevel) return undefined;
   const parts = origin.target.split(".");
   const kept = level - context.articleLevel + 1;
-  return parts.length < kept ? undefined : { target: parts.slice(0, kept).join("."), document: origin.document };
+  return parts.length < kept ? undefined : { ...origin, target: parts.slice(0, kept).join(".") };
 };
 
 /**
  * 参照が名指しした深さを覚える。名指しは、書き出しの深さ（第〇条なら条、前項なら項）から番地の深さまで。
  * 番地を深さで切る起点は条。条より外のまとまりは番地の形が違うので覚えない。
  */
-const remember = (context: Context, target: string, document: string | undefined, from: number): void => {
+const remember = (context: Context, reference: Last, from: number): void => {
   const articleLevel = context.articleLevel;
   // 章・節は番地の形（ch1）が条から始まらないので覚えない。覚えると「同条」が章を指してしまう。
   if (articleLevel === undefined || from < articleLevel) return;
-  const parts = target.split(".");
+  const parts = reference.target.split(".");
   parts.forEach((_, index) => {
     const level = articleLevel + index;
-    if (level >= from) context.last.set(level, { target: parts.slice(0, index + 1).join("."), document });
+    if (level >= from) context.last.set(level, { ...reference, target: parts.slice(0, index + 1).join(".") });
   });
 };
 
@@ -145,12 +151,17 @@ const baseOf = (way: string, node: StructureNode, anchor: Anchor | undefined, la
   return anchor === undefined ? undefined : byPosition(way, Number(node.attrs["count"]), anchor);
 };
 
-/** 同と並びの続きは、元の参照が他の文書を指していれば、それを引き継ぐ。前・次・本はこの文書の中。 */
-const documentOf = (way: string, node: StructureNode, last: Last | undefined, context: Context): string | undefined => {
-  if (way === "same") return last?.document;
-  return way === "continue" ? continued(Number(node.attrs["level"]), node, context)?.document : undefined;
+/** 同と並びの続きは、元の参照が指した文書を引き継ぐ。前・次・本はこの文書の中。 */
+const citedOf = (way: string, node: StructureNode, last: Last | undefined, context: Context): Cited | undefined => {
+  if (way === "same") return last;
+  return way === "continue" ? continued(Number(node.attrs["level"]), node, context) : undefined;
 };
 
+const citedAttrs = (cited: Cited | undefined): Record<string, string> => ({
+  ...(cited?.document === undefined ? {} : { document: cited.document }),
+  ...(cited?.unitWord === undefined ? {} : { unitWord: cited.unitWord }),
+});
+
 const resolveOne = (node: StructureNode, path: readonly StructureNode[], context: Context): StructureNode | undefined => {
   const way = String(node.attrs["relative"]);
   const level = Number(node.attrs["level"]);
@@ -160,12 +171,11 @@ const resolveOne = (node: StructureNode, path: readonly StructureNode[], context
   const place = base === undefined ? undefined : withSuffix(base, String(node.attrs["suffix"] ?? ""), context.implicitLevel);
   if (place === undefined) return undefined;
   const fallback = fallbackOf(place, context.implicitLevel);
-  const document = documentOf(way, node, last, context);
   const attrs = {
     target: place.address,
     label: String(node.attrs["label"]),
     ...(fallback === undefined ? {} : { fallback }),
-    ...(document === undefined ? {} : { document }),
+    ...citedAttrs(citedOf(way, node, last, context)),
   };
   return { ...node, attrs };
 };
@@ -176,17 +186,16 @@ const isRelative = (node: StructureNode): boolean => node.kind === "reference" &
 const walk = (node: StructureNode, path: readonly StructureNode[], context: Context): StructureNode | undefined => {
   if (isRelative(node)) {
     const resolved = resolveOne(node, path, context);
-    const target = resolved === undefined ? undefined : String(resolved.attrs["target"]);
-    const document = optional(resolved?.attrs["document"]);
-    if (target !== undefined) context.byStart.set(node.span.start, { target, document });
-    if (target !== undefined) remember(context, target, document, Number(node.attrs["names"] ?? node.attrs["level"]));
+    if (resolved === undefined) return undefined;
+    const reference = lastOf(resolved);
+    context.byStart.set(node.span.start, reference);
+    remember(context, reference, Number(node.attrs["names"] ?? node.attrs["level"]));
     return resolved;
   }
   if (node.kind === "reference") {
-    const target = String(node.attrs["target"]);
-    const document = optional(node.attrs["document"]);
-    context.byStart.set(node.span.start, { target, document });
-    remember(context, target, document, context.articleLevel ?? 0);
+    const reference = lastOf(node);
+    context.byStart.set(node.span.start, reference);
+    remember(context, reference, context.articleLevel ?? 0);
   }
   const inside = [...path, node];
   const children = node.children.flatMap((child) => {
@@ -198,6 +207,12 @@ const walk = (node: StructureNode, path: readonly StructureNode[], context: Cont
 
