@m4trix/evals 0.15.0 → 0.17.0

package/dist/cli.js

@@ -11,7 +11,7 @@ import { existsSync } from 'fs';
 import * as jitiModule from 'jiti';
 import { readdir, readFile, mkdir, appendFile } from 'fs/promises';
 import { pathToFileURL } from 'url';
-import { diffLines } from 'diff';
+import { diffString } from 'json-diff';
 
 var SEP = "  ";
 var ARROW = "\u203A";
@@ -978,45 +978,46 @@ async function collectTestCasesFromFiles(config) {
   );
   return found.flat();
 }
-function toJsonLines(value) {
+function createDiffString(expected, actual, diffOptions) {
+  const opts = { ...diffOptions, color: false };
+  const result = diffString(expected, actual, opts);
+  return typeof result === "string" ? result : "";
+}
+function formatLogMessage(msg) {
+  if (typeof msg === "string")
+    return msg;
   try {
-    return JSON.stringify(value, null, 2);
+    if (msg !== null && typeof msg === "object") {
+      return JSON.stringify(msg, null, 2);
+    }
+    return String(msg);
   } catch {
-    return String(value);
+    return String(msg);
   }
 }
-function formatDiffString(changes) {
-  const lines = [];
-  for (const part of changes) {
-    const prefix = part.added ? "+" : part.removed ? "-" : " ";
-    const partLines = part.value.split("\n");
-    if (partLines[partLines.length - 1] === "") {
-      partLines.pop();
-    }
-    for (const line of partLines) {
-      lines.push(`${prefix} ${line}`);
-    }
-  }
-  return lines.join("\n");
+function createLogEntry(message, options) {
+  return {
+    type: "log",
+    label: options?.label,
+    message: formatLogMessage(message)
+  };
 }
-function createDiffString(expected, actual) {
-  const expectedStr = toJsonLines(expected);
-  const actualStr = toJsonLines(actual);
-  const changes = diffLines(expectedStr, actualStr);
-  return formatDiffString(changes);
+function getLogLines(entry) {
+  return entry.message.split("\n");
 }
 function createDiffLogEntry(expected, actual, options) {
-  const diff = createDiffString(expected, actual);
+  const { label, ...diffOpts } = options ?? {};
+  const diff = createDiffString(expected, actual, diffOpts);
   return {
     type: "diff",
-    label: options?.label,
+    label,
     expected,
     actual,
     diff: diff || "(no differences)"
   };
 }
 function getDiffLines(entry) {
-  const raw = createDiffString(entry.expected, entry.actual) || "(no differences)";
+  const raw = entry.diff || "(no differences)";
   return raw.split("\n").map((line) => {
     const trimmed = line.trimStart();
     if (trimmed.startsWith("-") && !trimmed.startsWith("---")) {
@@ -1260,6 +1261,7 @@ function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent,
     const reruns = typeof testCaseItem.testCase.getReruns === "function" ? testCaseItem.testCase.getReruns() : 1;
     const rerunPassed = [];
     for (let r = 0; r < reruns; r++) {
+      const evaluatorRunId = `run-${randomUUID()}`;
       const started = Date.now();
       const evaluatorScores = [];
       let testCaseError;
@@ -1274,6 +1276,9 @@ function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent,
           const logDiff = (expected, actual, options) => {
             logs.push(createDiffLogEntry(expected, actual, options));
           };
+          const log = (message, options) => {
+            logs.push(createLogEntry(message, options));
+          };
           const ctx = yield* Effect.promise(
             () => Promise.resolve(evaluator.resolveContext())
           );
@@ -1283,7 +1288,13 @@ function processOneTestCase(task, testCaseItem, totalEvaluations, publishEvent,
                 input: testCaseItem.testCase.getInput(),
                 ctx,
                 output,
-                logDiff
+                meta: {
+                  triggerId: task.triggerId,
+                  runId: evaluatorRunId,
+                  datasetId: task.datasetId
+                },
+                logDiff,
+                log
               })
             )
           );
@@ -1771,6 +1782,7 @@ var EffectRunner = class {
       (sum, tc) => sum + (typeof tc.testCase.getReruns === "function" ? tc.testCase.getReruns() : 1),
       0
     );
+    const triggerId = request.triggerId ?? `trg-${randomUUID()}`;
     const runId = `run-${randomUUID()}`;
     const artifactPath = createArtifactPath(
       this.config.artifactDirectory,
@@ -1812,6 +1824,7 @@ var EffectRunner = class {
     await Effect.runPromise(
       Queue.offer(this.runQueue, {
         runId,
+        triggerId,
         datasetId: request.datasetId,
         dataset: dataset.dataset,
         evaluators: selectedEvaluators,
@@ -2360,6 +2373,23 @@ function buildDetailRows(run, testCases, evaluatorNameById) {
                   )
                 );
               }
+            } else if (log.type === "log") {
+              const logLines = getLogLines(log);
+              for (let lineIdx = 0; lineIdx < logLines.length; lineIdx++) {
+                rows.push(
+                  /* @__PURE__ */ jsxs(
+                    Text,
+                    {
+                      color: "gray",
+                      children: [
+                        "      ",
+                        logLines[lineIdx]
+                      ]
+                    },
+                    `tc-${tc.testCaseId}-${item.evaluatorId}-${logIdx}-${lineIdx}`
+                  )
+                );
+              }
             }
           }
         }