agentv 4.7.0 → 4.8.0

package/dist/{chunk-AX4CQS45.js → chunk-QBZJSQXV.js}

@@ -32,12 +32,12 @@ import {
   subscribeToCopilotCliLogEntries,
   subscribeToCopilotSdkLogEntries,
   subscribeToPiLogEntries
-} from "./chunk-I6UE4LHZ.js";
+} from "./chunk-H4GQXK5M.js";
 
 // package.json
 var package_default = {
   name: "agentv",
-  version: "4.7.0",
+  version: "4.8.0",
   description: "CLI entry point for AgentV",
   type: "module",
   repository: {
@@ -1712,8 +1712,9 @@ var JunitWriter = class _JunitWriter {
         }
         return `    <testcase name="${escapeXml(r.testId)}" classname="${escapeXml(suiteName)}" time="${time}">${inner}</testcase>`;
       });
+      const suiteTime = results.reduce((sum, r) => sum + (r.durationMs ?? 0), 0) / 1e3;
       suiteXmls.push(
-        `  <testsuite name="${escapeXml(suiteName)}" tests="${results.length}" failures="${failures}" errors="${errors}">
+        `  <testsuite name="${escapeXml(suiteName)}" tests="${results.length}" failures="${failures}" errors="${errors}" time="${suiteTime.toFixed(3)}">
 ${testCases.join("\n")}
   </testsuite>`
       );
@@ -1723,8 +1724,9 @@ ${testCases.join("\n")}
     const totalFailures = this.results.filter(
       (r) => r.executionStatus !== "execution_error" && r.score < this.threshold
     ).length;
+    const totalTime = this.results.reduce((sum, r) => sum + (r.durationMs ?? 0), 0) / 1e3;
     const xml = `<?xml version="1.0" encoding="UTF-8"?>
-<testsuites tests="${totalTests}" failures="${totalFailures}" errors="${totalErrors}">
+<testsuites tests="${totalTests}" failures="${totalFailures}" errors="${totalErrors}" time="${totalTime.toFixed(3)}">
 ${suiteXmls.join("\n")}
 </testsuites>
 `;
@@ -1823,17 +1825,6 @@ function createWriterFromPath(filePath, options) {
       );
   }
 }
-async function createMultiWriter(filePaths, options) {
-  const writers = await Promise.all(filePaths.map((fp) => createWriterFromPath(fp, options)));
-  return {
-    async append(result) {
-      await Promise.all(writers.map((w) => w.append(result)));
-    },
-    async close() {
-      await Promise.all(writers.map((w) => w.close()));
-    }
-  };
-}
 
 // src/commands/eval/progress-display.ts
 var ANSI_BOLD = "\x1B[1m";
@@ -3141,7 +3132,8 @@ var COPILOT_SDK_SETTINGS = /* @__PURE__ */ new Set([
   "log_dir",
   "log_format",
   "system_prompt",
-  "workspace_template"
+  "workspace_template",
+  "byok"
 ]);
 var COPILOT_CLI_SETTINGS = /* @__PURE__ */ new Set([
   ...COMMON_SETTINGS,
@@ -4015,15 +4007,12 @@ function trimOutputMessages(output, outputMessages) {
   return sliced.map((m) => ({ role: m.role, content: m.content }));
 }
 function normalizeOptions(rawOptions, config, yamlExecution) {
-  const cliFormat = normalizeString(rawOptions.outputFormat);
-  const configFormat = config?.output?.format;
-  const formatStr = cliFormat ?? configFormat ?? "jsonl";
-  const format = formatStr === "yaml" ? "yaml" : "jsonl";
   const cliWorkers = normalizeOptionalNumber(rawOptions.workers);
   const configWorkers = config?.execution?.workers;
   const workers = cliWorkers ?? configWorkers ?? 0;
-  const rawOutputPaths = rawOptions.output;
-  const outputPaths = Array.isArray(rawOutputPaths) ? rawOutputPaths.filter((v) => typeof v === "string" && v.trim().length > 0) : [];
+  const cliOutputDir = normalizeString(rawOptions.output);
+  const rawExportPaths = rawOptions.export;
+  const exportPaths = Array.isArray(rawExportPaths) ? rawExportPaths.filter((v) => typeof v === "string" && v.trim().length > 0) : [];
   const rawTarget = rawOptions.target;
   let cliTargets = [];
   let singleTarget;
@@ -4065,9 +4054,9 @@ function normalizeOptions(rawOptions, config, yamlExecution) {
     targetsPath: normalizeString(rawOptions.targets),
     filter: normalizeFilter(rawOptions.filter),
     workers: workers > 0 ? workers : void 0,
+    outputDir: cliOutputDir,
     outPath: cliOut ?? configOut,
-    outputPaths,
-    format,
+    exportPaths,
     dryRun: normalizeBoolean(rawOptions.dryRun),
     dryRunDelay: normalizeNumber(rawOptions.dryRunDelay, 0),
     dryRunDelayMin: normalizeNumber(rawOptions.dryRunDelayMin, 0),
@@ -4484,13 +4473,48 @@ async function runEvalCommand(input) {
   if (options.verbose) {
     console.log(`Repository root: ${repoRoot}`);
   }
-  const usesDefaultArtifactWorkspace = !options.outPath;
-  const outputPath = options.outPath ? path15.resolve(options.outPath) : buildDefaultOutputPath(cwd);
+  if (options.outPath) {
+    console.warn("Warning: --out is deprecated. Use --output <dir> to set the artifact directory.");
+  }
+  if (options.artifacts) {
+    console.warn(
+      "Warning: --artifacts is deprecated. Use --output <dir> to set the artifact directory."
