@maycombhq/atticus 2.0.0-beta.2 → 2.0.0-beta.22

package/dist/{chunk-T6AIP7H5.js → chunk-ISUAAMGS.js}

@@ -6,14 +6,22 @@ var __atticusPath = (p) => __atticusFileURLToPath(__atticusUrl(p));
 import {
   dotInt8,
   textSha
-} from "./chunk-CNUKCS2T.js";
+} from "./chunk-P32RUW3B.js";
 import {
+  JOINED_BY_PERSON,
   Question,
   RECORD_KINDS,
   RecordId,
+  answerByWords,
   changesDirOf,
+  decodeText,
   deserializeRecord,
+  globMatcher,
   identifierWords,
+  ignoreRules,
+  ignoredBy,
+  isBinaryBytes,
+  isBinaryPath,
   knowledgeDirOf,
   knowledgeKindOf,
   layoutOf,
@@ -21,14 +29,17 @@ import {
   parseRecord,
   questionsDirOf,
   recordReferences,
-  require_dist,
   serializeRecord,
   specsDirOf,
   specsRelOf,
   splitForms,
   stateDirOf,
-  toCanonicalYaml
-} from "./chunk-S2A7YC3Q.js";
+  toCanonicalYaml,
+  walkRepo
+} from "./chunk-Q2TECDIT.js";
+import {
+  require_dist
+} from "./chunk-BQGDB6ZL.js";
 import {
   __toESM
 } from "./chunk-TV45EUGM.js";
@@ -870,9 +881,9 @@ function questionNodes(path, text) {
         name: q.question,
         file: path,
         verified: false,
-        line: `answered question ${q.id}: ${q.question} Answer: ${answer} (${a.by}, ${a.at}) [${path}]`,
+        line: `answered question ${q.id}: ${q.question} Answer: ${answer} (${answerByWords(a.by)}, ${a.at}) [${path}]`,
         detail: `Question ${q.id} (${path}), asked by ${q.askedBy.step} about repo ${q.repo}: ${q.question}
-Answer, by ${a.by} at ${a.at}: ${answer}
+Answer, by ${answerByWords(a.by)} at ${a.at}: ${answer}
 Evidence: ${q.evidence.join(", ")}`,
         names: "",
         ids: q.id,
@@ -1004,8 +1015,8 @@ var DOCUMENT_EXTENSIONS = /\.(docx|pptx|xlsx|pdf|md|markdown|html?|txt)$/i;
 async function documentNodes(path, bytes) {
   const out = { nodes: [], names: [], links: [] };
   const [{ readDocument }, { screenFile }] = await Promise.all([
-    import("./chunk-4NDAAFQT.js"),
-    import("./chunk-6JCPI55B.js")
+    import("./chunk-WJF2CS6M.js"),
+    import("./chunk-RLDGWLQ7.js")
   ]);
   const screen = screenFile(path, bytes);
   if (screen.problem)
@@ -2284,7 +2295,7 @@ function openDb(file) {
 // packages/index/dist/catalog-index.js
 var vectorsFileOf = (file) => join3(dirname2(file), "vectors.sqlite");
 var notInstalled = (e, pkg) => e.code === "ERR_MODULE_NOT_FOUND" && String(e.message).includes(pkg);
-async function embedderOf(root, setting, load = () => import("./chunk-TANFV3MZ.js")) {
+async function embedderOf(root, setting, load = () => import("./chunk-WERM6KPJ.js")) {
   if (setting === "off")
     return { why: "turned off for this command" };
   if (setting && setting !== "auto")
@@ -2525,6 +2536,10 @@ var CatalogIndex = class {
   list(kind, o = {}) {
     const where = ["kind = ?"];
     const params = [kind];
+    if (o.confidence) {
+      where.push("confidence = ?");
+      params.push(o.confidence);
+    }
     if (o.status) {
       where.push("status = ?");
       params.push(o.status);
@@ -2713,7 +2728,8 @@ function impact(ix, rootId, depth = 4) {
   };
   for (const a of affected) {
     const rec = recordOf(a.kind, a.id);
-    if (a.kind === "Interface" && rec.confidence === "inferred")
+    const paired = Array.isArray(rec.flags) && rec.flags.includes(JOINED_BY_PERSON);
+    if (a.kind === "Interface" && rec.confidence === "inferred" && !paired)
       addGap({
         record: a.id,
         field: "confidence",
@@ -2867,8 +2883,11 @@ function translate(question, lookup) {
       terms.push(`refs:${rec.id}`);
     if (has(/\b(without|no|missing) (a |an )?pattern\b/))
       terms.push("missing:pattern");
-    if (has(/\b(inferred|unconfirmed|one end)\b/))
