hanlint 0.0.7 → 0.0.9

package/src/rules/code/inputFileSource.js

@@ -1,39 +1,23 @@
 // @ts-check
+import { createdIn, fileName, readsIn, writeTargets } from "../../fingerprint/codeMarkers.js";
 import { DOCUMENT, NOTICE, finding } from "../finding.js";
-import { codeBlocksOf } from "../shared/codeBlocks.js";
 
 export const name = "inputFileSource";
+export const mechanism = "reader";
 
-const QUOTED = "\\(\\s*[\"']([^\"'\\n]+)[\"']";
-const READS = new RegExp(
-  "\\b(?:read_csv|read_excel|read_parquet|read_json|read_sql|read_table|read_ndjson|scan_csv|scan_parquet|scan_ndjson|" +
-    "read_text|read_bytes|loadtxt|imread|load_workbook)" + QUOTED,
-  "g",
-);
-const OPENS = new RegExp("\\bopen" + QUOTED + "\\s*(?:,\\s*(?:mode\\s*=\\s*)?[\"']([^\"']*)[\"'])?", "g");
-const WRITES = new RegExp(
-  "\\b(?:to_csv|to_excel|to_parquet|to_json|to_sql|write_csv|write_parquet|write_json|write_ndjson|sink_parquet|sink_csv|" +
-    "write_text|write_bytes|savefig|save|imwrite|savetxt|dump)" + QUOTED,
-  "g",
-);
-const SHELL_WRITES = /(?:>>?|-o|-O|--output|Out-File)\s*["']?([\w./\\-]+\.[A-Za-z0-9]+)/g;
-const SQL_PATHS = /FROM\s+'([^']+\.(?:csv|parquet|json|xlsx))'/gi;
-const MAKES_DIR = /(?:mkdir(?:\s+-p)?|makedirs|md)\s*\(?\s*["']?([\w./\\-]+)|Path\(\s*["']([^"']+)["']\s*\)\.mkdir/g;
 const EXTENSIONS = new Set(
   "csv xlsx xls parquet json jsonl txt png jpg jpeg svg gif db sqlite md py yaml yml toml html pdf zip ndjson".split(" "),
 );
-
-/** @param {string} path */
-function fileName(path) {
-  const parts = path.replace(/\\/g, "/").replace(/\/+$/, "").split("/");
-  return parts[parts.length - 1];
-}
+/** 별표, 서식 자리표시자, URL 이 든 경로는 파일 하나를 가리키지 않는다. */
+const NOT_A_FILE = ["*", "{", "http", "://", "<", ">"];
+/** 폴더를 볼 때는 물음표까지 뺀다. */
+const NOT_A_PATH = [...NOT_A_FILE, "?"];
 
 /** @param {string} path */
 function isDataFile(path) {
-  if (["*", "{", "http", "://", "<", ">"].some((mark) => path.includes(mark))) return false;
-  const name = fileName(path);
-  return name.includes(".") && EXTENSIONS.has(name.split(".").pop()?.toLowerCase() ?? "");
+  if (NOT_A_FILE.some((mark) => path.includes(mark))) return false;
+  const name2 = fileName(path);
+  return name2.includes(".") && EXTENSIONS.has(name2.split(".").pop()?.toLowerCase() ?? "");
 }
 
 /** @param {string} path */
@@ -42,62 +26,33 @@ function directoryOf(path) {
   return parts.length > 1 && !["", ".", ".."].includes(parts[0]) ? parts[0] : null;
 }
 