+    );
+  }
+  if (options.benchmarkJson) {
+    console.warn(
+      "Warning: --benchmark-json is deprecated. benchmark.json is always written to the artifact directory."
+    );
+  }
+  if (normalizeString(input.rawOptions.outputFormat)) {
+    console.warn(
+      "Warning: --output-format is deprecated. The artifact directory always uses JSONL."
+    );
+  }
+  const explicitDir = options.outputDir ?? options.artifacts;
+  let runDir;
+  let outputPath;
+  let usesDefaultArtifactWorkspace;
+  if (explicitDir) {
+    runDir = path15.resolve(explicitDir);
+    mkdirSync(runDir, { recursive: true });
+    outputPath = path15.join(runDir, "index.jsonl");
+    usesDefaultArtifactWorkspace = true;
+  } else if (options.outPath) {
+    outputPath = path15.resolve(options.outPath);
+    runDir = path15.dirname(outputPath);
+    mkdirSync(runDir, { recursive: true });
+    usesDefaultArtifactWorkspace = false;
+  } else {
+    outputPath = buildDefaultOutputPath(cwd);
+    runDir = path15.dirname(outputPath);
+    usesDefaultArtifactWorkspace = true;
+  }
   let otelExporter = null;
   const useFileExport = !!options.otelFile;
   if (options.exportOtel || useFileExport) {
     try {
-      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-XRVHRBJF.js");
+      const { OtelTraceExporter, OTEL_BACKEND_PRESETS } = await import("./dist-QXVR2ZRH.js");
       let endpoint = process.env.OTEL_EXPORTER_OTLP_ENDPOINT;
       let headers = {};
       if (options.otelBackend) {
@@ -4531,16 +4555,11 @@ async function runEvalCommand(input) {
     }
   }
   const primaryWritePath = outputPath;
-  const extraOutputPaths = options.outputPaths.map((p) => path15.resolve(p));
-  const allOutputPaths = extraOutputPaths.length > 0 ? [primaryWritePath, ...extraOutputPaths] : [primaryWritePath];
-  const uniqueOutputPaths = [...new Set(allOutputPaths)];
-  const reportedOutputPaths = extraOutputPaths.length > 0 ? [outputPath, ...extraOutputPaths] : [outputPath];
-  const uniqueReportedOutputPaths = [...new Set(reportedOutputPaths)];
-  if (uniqueOutputPaths.length === 1) {
-    console.log(`Output path: ${outputPath}`);
-  } else {
-    console.log("Output paths:");
-    for (const p of uniqueReportedOutputPaths) {
+  const resolvedExportPaths = options.exportPaths.map((p) => path15.resolve(p));
+  console.log(`Artifact directory: ${runDir}`);
+  if (resolvedExportPaths.length > 0) {
+    console.log("Export files:");
+    for (const p of resolvedExportPaths) {
       console.log(`  ${p}`);
     }
   }
@@ -4621,12 +4640,7 @@ async function runEvalCommand(input) {
     throw new Error("--threshold must be between 0 and 1");
   }
   const writerOptions = resolvedThreshold !== void 0 ? { threshold: resolvedThreshold } : void 0;
-  let outputWriter;
-  if (uniqueOutputPaths.length === 1) {
-    outputWriter = await createOutputWriter(primaryWritePath, options.format);
-  } else {
-    outputWriter = await createMultiWriter(uniqueOutputPaths, writerOptions);
-  }
+  const outputWriter = await createOutputWriter(primaryWritePath, "jsonl");
   const isMatrixMode = Array.from(fileMetadata.values()).some((meta) => meta.selections.length > 1);
   let totalEvalCount = 0;
   for (const meta of fileMetadata.values()) {
@@ -4694,7 +4708,7 @@ async function runEvalCommand(input) {
   const activeTestFiles = resolvedTestFiles.filter((f) => fileMetadata.has(f));
   let transcriptProviderFactory;
   if (options.transcript) {