+    if (has(/\b(inferred|unconfirmed|one end)\b/)) {
       terms.push("confidence:inferred");
+      if (has(/\bone end\b/))
+        terms.push("-flags:joined-by-person");
+    }
     return {
       action: "query",
       query: terms.join(" "),
@@ -2886,8 +2905,9 @@ function translate(question, lookup) {
 
 // packages/index/dist/tools.js
 import { existsSync, readdirSync } from "node:fs";
-import { readdir as readdir2, readFile as readFile3, realpath, stat as stat2 } from "node:fs/promises";
+import { readFile as readFile3, realpath, stat as stat2 } from "node:fs/promises";
 import { isAbsolute, join as join4, relative as relative2, resolve, sep as sep2 } from "node:path";
+import { createContext, runInContext, Script } from "node:vm";
 var ContextToolError = class extends Error {
   constructor(message) {
     super(message);
@@ -2918,6 +2938,12 @@ var CONTEXT_TOOL_BOUNDS = {
   /** The longest pattern grep takes, and the most of a hit's line it shows. */
   grepPattern: 200,
   grepLineChars: 200,
+  /**
+   * The most time a regular expression may spend on one file, and on one grep call in all,
+   * so a pattern that backtracks without end (`(a+)+$`) can't hang the MCP server or a scan.
+   */
+  grepRegexMs: 1e3,
+  grepRegexTotalMs: 5e3,
   /** Files one grep or files call walks at most, and bytes grep reads at most. */
   walkFiles: 2e4,
   grepBytes: 64e6,
@@ -3286,7 +3312,7 @@ async function excerpt(root, c, around, o) {
   const size = (await stat2(file)).size;
   if (size > CONTEXT_TOOL_BOUNDS.codeFileBytes)
     return `${c.what} ${c.verb ?? "cites"} ${c.repo}/${c.path}:${c.line}, a file too big to show (${size} bytes).`;
-  const lines = (await readFile3(file, "utf8")).split(/\r?\n/);
+  const lines = decodeText(await readFile3(file)).split(/\r?\n/);
   if (c.line > lines.length)
     return `${c.what} ${c.verb ?? "cites"} ${c.repo}/${c.path}:${c.line}, but the file has ${lines.length} lines now: the cite is stale.`;
   const from = Math.max(1, c.line - around);
@@ -3415,17 +3441,20 @@ async function namesInCode(root, names, o = {}) {
       const names2 = open.filter((n) => !found.has(n) && DECLARABLE.test(n));
       if (!names2.length)
         return;
-      const buf = await readFile3(join4(dir, ...p.split("/"))).catch(() => void 0);
+      const file = join4(dir, ...p.split("/"));
+      if (!await readableCode(file, p))
+        return;
+      const buf = await readFile3(file).catch(() => void 0);
       if (!buf || buf.length > B.codeFileBytes)
         return;
-      if (buf.subarray(0, 8e3).includes(0))
+      if (isBinaryBytes(buf))
         return;
       bytes += buf.length;
       if (bytes > B.grepBytes) {
         cut = true;
         return false;
       }
-      const text = buf.toString("utf8");
+      const text = decodeText(buf);
       if (!names2.some((n) => text.includes(n)))
         return;
       const decls = await o.outline(p, text).catch(() => void 0);
@@ -3488,90 +3517,12 @@ function reposOf(root, input, o) {
     throw new ContextToolError("No repo is cloned in the workspace.");
   return out;
 }
-function globRe(glob2, anyDepth = !glob2.replace(/\/$/, "").includes("/")) {
-  let g = glob2.trim().replace(/\\/g, "/").replace(/^\.\//, "");
-  if (g.startsWith("/")) {
-    g = g.slice(1);
-    anyDepth = false;
-  }
-  let re = "";
-  for (let i = 0; i < g.length; i++) {
-    const ch = g[i];
-    if (ch === "*") {
-      if (g[i + 1] === "*") {
-        const slash = g[i + 2] === "/";
-        re += slash ? "(?:.*/)?" : ".*";
-        i += slash ? 2 : 1;
-      } else
-        re += "[^/]*";
-    } else if (ch === "?")