-/** @param {import("../../fingerprint/build.js").DocumentPrint} doc @param {string} needle @param {number} line */
-function mentionedBefore(doc, needle, line) {
-  return doc.sentences.some((s) => s.line < line && s.text.includes(needle));
-}
-
-/** @param {RegExp} pattern @param {string} text @param {number} group */
-function all(pattern, text, group) {
-  pattern.lastIndex = 0;
-  const found = [];
-  for (const match of text.matchAll(pattern)) if (match[group] !== undefined) found.push(match[group]);
-  return found;
-}
-
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {
   const findings = [];
-  /** @type {Set<string>} */
-  const created = new Set();
-  for (const block of codeBlocksOf(doc)) {
-    /** @type {Set<string>} */
-    const blockWrites = new Set();
-    /** @type {Set<string>} */
-    const blockDirs = new Set();
-    for (const [, code] of block.lines) {
-      for (const name2 of all(WRITES, code, 1)) blockWrites.add(fileName(name2));
-      for (const name2 of all(SHELL_WRITES, code, 1)) blockWrites.add(fileName(name2));
-      OPENS.lastIndex = 0;
-      for (const match of code.matchAll(OPENS)) {
-        const mode = match[2] ?? "r";
-        if ([..."wax"].some((flag) => mode.includes(flag))) blockWrites.add(fileName(match[1]));
-      }
-      MAKES_DIR.lastIndex = 0;
-      for (const match of code.matchAll(MAKES_DIR)) blockDirs.add((match[1] ?? match[2]).replace(/\\/g, "/").split("/")[0]);
-    }
-    for (const dir of blockDirs) created.add(dir);
+  /** 이미 짚은 폴더. 같은 폴더를 두 번 짚지 않는다. @type {Set<string>} */
+  const reported = new Set();
+  for (const block of doc.codeBlocks) {
+    const reader = doc.reader.beforeBlock[block.index];
+    const [blockWrites, blockDirs] = createdIn(block.lines.map(([, line]) => line));
+    // 이 블록이 만드는 폴더는 같은 블록의 읽기에도 미리 센다. 파일은 블록 안에서 쓴 것 (blockWrites) 만 따로 본다.
+    const have = new Set([...reader.files, ...blockDirs, ...reported]);
     for (const [line, code] of block.lines) {
-      OPENS.lastIndex = 0;
-      const reads = [...all(READS, code, 1), ...all(SQL_PATHS, code, 1)];
-      for (const match of code.matchAll(OPENS)) {
-        if (![..."wax"].some((flag) => (match[2] ?? "r").includes(flag))) reads.push(match[1]);
-      }
+      const reads = readsIn(code);
       for (const path of reads) {
         if (!isDataFile(path)) continue;
         const name2 = fileName(path);
-        if (created.has(name2) || blockWrites.has(name2) || mentionedBefore(doc, name2, block.startLine)) continue;
+        if (have.has(name2) || blockWrites.has(name2) || doc.reader.mentionedBefore(block.index, name2)) continue;
         findings.push(finding(name, line, code.trim(), `\`${name2}\` 을 읽는데 글 어디에서도 만들지 않았다. 독자는 여기서 파일 없음 오류로 멈춘다`, null, "error", DOCUMENT, block.index));
       }
-      for (const path of [...all(WRITES, code, 1), ...reads]) {
-        if (["http", "://", "{", "<", ">", "*", "?"].some((mark) => path.includes(mark))) continue;
+      for (const path of [...writeTargets(code), ...reads]) {
+        if (NOT_A_PATH.some((mark) => path.includes(mark))) continue;
         const directory = directoryOf(path);
-        if (!directory || created.has(directory) || mentionedBefore(doc, directory, block.startLine)) continue;
-        created.add(directory);
+        if (!directory || have.has(directory) || doc.reader.mentionedBefore(block.index, directory)) continue;
+        have.add(directory);
+        reported.add(directory);
         findings.push(finding(name, line, code.trim(), `\`${directory}\` 폴더를 쓰는데 글 어디에서도 만들지 않았다. 없으면 파일을 쓰다 멈춘다`, null, NOTICE, DOCUMENT, block.index));
       }
     }
-    for (const write of blockWrites) created.add(write);
   }
   return findings;
 }

package/src/rules/code/installImport.js

@@ -1,11 +1,12 @@
 // @ts-check
 import { loadLines } from "../../data/load.js";
 import { DOCUMENT, NOTICE, finding } from "../finding.js";
-import { codeBlocksOf } from "../shared/codeBlocks.js";
 
 export const name = "installImport";
+export const mechanism = "contrast";
 
