@m4trix/evals 0.21.0 → 0.21.1

package/dist/cli-simple.js

@@ -6,7 +6,9 @@ import { resolve, relative, join, parse, dirname } from 'path';
 import * as jitiModule from 'jiti';
 import { writeFile, readdir, readFile, mkdir, appendFile } from 'fs/promises';
 import { pathToFileURL } from 'url';
-import { diffString } from 'json-diff';
+import { diffLines } from 'diff';
+import stringify from 'fast-json-stable-stringify';
+import { cpus } from 'os';
 import * as React2 from 'react';
 import React2__default, { useState, useEffect, useCallback } from 'react';
 import { render, Box, Text } from 'ink';
@@ -261,10 +263,102 @@ async function collectTestCasesFromFiles(config) {
   );
   return found.flat();
 }
+function preprocessForDiff(value, options) {
+  if (options?.sort && Array.isArray(value)) {
+    return [...value].sort((a, b) => {
+      const aStr = stringify(preprocessForDiff(a, options));
+      const bStr = stringify(preprocessForDiff(b, options));
+      return aStr.localeCompare(bStr);
+    }).map((item) => preprocessForDiff(item, options));
+  }
+  if (value !== null && typeof value === "object" && !Array.isArray(value) && options?.excludeKeys) {
+    const keys = Array.isArray(options.excludeKeys) ? options.excludeKeys : options.excludeKeys.split(",").map((k) => k.trim());
+    const filtered = {};
+    for (const [k, v] of Object.entries(value)) {
+      if (!keys.includes(k)) {
+        filtered[k] = preprocessForDiff(v, options);
+      }
+    }
+    return filtered;
+  }
+  if (value !== null && typeof value === "object" && !Array.isArray(value)) {
+    const result = {};
+    for (const [k, v] of Object.entries(value)) {
+      result[k] = preprocessForDiff(v, options);
+    }
+    return result;
+  }
+  if (typeof value === "number" && options?.precision !== void 0) {
+    return Number(value.toFixed(options.precision));
+  }
+  return value;
+}
+function toPrettyJson(value) {
+  const str = stringify(value);
+  try {
+    const parsed = JSON.parse(str);
+    return JSON.stringify(parsed, null, 2);
+  } catch {
+    return str;
+  }
+}
+function formatDiffParts(parts) {
+  const lines = [];
+  for (const part of parts) {
+    const prefix = part.added ? "+ " : part.removed ? "- " : "";
+    const partLines = part.value.split("\n");
+    for (let i = 0; i < partLines.length; i++) {
+      const line = partLines[i];
+      if (i === partLines.length - 1 && line === "")
+        continue;
+      lines.push(prefix + line);
+    }
+  }
+  return lines.join("\n");
+}
 function createDiffString(expected, actual, diffOptions) {
-  const opts = { ...diffOptions, color: false };
-  const result = diffString(expected, actual, opts);
-  return typeof result === "string" ? result : "";
+  const expectedProcessed = preprocessForDiff(expected, diffOptions);
+  const actualProcessed = preprocessForDiff(actual, diffOptions);
+  if (diffOptions?.keysOnly) {
+    const expectedKeys = JSON.stringify(
+      extractKeys(expectedProcessed),
+      null,
+      2
+    );
+    const actualKeys = JSON.stringify(
+      extractKeys(actualProcessed),
+      null,
+      2
+    );
+    const parts2 = diffLines(expectedKeys, actualKeys);
+    return formatDiffParts(parts2);
+  }
+  const expectedStr = toPrettyJson(expectedProcessed);
+  const actualStr = toPrettyJson(actualProcessed);
+  if (expectedStr === actualStr) {
+    return "";
+  }
+  const parts = diffLines(expectedStr, actualStr);
+  if (diffOptions?.outputNewOnly) {
+    const filtered = parts.filter(
+      (p) => p.added === true
+    );
+    return formatDiffParts(filtered);
+  }
+  return formatDiffParts(parts);
+}
+function extractKeys(value) {
+  if (value === null || typeof value !== "object") {
+    return "\xB7";
+  }