-      re += "[^/]";
-    else if (ch === "[") {
-      const end = g.indexOf("]", i + 2);
-      if (end < 0)
-        re += "\\[";
-      else {
-        const body = g.slice(i + 1, end).replace(/^!/, "^").replace(/\\/g, "\\\\");
-        re += `[${body}]`;
-        i = end;
-      }
-    } else if (ch === "{") {
-      const end = g.indexOf("}", i);
-      if (end < 0)
-        re += "\\{";
-      else {
-        re += `(?:${g.slice(i + 1, end).split(",").map((p) => globRe(p, false).source.slice(1, -1)).join("|")})`;
-        i = end;
-      }
-    } else
-      re += ch.replace(/[.+^$()|\\]/g, "\\$&");
-  }
-  return new RegExp(`^${anyDepth ? "(?:.*/)?" : ""}${re}$`);
-}
-function ignoreRules(text, base) {
-  const out = [];
-  for (let line of text.split(/\r?\n/)) {
-    line = line.replace(/(?<!\\)\s+$/, "");
-    if (!line || line.startsWith("#"))
-      continue;
-    let negate = false;
-    if (line.startsWith("!")) {
-      negate = true;
-      line = line.slice(1);
-    } else if (line.startsWith("\\!") || line.startsWith("\\#"))
-      line = line.slice(1);
-    const dirOnly = line.endsWith("/");
-    if (dirOnly)
-      line = line.slice(0, -1);
-    if (!line)
-      continue;
-    const anchored = line.includes("/");
-    try {
-      out.push({ base, re: globRe(line.replace(/^\//, ""), !anchored), negate, dirOnly });
-    } catch {
-    }
-  }
-  return out;
-}
-function ignored(rules, path, isDir) {
-  let out = false;
-  for (const r of rules) {
-    if (r.dirOnly && !isDir)
-      continue;
-    if (r.base && !path.startsWith(`${r.base}/`))
-      continue;
-    const rel2 = r.base ? path.slice(r.base.length + 1) : path;
-    if (r.re.test(rel2))
-      out = !r.negate;
-  }
-  return out;
-}
 async function rulesFor(dir, path) {
   const parts = path.split("/").slice(0, -1);
   const rules = [];
   for (let i = 0; i <= parts.length; i++) {
     const base = parts.slice(0, i).join("/");
-    const text = await readFile3(join4(dir, ...parts.slice(0, i), ".gitignore"), "utf8").catch(() => void 0);
+    const text = await readFile3(join4(dir, ...parts.slice(0, i), ".gitignore")).then(decodeText, () => void 0);
     if (text)
       rules.push(...ignoreRules(text, base));
   }
@@ -3581,50 +3532,18 @@ async function isIgnored(dir, path) {
   const rules = await rulesFor(dir, path);
   const parts = path.split("/");
   for (let i = 1; i < parts.length; i++)
-    if (ignored(rules, parts.slice(0, i).join("/"), true))
+    if (ignoredBy(rules, parts.slice(0, i).join("/"), true))
       return true;
-  return ignored(rules, path, false);
+  return ignoredBy(rules, path, false);
 }
-async function walkClone(dir, max, visit) {
-  let files = 0;
-  let cut = false;
-  const go = async (rel2, inherited) => {
-    let entries;
-    try {
-      entries = await readdir2(rel2 ? join4(dir, ...rel2.split("/")) : dir, { withFileTypes: true });
-    } catch {
-      return true;
-    }
-    const here = entries.find((e) => e.name === ".gitignore" && e.isFile());
-    const rules = here ? [
-      ...inherited,
-      ...ignoreRules(await readFile3(join4(dir, ...rel2.split("/"), ".gitignore"), "utf8"), rel2)
-    ] : inherited;
-    for (const e of entries.sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0)) {
-      if (e.isSymbolicLink() || e.name === ".git")