-const INSTALL = /(?:pip\s+install|uv\s+add|uv\s+pip\s+install|conda\s+install|poetry\s+add)\s+([^\n#|&;]+)/g;
+// 캡처는 백틱과 한글에서 멈춘다. 안 멈추면 설치 줄 뒤의 산문을 패키지로 등록한다. 파이썬과 같다.
+const INSTALL = /(?:pip\s+install|uv\s+add|uv\s+pip\s+install|conda\s+install|poetry\s+add)\s+([^\n#|&;`가-힣]+)/g;
 const IMPORT = /^\s*(?:import\s+([\w.]+(?:\s*,\s*[\w.]+)*)|from\s+([\w.]+)\s+import\b)/;
 const REQUIREMENT = /^([A-Za-z0-9][A-Za-z0-9._-]*)(?:\[([^\]]*)\])?/;
 const PY_FILE = /\b(\w+)\.py\b/g;
@@ -49,7 +50,7 @@ function normalize(value) {
 
 /** 설치 줄이 말한 패키지 → extras. 설치 줄이 하나도 없으면 null. @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 function installed(doc) {
-  const texts = [...codeBlocksOf(doc).flatMap((b) => b.lines.map(([, line]) => line)), ...doc.sentences.map((s) => s.text)];
+  const texts = [...doc.codeBlocks.flatMap((b) => b.lines.map(([, line]) => line)), ...doc.sentences.map((s) => s.text)];
   /** @type {Map<string, Set<string>>} */
   const packages = new Map();
   let seen = false;
@@ -80,7 +81,7 @@ function localModules(doc) {
     PY_FILE.lastIndex = 0;
     for (const match of sentence.text.matchAll(PY_FILE)) names.add(match[1]);
   }
-  for (const block of codeBlocksOf(doc)) {
+  for (const block of doc.codeBlocks) {
     for (const [, line] of block.lines) {
       PY_FILE.lastIndex = 0;
       for (const match of line.matchAll(PY_FILE)) names.add(match[1]);
@@ -95,7 +96,7 @@ export function run(doc) {
   if (packages === null) return [];
   const local = localModules(doc);
   const findings = [];
-  for (const block of codeBlocksOf(doc)) {
+  for (const block of doc.codeBlocks) {
     if (!["python", "py"].includes(block.language)) continue;
     for (const [line, code] of block.lines) {
       const match = IMPORT.exec(code);

package/src/rules/code/platformApi.js

@@ -1,17 +1,18 @@
 // @ts-check
 import { loadLines } from "../../data/load.js";
 import { DOCUMENT, NOTICE, finding } from "../finding.js";
-import { codeBlocksOf } from "../shared/codeBlocks.js";
 
 export const name = "platformApi";
+export const mechanism = "dictionary";
 
-/** @type {[RegExp, string, string][] | null} */
+/** @type {[RegExp, string, string, string][] | null} */
 let cache = null;
 function platformApis() {
   if (!cache) {
     cache = loadLines("platformApis.txt").map((line) => {
       const [pattern, platform, alternative] = line.split("\t");
-      return /** @type {[RegExp, string, string]} */ ([new RegExp(pattern), platform, alternative]);
+      // 지적문에는 정규식 원문을 적는다. V8 의 RegExp.source 는 / 를 \/ 로 돌려줘 파이썬의 pattern.pattern 과 갈린다.
+      return /** @type {[RegExp, string, string, string]} */ ([new RegExp(pattern), pattern, platform, alternative]);
     });
   }
   return cache;
@@ -20,13 +21,13 @@ function platformApis() {
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {
   const findings = [];
-  for (const block of codeBlocksOf(doc)) {
+  for (const block of doc.codeBlocks) {
     if (block.isOutput) continue;
     for (const [line, code] of block.lines) {
-      for (const [pattern, platform, alternative] of platformApis()) {
+      for (const [pattern, source, platform, alternative] of platformApis()) {
         pattern.lastIndex = 0;
         if (pattern.test(code)) {
-          findings.push(finding(name, line, code.trim(), `\`${pattern.source}\` 은 ${platform} 전용이다. 다른 운영체제 독자는 여기서 멈춘다. ${alternative}`, null, NOTICE, DOCUMENT, block.index));
+          findings.push(finding(name, line, code.trim(), `\`${source}\` 은 ${platform} 전용이다. 다른 운영체제 독자는 여기서 멈춘다. ${alternative}`, null, NOTICE, DOCUMENT, block.index));
           break;
         }
       }

package/src/rules/document/countMismatch.js

@@ -2,6 +2,7 @@
 import { DOCUMENT, finding } from "../finding.js";
 
 export const name = "countMismatch";
+export const mechanism = "contrast";
 