+  if (Array.isArray(value)) {
+    return value.map(extractKeys);
+  }
+  const result = {};
+  for (const [k, v] of Object.entries(value)) {
+    result[k] = extractKeys(v);
+  }
+  return result;
 }
 function formatLogMessage(msg) {
   if (typeof msg === "string")
@@ -1304,8 +1398,9 @@ var EffectRunner = class {
     );
   }
 };
-
-// src/cli-simple/args.ts
+function getDefaultConcurrency() {
+  return Math.max(1, cpus().length);
+}
 function parseSimpleCliArgs(argv) {
   const args = {
     help: false,
@@ -1332,6 +1427,14 @@ function parseSimpleCliArgs(argv) {
       index += 1;
       continue;
     }
+    if ((token === "--concurrency" || token === "-c") && argv[index + 1]) {
+      const n = parseInt(argv[index + 1], 10);
+      if (!Number.isNaN(n) && n >= 1) {
+        args.concurrency = n;
+      }
+      index += 1;
+      continue;
+    }
     args.unknownArgs.push(token);
   }
   return args;
@@ -1339,9 +1442,12 @@ function parseSimpleCliArgs(argv) {
 function getSimpleCliUsage() {
   return [
     "Usage:",
-    "  eval-agents-simple run --dataset <datasetName> --evaluator <name-or-pattern>",
+    "  eval-agents-simple run --dataset <datasetName> --evaluator <name-or-pattern> [--concurrency N]",
     "  eval-agents-simple generate --dataset <datasetName>",
     "",
+    "Options:",
+    "  --concurrency, -c N   Max concurrent test cases (default: CPU count). Use 1 for sequential.",
+    "",
     "Pattern examples for --evaluator:",
     "  score-evaluator       exact name (case-insensitive)",
     '  "*score*"             wildcard pattern',
@@ -1630,6 +1736,7 @@ function RunView({
   runner,
   datasetName,
   evaluatorPattern,
+  concurrency,
   onComplete
 }) {
   const [phase, setPhase] = useState(
@@ -1777,7 +1884,8 @@ function RunView({
     });
     const snapshot = await runner.runDatasetWith({
       datasetId: dataset.id,
-      evaluatorIds: evaluators.map((item) => item.id)
+      evaluatorIds: evaluators.map((item) => item.id),
+      concurrency
     });
     setRunInfo({
       runId: snapshot.runId,
@@ -1805,7 +1913,7 @@ function RunView({
     });
     setPhase("completed");
     setTimeout(() => onComplete(), 200);
-  }, [runner, datasetName, evaluatorPattern, onComplete]);
+  }, [runner, datasetName, evaluatorPattern, concurrency, onComplete]);
   useEffect(() => {
     void runEval();
   }, [runEval]);
@@ -1848,22 +1956,30 @@ function RunView({
           label: `Evaluations ${completedEvaluations}/${runInfo?.totalTestCases ?? 0} completed \u2022 ${startedEvaluations}/${runInfo?.totalTestCases ?? 0} started`
         }
       ),
-      runningEvaluations.length > 0 && /* @__PURE__ */ jsx(Box, { flexDirection: "column", marginTop: 1, children: runningEvaluations.map((item) => /* @__PURE__ */ jsxs(Text, { color: "yellow", children: [
-        "[running ",
-        item.startedTestCases,
-        "/",
-        item.totalTestCases,
-        "] ",
-        item.name,
-        " ",
-        /* @__PURE__ */ jsxs(Text, { color: "gray", children: [
-          "(",
-          item.rerunIndex,
-          "/",
-          item.rerunTotal,
-          ")"
-        ] })
-      ] }, `${item.testCaseId}:${item.rerunIndex}`)) })
+      runningEvaluations.length > 0 && /* @__PURE__ */ jsx(Box, { flexDirection: "column", marginTop: 1, children: runningEvaluations.map((item) => /* @__PURE__ */ jsxs(
+        Text,
+        {
+          color: "yellow",
+          children: [
+            "[running ",
+            item.startedTestCases,
+            "/",
+            item.totalTestCases,
+            "]",
+            " ",
+            item.name,
+            " ",
+            /* @__PURE__ */ jsxs(Text, { color: "gray", children: [
+              "(",
+              item.rerunIndex,