-    const { TranscriptProvider } = await import("./dist-XRVHRBJF.js");
+    const { TranscriptProvider } = await import("./dist-QXVR2ZRH.js");
     const transcriptProvider = await TranscriptProvider.fromFile(options.transcript);
     const totalTests = [...fileMetadata.values()].reduce(
       (sum, meta) => sum + meta.testCases.length,
@@ -4808,18 +4822,17 @@ async function runEvalCommand(input) {
       await writeBenchmarkJson(benchmarkPath, allResults);
       console.log(`Benchmark written to: ${benchmarkPath}`);
     }
-    if (usesDefaultArtifactWorkspace) {
+    if (usesDefaultArtifactWorkspace && allResults.length > 0) {
       const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
-      const workspaceDir = path15.dirname(outputPath);
       const {
         testArtifactDir,
         timingPath,
         benchmarkPath: workspaceBenchmarkPath,
         indexPath
-      } = await writeArtifactsFromResults(allResults, workspaceDir, {
+      } = await writeArtifactsFromResults(allResults, runDir, {
         evalFile
       });
-      console.log(`Artifact workspace written to: ${workspaceDir}`);
+      console.log(`Artifact workspace written to: ${runDir}`);
       console.log(`  Index: ${indexPath}`);
       console.log(
         `  Per-test artifacts: ${testArtifactDir} (${allResults.length} test directories)`
@@ -4827,24 +4840,17 @@ async function runEvalCommand(input) {
       console.log(`  Timing: ${timingPath}`);
       console.log(`  Benchmark: ${workspaceBenchmarkPath}`);
     }
-    if (options.artifacts) {
-      const artifactsDir = path15.resolve(options.artifacts);
-      const evalFile = activeTestFiles.length === 1 ? activeTestFiles[0] : "";
-      const {
-        testArtifactDir,
-        indexPath,
-        timingPath,
-        benchmarkPath: abp
-      } = await writeArtifactsFromResults(allResults, artifactsDir, {
-        evalFile
-      });
-      console.log(`Artifacts written to: ${artifactsDir}`);
-      console.log(`  Index: ${indexPath}`);
+    if (resolvedExportPaths.length > 0 && allResults.length > 0) {
+      for (const exportPath of resolvedExportPaths) {
+        const writer = await createWriterFromPath(exportPath, writerOptions);
+        for (const result of allResults) {
+          await writer.append(result);
+        }
+        await writer.close();
+      }
       console.log(
-        `  Per-test artifacts: ${testArtifactDir} (${allResults.length} test directories)`
+        `Export file(s) written: ${resolvedExportPaths.map((p) => path15.relative(cwd, p)).join(", ")}`
       );
-      console.log(`  Timing:  ${timingPath}`);
-      console.log(`  Benchmark: ${abp}`);
     }
     const failedWithWorkspaces = allResults.filter(
       (r) => r.workspacePath && (r.error || r.score < 0.5)
@@ -4856,15 +4862,8 @@ async function runEvalCommand(input) {
       }
     }
     if (allResults.length > 0) {
-      if (uniqueReportedOutputPaths.length === 1) {
-        console.log(`
+      console.log(`
 Results written to: ${outputPath}`);
-      } else {
-        console.log("\nResults written to:");
-        for (const p of uniqueReportedOutputPaths) {
-          console.log(`  ${p}`);
-        }
-      }
       await saveRunCache(cwd, outputPath).catch(() => void 0);
     }
     if (summary.executionErrorCount > 0 && !options.retryErrors) {
@@ -4985,4 +4984,4 @@ export {
   getCategories,
   filterByCategory
 };
-//# sourceMappingURL=chunk-AX4CQS45.js.map
+//# sourceMappingURL=chunk-QBZJSQXV.js.map