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc @param {number} line */
 function sectionIndexAt(doc, line) {
@@ -10,16 +11,31 @@ function sectionIndexAt(doc, line) {
   return index;
 }
 
-/** 같은 절 안이거나, 도입과 마지막 절 사이일 때만 같은 목록을 센다고 본다. */
-function comparable(doc, lineA, lineB) {
+/** 줄이 든 문단의 차례. sectionIndexAt 과 같은 꼴로 훑는다. */
+function blockIndexAt(doc, line) {
+  let index = 0;
+  for (const block of doc.blocks) if (block.startLine <= line) index = block.index;
+  return index;
+}
+
+/** 도입의 약속과 마지막 절의 결산만 같은 목록으로 본다. 절이 없으면 첫 문단과 마지막 문단이다. 파이썬과 같다. */
+function comparable(doc, lineA, lineB, span) {
   const a = sectionIndexAt(doc, lineA);
   const b = sectionIndexAt(doc, lineB);
   const last = doc.sections[doc.sections.length - 1].index;
-  return a === b || (Math.min(a, b) === 0 && Math.max(a, b) === last && a !== b);
+  if (last === 0) {
+    if (Math.max(...doc.blocks.map((block) => block.endLine)) > span) return false;
+    const lastBlock = doc.blocks[doc.blocks.length - 1].index;
+    return lastBlock !== 0 && blockIndexAt(doc, lineA) === 0 && blockIndexAt(doc, lineB) === lastBlock;
+  }
+  return Math.min(a, b) === 0 && Math.max(a, b) === last && a !== b;
 }
 
-/** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
-export function run(doc) {
+/**
+ * @param {import("../../fingerprint/build.js").DocumentPrint} doc
+ * @param {import("../../config/settings.js").Config} config
+ */
+export function run(doc, config) {
   const findings = [];
   /** @type {Map<string, [number, number, string][]>} */
   const byUnit = new Map();
@@ -31,7 +47,7 @@ export function run(doc) {
     promises.sort((a, b) => a[0] - b[0] || a[1] - b[1] || (a[2] < b[2] ? -1 : a[2] > b[2] ? 1 : 0));
     for (let i = 0; i < promises.length; i++) {
       const [lineA, numberA, textA] = promises[i];
-      const conflict = promises.slice(i + 1).find((p) => p[1] !== numberA && comparable(doc, lineA, p[0]));
+      const conflict = promises.slice(i + 1).find((p) => p[1] !== numberA && comparable(doc, lineA, p[0], config.countMismatchSpan));
       if (!conflict) continue;
       const [lineB, , textB] = conflict;
       findings.push(finding(name, lineB, textB, `${lineA}번째 줄은 \`${textA}\` 인데 여기는 \`${textB}\` 다. 같은 단위로 다른 수를 약속하면 독자가 어느 쪽을 믿을지 모른다`, null, "error", DOCUMENT, -1));