 const optional = (value: string | number | undefined): string | undefined => (value === undefined ? undefined : String(value));
 
+const lastOf = (reference: StructureNode): Last => ({
+  target: String(reference.attrs["target"]),
+  document: optional(reference.attrs["document"]),
+  unitWord: optional(reference.attrs["unitWord"]),
+});
+
 const articlesOf = (node: StructureNode): Place[] => {
   const place = node.kind === "article" ? placeOf(node) : undefined;
   return [...(place === undefined ? [] : [place]), ...node.children.flatMap(articlesOf)];

package/src/team-names.ts

@@ -1,4 +1,5 @@
 import type { Span } from "./plugin.ts";
+import { mergeSpans } from "./span-merge.ts";
 
 const SPACE = /\s/u;
 const ESCAPE = /[.*+?^${}()|[\]\\]/gu;
@@ -9,9 +10,14 @@ const WRAP = String.raw`(?:\r?\n[ \t]*)?`;
 const patternOf = (name: string): RegExp => {
   const pieces = Array.from(name.trim()).reduce<string[]>((acc, char) => {
     const previous = acc.at(-1);
-    if (SPACE.test(char)) return previous === WHITESPACE_RUN ? acc : [...acc, WHITESPACE_RUN];
+    if (SPACE.test(char)) {
+      if (previous !== WHITESPACE_RUN) acc.push(WHITESPACE_RUN);
+      return acc;
+    }
     const escaped = char.replace(ESCAPE, "\\$&");
-    return previous === undefined || previous === WHITESPACE_RUN ? [...acc, escaped] : [...acc, WRAP, escaped];
+    if (previous !== undefined && previous !== WHITESPACE_RUN) acc.push(WRAP);
+    acc.push(escaped);
+    return acc;
   }, []);
   return new RegExp(pieces.join(""), "gu");
 };
@@ -19,21 +25,15 @@ const patternOf = (name: string): RegExp => {
 const occurrences = (text: string, name: string): Span[] =>
   name.trim().length === 0 ? [] : [...text.matchAll(patternOf(name))].map((match) => ({ start: match.index, end: match.index + match[0].length }));
 
-const merged = (spans: readonly Span[]): Span[] =>
-  spans
-    .toSorted((left, right) => left.start - right.start)
-    .reduce<Span[]>((acc, span) => {
-      const last = acc.at(-1);
-      if (last !== undefined && span.start < last.end) acc[acc.length - 1] = { start: last.start, end: Math.max(last.end, span.end) };
-      else acc.push(span);
-      return acc;
-    }, []);
-
 /**
  * チームが chaff.yaml の names に並べた名前が、text のどこにあるか。綴りは書かれたとおり（Apple と apple は別）。
  * 重なる出現は 1 つの範囲にまとめる。接するだけなら別の名前（並んだ 2 つの名前は 2 つと数える）。位置は text の先頭を 0 とする。
  */
-export const nameSpans = (text: string, names: readonly string[]): Span[] => merged(names.flatMap((name) => occurrences(text, name)));
+export const nameSpans = (text: string, names: readonly string[]): Span[] =>
+  mergeSpans(
+    names.flatMap((name) => occurrences(text, name)),
+    false,
+  );
 
 /** [start, end) が names の範囲のどれかと重なるか。 */
 export const touchesAny = (spans: readonly Span[], start: number, end: number): boolean => spans.some((span) => start < span.end && span.start < end);