-        continue;
-      const p = rel2 ? `${rel2}/${e.name}` : e.name;
-      if (e.isDirectory()) {
-        if (ignored(rules, p, true))
-          continue;
-        if (!await go(p, rules))
-          return false;
-      } else if (e.isFile()) {
-        if (ignored(rules, p, false))
-          continue;
-        if (files >= max) {
-          cut = true;
-          return false;
-        }
-        files++;
-        if (await visit(p) === false)
-          return false;
-      }
-    }
-    return true;
-  };
-  await go("", []);
-  return { files, cut };
+function walkClone(dir, max, visit) {
+  return walkRepo(dir, visit, { max });
+}
+async function readableCode(file, path) {
+  if (isBinaryPath(path))
+    return false;
+  const size = (await stat2(file).catch(() => void 0))?.size;
+  return size !== void 0 && size <= CONTEXT_TOOL_BOUNDS.codeFileBytes;
 }
 var plural = (n, one, many = `${one}s`) => `${n} ${n === 1 ? one : many}`;
 async function runCodeTool(root, name, input, o) {
@@ -3634,6 +3553,34 @@ async function runCodeTool(root, name, input, o) {
     return filesTool(root, input, o);
   return outlineTool(root, input, o);
 }
+var REGEX_LINES = new Script("(() => { const out = []; for (let i = 0; i < lines.length; i++) if (re.test(lines[i])) out.push(i); return out; })()");
+function regexMatcher(pattern, flags) {
+  const B = CONTEXT_TOOL_BOUNDS;
+  const context = createContext({ pattern, flags, lines: [] });
+  runInContext("globalThis.re = new RegExp(pattern, flags);", context);
+  let spent = 0;
+  return (lines) => {
+    const left = B.grepRegexTotalMs - spent;
+    if (left < 1)
+      return void 0;
+    const timeout = Math.min(B.grepRegexMs, left);
+    context.lines = lines;
+    const started = performance.now();
+    try {
+      return Array.from(REGEX_LINES.runInContext(context, { timeout }));
+    } catch (e) {
+      if (e.code === "ERR_SCRIPT_EXECUTION_TIMEOUT") {
+        if (timeout < B.grepRegexMs)
+          return void 0;
+        throw new ContextToolError(`The pattern ${pattern} took too long; make it simpler (no quantifier inside a repeated group, such as (a+)+).`);
+      }
+      throw e;
+    } finally {
+      context.lines = [];
+      spent += performance.now() - started;
+    }
+  };
+}
 async function grepTool(root, input, o) {
   const B = CONTEXT_TOOL_BOUNDS;
   const pattern = typeof input.pattern === "string" && input.pattern ? input.pattern : void 0;
@@ -3642,22 +3589,26 @@ async function grepTool(root, input, o) {
   if (pattern.length > B.grepPattern)
     throw new ContextToolError(`A grep pattern is at most ${B.grepPattern} characters.`);
   const ignoreCase = input.ignoreCase === true;
-  let test;
+  let matching;
   if (input.regex === true) {
-    let re;
     try {
-      re = new RegExp(pattern, ignoreCase ? "i" : "");
+      new RegExp(pattern, ignoreCase ? "i" : "");
     } catch (e) {
       throw new ContextToolError(`${pattern} isn't a regular expression: ${e.message}.`);
     }
-    test = (l) => re.test(l);
+    matching = regexMatcher(pattern, ignoreCase ? "i" : "");
   } else {
     const p = ignoreCase ? pattern.toLowerCase() : pattern;
-    test = ignoreCase ? (l) => l.toLowerCase().includes(p) : (l) => l.includes(p);