package/src/rules/document/enoughOnce.js

@@ -0,0 +1,25 @@
+// @ts-check
+import { matchedTexts } from "../../fingerprint/markers.js";
+import { SENTENCE, finding } from "../finding.js";
+
+export const name = "enoughOnce";
+export const mechanism = "repeat";
+
+/**
+ * @param {import("../../fingerprint/build.js").DocumentPrint} doc
+ * @param {import("../../config/settings.js").Config} config
+ */
+export function run(doc, config) {
+  void config;
+  /** @type {[import("../../fingerprint/build.js").SentencePrint, string][]} */
+  const breaks = [];
+  for (const sentence of doc.sentences) {
+    const texts = matchedTexts(sentence.text, "enoughMarkers.txt");
+    if (texts.length) breaks.push([sentence, texts[0]]);
+  }
+  if (breaks.length < 2) return [];
+  const first = breaks[0][0];
+  return breaks.slice(1).map(([sentence, text]) =>
+    finding(name, sentence.line, sentence.text, `\`${text}\` 로 끊는 문장이 ${first.line}행에 이미 있다. 글 전체에 한 번만 쓴다. 계속 끝나는 척하다 이어지는 꼴이 된다`, null, "error", SENTENCE, sentence.index),
+  );
+}

package/src/rules/document/fieldEcho.js

@@ -0,0 +1,38 @@
+// @ts-check
+import { topicsOf } from "../../fingerprint/topics.js";
+import { DOCUMENT, NOTICE, finding } from "../finding.js";
+
+export const name = "fieldEcho";
+export const mechanism = "contrast";
+
+/**
+ * @param {import("../../fingerprint/build.js").DocumentPrint} doc
+ * @param {import("../../fingerprint/build.js").SectionPrint} section
+ * @param {string[]} fields
+ * @param {string} where
+ */
+function check(doc, section, fields, where) {
+  const findings = [];
+  for (const name_ of fields) {
+    const value = doc.frontmatter[name_];
+    if (!value) continue;
+    const promised = topicsOf(value);
+    if (!promised.size) continue;
+    if ([...promised].some((word) => section.topics.has(word))) continue;
+    findings.push(
+      finding(name, section.startLine, value, `frontmatter 의 \`${name_}\` 가 약속한 말이 ${where}에 하나도 없다. 약속한 것을 ${where}에서 그 말로 답한다`, null, NOTICE, DOCUMENT, section.index),
+    );
+  }
+  return findings;
+}
+
+/**
+ * @param {import("../../fingerprint/build.js").DocumentPrint} doc
+ * @param {import("../../config/settings.js").Config} config
+ */
+export function run(doc, config) {
+  if (!config.introFields.length && !config.endingFields.length) return [];
+  const findings = check(doc, doc.intro, config.introFields, "도입");
+  if (doc.sections.length > 1) findings.push(...check(doc, doc.sections[doc.sections.length - 1], config.endingFields, "마지막 절"));
+  return findings;
+}

package/src/rules/document/keywordHeading.js

@@ -0,0 +1,32 @@
+// @ts-check
+import { DOCUMENT, NOTICE, finding } from "../finding.js";
+
+export const name = "keywordHeading";
+export const mechanism = "contrast";
+/** 목차라고 부를 수 있는 최소 H2 수. 둘 이하는 훑을 목차가 아니다. */
+const MIN_HEADINGS = 3;
+/** 한 글자 낱말은 어느 제목에나 우연히 들어 있어 가르지 못한다. */
+const MIN_WORD = 2;
+
+/**
+ * @param {import("../../fingerprint/build.js").DocumentPrint} doc
+ * @param {import("../../config/settings.js").Config} config
+ */
+export function run(doc, config) {
+  if (!config.keywordField) return [];
+  const keyword = (doc.frontmatter[config.keywordField] ?? "").trim();
+  const title = doc.frontmatter.title ?? "";
+  const headings = doc.headings.filter(([level]) => level === 2);
+  if (!keyword || headings.length < MIN_HEADINGS) return [];
+  const joined = headings.map(([, text]) => text).join(" / ");
+  // keywordMissing 과 같은 자리를 본다. 제목만 보면 첫 문단에만 있는 검색어가 두 규칙 사이로 빠진다. 파이썬과 같다.
+  const first = doc.paragraphs.length ? doc.paragraphs[0].text : "";
+  const promised = `${title}\n${first}`;
+  const missing = keyword
+    .split(/\s+/)
+    .filter((word) => word.length >= MIN_WORD && promised.includes(word) && !joined.includes(word));
+  if (!missing.length) return [];
+  return [
+    finding(name, headings[0][2], joined, `글이 내건 \`${missing[0]}\` 가 절 제목 어디에도 없다. 목차를 훑는 독자가 이 글이 그것을 다루는지 못 본다`, null, NOTICE, DOCUMENT, -1),
+  ];
+}

package/src/rules/document/keywordMissing.js

@@ -2,6 +2,7 @@
 import { DOCUMENT, finding } from "../finding.js";
 
 export const name = "keywordMissing";
+export const mechanism = "contrast";
 