+              "/",
+              item.rerunTotal,
+              ")"
+            ] })
+          ]
+        },
+        `${item.testCaseId}:${item.rerunIndex}`
+      )) })
     ] }),
     testCases.length > 0 && /* @__PURE__ */ jsx(Box, { flexDirection: "column", marginBottom: 1, children: testCases.map((tc) => /* @__PURE__ */ jsxs(Box, { flexDirection: "column", marginBottom: 0, children: [
       /* @__PURE__ */ jsxs(Text, { children: [
@@ -1945,7 +2061,7 @@ function RunView({
                 },
                 `${item.evaluatorId}-${s.id}-${idx}`
               );
-            }) : /* @__PURE__ */ jsx(Text, { color: "gray", children: "      n/a" }),
+            }) : /* @__PURE__ */ jsx(Text, { color: "gray", children: " n/a" }),
             !item.passed && item.logs && item.logs.length > 0 && /* @__PURE__ */ jsx(Box, { marginLeft: 2, flexDirection: "column", children: item.logs.map(
               (log, logIdx) => log.type === "diff" ? /* @__PURE__ */ jsx(Box, { flexDirection: "column", children: getDiffLines(log).map(
                 ({ type, line }, lineIdx) => /* @__PURE__ */ jsx(
@@ -2003,9 +2119,9 @@ function RunView({
         /* @__PURE__ */ jsx(Text, { color: "magenta", children: "evaluator averages" }),
         Array.from(evaluatorNameById.entries()).map(([id, name]) => {
           const agg = summary.aggregates.get(id);
-          const scoreKeys = [...summary.scoreItemsByEvaluatorScore?.keys() ?? []].filter(
-            (k) => k.startsWith(`${id}:`)
-          );
+          const scoreKeys = [
+            ...summary.scoreItemsByEvaluatorScore?.keys() ?? []
+          ].filter((k) => k.startsWith(`${id}:`));
           if (scoreKeys.length === 0) {
             return /* @__PURE__ */ jsxs(Text, { color: "gray", children: [
               "- ",
@@ -2313,7 +2429,7 @@ function formatEvaluatorScoreLine(name, scores, passed, metrics, options) {
   }
   return lines;
 }
-async function runSimpleEvalCommandPlain(runner, datasetName, evaluatorPattern) {
+async function runSimpleEvalCommandPlain(runner, datasetName, evaluatorPattern, concurrency) {
   const dataset = await runner.resolveDatasetByName(datasetName);
   if (!dataset) {
     const known = await runner.collectDatasets();
@@ -2503,7 +2619,8 @@ async function runSimpleEvalCommandPlain(runner, datasetName, evaluatorPattern)
   });
   const snapshot = await runner.runDatasetWith({
     datasetId: dataset.id,
-    evaluatorIds: evaluators.map((item) => item.id)
+    evaluatorIds: evaluators.map((item) => item.id),
+    concurrency
   });
   totalCount = snapshot.totalTestCases;
   console.log(colorize("=== Eval Run Started ===", `${ansi2.bold}${ansi2.cyan}`));
@@ -2592,13 +2709,14 @@ async function runSimpleEvalCommandPlain(runner, datasetName, evaluatorPattern)
   }
   console.log(`- artifact: ${colorize(completed.artifactPath, ansi2.dim)}`);
 }
-async function runSimpleEvalCommandInk(runner, datasetName, evaluatorPattern) {
+async function runSimpleEvalCommandInk(runner, datasetName, evaluatorPattern, concurrency) {
   return new Promise((resolve5, reject) => {
     const app = render(
       React2.createElement(RunView, {
         runner,
         datasetName,
         evaluatorPattern,
+        concurrency,
         onComplete: (err) => {
           app.unmount();
           if (err) {
@@ -2645,10 +2763,12 @@ async function main() {
   const runner = createRunner();
   try {
     if (args.command === "run") {
+      const concurrency = args.concurrency ?? getDefaultConcurrency();
       await (useInk ? runSimpleEvalCommandInk : runSimpleEvalCommandPlain)(
         runner,
         args.datasetName,
-        args.evaluatorPattern
+        args.evaluatorPattern,
+        concurrency
       );
       return;
     }