+    const test = ignoreCase ? (l) => l.toLowerCase().includes(p) : (l) => l.includes(p);
+    matching = function* (lines) {
+      for (let i = 0; i < lines.length; i++)
+        if (test(lines[i]))
+          yield i;
+    };
   }
   const limit = int(input.limit, 1, B.grepLimit, B.grepDefault);
   const paths = str2(input.paths);
-  const only = paths ? globRe(paths) : void 0;
+  const only = paths ? globMatcher(paths) : void 0;
   const repos = reposOf(root, input, o);
   const hits = [];
   let more = false;
@@ -3668,10 +3619,13 @@ async function grepTool(root, input, o) {
     const walked = await walkClone(dir, B.walkFiles - read, async (p) => {
       if (only && !only.test(p))
         return;
-      const buf = await readFile3(join4(dir, ...p.split("/"))).catch(() => void 0);
+      const file = join4(dir, ...p.split("/"));
+      if (!await readableCode(file, p))
+        return;
+      const buf = await readFile3(file).catch(() => void 0);
       if (!buf || buf.length > B.codeFileBytes)
         return;
-      if (buf.subarray(0, 8e3).includes(0))
+      if (isBinaryBytes(buf))
         return;
       bytes += buf.length;
       if (bytes > B.grepBytes) {
@@ -3679,11 +3633,14 @@ async function grepTool(root, input, o) {
         return false;
       }
       read++;
-      const lines = buf.toString("utf8").split(/\r?\n/);
-      for (let i = 0; i < lines.length; i++) {
-        const line = lines[i].slice(0, 2e3);
-        if (!test(line))
-          continue;
+      const lines = decodeText(buf).split(/\r?\n/).map((l) => l.slice(0, 2e3));
+      const found = matching(lines);
+      if (!found) {
+        cut = true;
+        return false;
+      }
+      for (const i of found) {
+        const line = lines[i];
         if (hits.length >= limit) {
           more = true;
           return false;
@@ -3707,7 +3664,7 @@ async function grepTool(root, input, o) {
     `${plural(hits.length, "hit")} for "${pattern}"${where}${scope} (${plural(read, "file")} read):`,
     ...hits,
     ...more ? [`\u2026 more hits past the ${limit} shown; narrow the pattern or the paths.`] : [],
-    ...cut ? [`\u2026 the search stopped at its cap of files or bytes; name a repo or narrower paths.`] : [],
+    ...cut ? [`\u2026 the search stopped at its cap of files, bytes or time; name a repo or narrower paths.`] : [],
     "Show the lines around a hit with show_code (repo, path, line). A hit is a pointer, not a fact: a fact cites its file and line, and Atticus checks the value there."
   ].join("\n");
 }
@@ -3717,7 +3674,7 @@ async function filesTool(root, input, o) {
   if (!glob2)
     throw new ContextToolError("files needs a glob, for example src/**/*.java.");
   const limit = int(input.limit, 1, B.filesLimit, B.filesDefault);
-  const re = globRe(glob2);
+  const re = globMatcher(glob2);
   const repos = reposOf(root, input, o);
   const out = [];
   let more = false;
@@ -3790,7 +3747,7 @@ async function outlineTool(root, input, o) {
   const size = (await stat2(file)).size;
   if (size > CONTEXT_TOOL_BOUNDS.codeFileBytes)
     throw new ContextToolError(`${path} is too big to outline (${size} bytes).`);
-  const text = await readFile3(file, "utf8");
+  const text = decodeText(await readFile3(file));
   const head = `Outline of ${repo}:${rel2}`;
   let how;
   let lines;