 /**
  * @param {import("../../fingerprint/build.js").DocumentPrint} doc

package/src/rules/document/noQuestion.js

@@ -2,10 +2,14 @@
 import { DOCUMENT, finding } from "../finding.js";
 
 export const name = "noQuestion";
+export const mechanism = "threshold";
 
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {
   if (doc.bodySections.length < 2 || doc.questionCount > 0) return [];
   const first = doc.sentences.length ? doc.sentences[0] : null;
-  return [finding(name, first ? first.line : 1, first ? first.text : "", "물음표가 한 번도 없다. 독자가 품을 의문을 한 번은 대신 묻고 다음 문장에서 답한다", null, "error", DOCUMENT, -1)];
+  const why = doc.readerCallCount === 0
+    ? "물음표가 한 번도 없고 독자를 부르는 말도 없다. 독자가 품을 의문을 한 번은 대신 묻고, 독자가 할 행동을 동사로 끝내는 문장을 넣는다"
+    : "물음표가 한 번도 없다. 독자가 품을 의문을 한 번은 대신 묻고 다음 문장에서 답한다";
+  return [finding(name, first ? first.line : 1, first ? first.text : "", why, null, "error", DOCUMENT, -1)];
 }

package/src/rules/document/promiseRecall.js

@@ -2,11 +2,13 @@
 import { DOCUMENT, NOTICE, finding } from "../finding.js";
 
 export const name = "promiseRecall";
+export const mechanism = "reader";
 
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {
-  if (!doc.promises.length || doc.recalls.length) return [];
-  return doc.promises.map(([line, text]) =>
+  const reader = doc.reader.final;
+  if (!reader.promises.length || reader.recalls.length) return [];
+  return reader.promises.map(([line, text]) =>
     finding(name, line, text, "뒤로 미룬 것을 앞에서 미룬 것으로 다시 부르는 자리가 글 어디에도 없다. 회수하거나 미루지 않는다", null, NOTICE, DOCUMENT, -1),
   );
 }

package/src/rules/document/tableOddCell.js

@@ -0,0 +1,61 @@
+// @ts-check
+import { TABLE } from "../../document/model.js";
+import { DOCUMENT, NOTICE, finding } from "../finding.js";
+
+export const name = "tableOddCell";
+export const mechanism = "contrast";
+const SEPARATOR = /^:?-{2,}:?$/;
+// 단위는 한 글자 이상이다. 0 을 허용하면 맨숫자 칸이 빈 단위를 받아 지적문이 깨진다. 파이썬과 같다.
+const MEASURE = /^([+-]?\d[\d,]*(?:\.\d+)?)\s*([^\s\d]{1,6})$/;
+const EMPTY = ["", "-", "--"];
+
+/** (블록 안 줄 오프셋, 칸 목록). 구분 줄은 뺀다. @param {string} text @returns {[number, string[]][]} */
+function cellsOf(text) {
+  /** @type {[number, string[]][]} */
+  const rows = [];
+  text.split("\n").forEach((line, offset) => {
+    const stripped = line.trim();
+    if (!stripped.startsWith("|")) return;
+    const cells = stripped.replace(/^\|+/, "").replace(/\|+$/, "").split("|").map((cell) => cell.trim());
+    if (cells.length && cells.filter((cell) => cell).every((cell) => SEPARATOR.test(cell))) return;
+    rows.push([offset, cells]);
+  });
+  return rows;
+}
+
+/** `3.72초` 는 초, `533MB` 는 MB, `직접 부르지 않음` 은 null. @param {string} cell */
+function unitOf(cell) {
+  const match = MEASURE.exec(cell);
+  return match ? match[2] : null;
+}
+
+/**
+ * @param {import("../../fingerprint/build.js").DocumentPrint} doc
+ * @param {import("../../config/settings.js").Config} config
+ */
+export function run(doc, config) {
+  const findings = [];
+  for (const block of doc.blocks) {
+    if (block.kind !== TABLE) continue;
+    const rows = cellsOf(block.text);
+    if (rows.length < 2) continue;
+    const body = rows.slice(1);
+    const width = Math.min(...body.map(([, cells]) => cells.length));
+    for (let column = 1; column < width; column += 1) {
+      /** @type {[number, string][]} */
+      const values = [];
+      for (const [offset, cells] of body) if (!EMPTY.includes(cells[column])) values.push([offset, cells[column]]);
+      if (values.length < config.tableOddCellMinRows) continue;
+      const units = values.map(([offset, cell]) => /** @type {[number, string, string | null]} */ ([offset, cell, unitOf(cell)]));
+      const odd = units.filter(([, , unit]) => unit === null);
+      const kinds = new Set(units.filter(([, , unit]) => unit !== null).map(([, , unit]) => unit));
+      if (odd.length !== 1 || kinds.size !== 1) continue;
+      const unit = [...kinds][0];
+      const [offset, cell] = odd[0];
+      findings.push(
+        finding(name, block.startLine + offset, cell, `이 열의 나머지 ${values.length - 1}칸은 \`${unit}\` 으로 잰 값 하나인데 이 칸만 다르다. 같은 잣대로 다시 재거나 그 줄을 본문 문장으로 옮긴다`, null, NOTICE, DOCUMENT, block.index),
+      );
+    }
+  }
+  return findings;
+}

package/src/rules/finding.js

@@ -9,6 +9,8 @@ export const PARAGRAPH = "paragraph";
 export const SECTION = "section";
 export const DOCUMENT = "document";
 
+const MIN_LOCAL_CUE = 2;
+
 /**
  * @typedef {object} Finding
  * @property {string} rule
@@ -21,8 +23,11 @@ export const DOCUMENT = "document";
  * @property {number} at scope 가 가리키는 지문의 index
  * @property {string | null} fragment 원문에서 바꿀 조각. hanlint fix 가 replacement 로 바꾼다
  * @property {string | null} replacement
+ * @property {Candidate[]} candidates 뜻을 정하지 않고 형태로 좁힐 수 있을 때만. 순위와 점수는 없다
  */
 
+/** @typedef {{text: string, why: string}} Candidate */
+
 /**
  * @param {string} rule
  * @param {number} line
@@ -34,10 +39,11 @@ export const DOCUMENT = "document";
  * @param {number} [at]
  * @param {string | null} [fragment]
  * @param {string | null} [replacement]
+ * @param {Candidate[]} [candidates]
  * @returns {Finding}
  */
-export function finding(rule, line, quote, why, fix = null, severity = ERROR, scope = SENTENCE, at = -1, fragment = null, replacement = null) {
-  return { rule, line, quote, why, fix, severity, scope, at, fragment, replacement };
+export function finding(rule, line, quote, why, fix = null, severity = ERROR, scope = SENTENCE, at = -1, fragment = null, replacement = null, candidates = []) {
+  return { rule, line, quote, why, fix, severity, scope, at, fragment, replacement, candidates };
 }
 
 /** @param {Finding} f */
@@ -49,5 +55,21 @@ export function findingAsDict(f) {
     data.fragment = f.fragment;
     data.replacement = f.replacement;
   }
+  if (f.candidates.length) data.candidates = f.candidates.map((candidate) => ({ text: candidate.text, why: candidate.why }));
   return data;
 }
+
+/** 규칙이 짚은 국소 표지. 너무 짧으면 오선택을 피하려고 인용 전체를 쓴다. @param {Finding} f */
+export function localCue(f) {
+  const flat = (text) => text.split(/\s+/).filter(Boolean).join(" ");
+  const quote = flat(f.quote);
+  if (f.fragment) {
+    const fragment = flat(f.fragment);
+    if (fragment.length >= MIN_LOCAL_CUE) return fragment;
+  }
+  const marked = [...f.why.matchAll(/`([^`\n]+)`/g)]
+    .map((match) => flat(match[1]))
+    .filter((candidate) => candidate.length >= MIN_LOCAL_CUE && quote.includes(candidate));
+  marked.sort((left, right) => right.length - left.length || (left < right ? 1 : left > right ? -1 : 0));
+  return marked[0] ?? quote;
+}

package/src/rules/orthography/confusable.js

@@ -2,6 +2,7 @@
 import { dictionaryFindings } from "../shared/dictionaryRule.js";
 
 export const name = "confusable";
+export const mechanism = "dictionary";
 
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {

package/src/rules/orthography/spacing.js

@@ -2,6 +2,7 @@
 import { dictionaryFindings } from "../shared/dictionaryRule.js";
 
 export const name = "spacing";
+export const mechanism = "dictionary";
 
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {

package/src/rules/orthography/spelling.js

@@ -2,6 +2,7 @@
 import { dictionaryFindings } from "../shared/dictionaryRule.js";
 
 export const name = "spelling";
+export const mechanism = "dictionary";
 
 /** @param {import("../../fingerprint/build.js").DocumentPrint} doc */
 export function run(doc) {

package/src/rules/paragraph/factListParagraph.js

@@ -2,6 +2,7 @@
 import { NOTICE, PARAGRAPH, finding } from "../finding.js";
 
 export const name = "factListParagraph";
+export const mechanism = "threshold";
 
 /** 문장 사이를 잇는 표지가 하나라도 있는가. @param {import("../../fingerprint/build.js").ParagraphPrint} paragraph */
 function isLinked(paragraph) {

package/src/rules/paragraph/paraFragment.js

@@ -1,7 +1,9 @@
 // @ts-check
 import { PARAGRAPH, finding } from "../finding.js";
+import { runsOf } from "../shared/repeat.js";
 
 export const name = "paraFragment";
+export const mechanism = "repeat";
 
 /**
  * @param {import("../../fingerprint/build.js").DocumentPrint} doc
@@ -10,19 +12,27 @@ export const name = "paraFragment";
 export function run(doc, config) {
   const findings = [];
   for (const section of doc.sections) {
-    /** @type {import("../../fingerprint/build.js").ParagraphPrint[]} */
-    let run = [];
-    for (const paragraph of section.paragraphs) {
+    const paragraphs = section.paragraphs;
+    // 열쇠는 짧은 문단의 구간 번호다. 코드나 표 뒤의 짧은 문단과 긴 문단 뒤의 짧은 문단은 새 구간을 연다.
+    /** @type {string[]} */
+    const keys = [];
+    let runId = 0;
+    let previousShort = false;
+    for (const paragraph of paragraphs) {
       const short = paragraph.sentenceCount <= 2;
-      if (short && (!run.length || paragraph.followsProseDirectly)) run.push(paragraph);
-      else if (short) run = [paragraph];
-      else run = [];
-      if (run.length === config.fragmentRun) {
-        const first = run[0];
-        findings.push(
-          finding(name, first.startLine, first.sentences.length ? first.sentences[0].text : "", `${run.length}개 문단이 연달아 한두 문장씩이다. 화제가 같으면 한 문단으로 묶는다. 줄바꿈은 화제가 바뀌는 자리에만`, null, "error", PARAGRAPH, first.index),
-        );
+      if (short) {
+        if (!previousShort || !paragraph.followsProseDirectly) runId += 1;
+        keys.push(`short${runId}`);
+      } else {
+        keys.push(`__${paragraph.index}`);
       }
+      previousShort = short;
+    }
+    for (const [start] of runsOf(keys, config.fragmentRun)) {
+      const first = paragraphs[start];
+      findings.push(
+        finding(name, first.startLine, first.sentences.length ? first.sentences[0].text : "", `${config.fragmentRun}개 문단이 연달아 한두 문장씩이다. 화제가 같으면 한 문단으로 묶는다. 줄바꿈은 화제가 바뀌는 자리에만`, null, "error", PARAGRAPH, first.index),
+      );
     }
   }
   return findings;