testdriverai 7.8.0 → 7.9.0-test.1

package/lib/github-comment.mjs

@@ -49,7 +49,7 @@ function generateTestResultsTable(testCases, testRunUrl) {
 
   // Filter out skipped tests
   const nonSkippedTests = testCases.filter(test => test.status !== 'skipped');
-  
+
   if (nonSkippedTests.length === 0) {
     return '_No test cases to display (all tests were skipped)_';
   }
@@ -62,12 +62,12 @@ function generateTestResultsTable(testCases, testRunUrl) {
     const name = test.testName || 'Unknown';
     const file = test.testFile || 'unknown';
     const duration = formatDuration(test.duration || 0);
-    
+
     // Use test run context URL instead of direct replay URL
     let replay = '-';
     if (test.replayUrl) {
       const linkUrl = (test.id && testRunUrl) ? `${testRunUrl}/${test.id}` : test.replayUrl;
-      
+
       // Extract replay ID and generate GIF URL
       const replayId = extractReplayId(test.replayUrl);
       if (replayId) {
@@ -97,28 +97,28 @@ function generateTestResultsTable(testCases, testRunUrl) {
  */
 function generateExceptionsSection(testCases, testRunUrl) {
   const failedTests = testCases.filter(t => t.status === 'failed' && t.errorMessage);
-  
+
   if (failedTests.length === 0) {
     return '';
   }
 
   let section = '\n## 🔴 Failures\n\n';
-  
+
   for (const test of failedTests) {
     section += `### ${test.testName}\n\n`;
     section += `**File:** \`${test.testFile}\`\n\n`;
-    
+
     // Use test run context URL instead of direct replay URL
     if (test.id && testRunUrl) {
       section += `**📹 [Watch Replay](${testRunUrl}/${test.id})**\n\n`;
     } else if (test.replayUrl) {
       section += `**📹 [Watch Replay](${test.replayUrl})**\n\n`;
     }
-    
+
     section += '```\n';
     section += test.errorMessage || 'Unknown error';
     section += '\n```\n\n';
-    
+
     if (test.errorStack) {
       section += '<details>\n';
       section += '<summary>Stack Trace</summary>\n\n';
@@ -140,22 +140,22 @@ function generateExceptionsSection(testCases, testRunUrl) {
  */
 function generateReplaySection(testCases, testRunUrl) {
   const testsWithReplays = testCases.filter(t => t.replayUrl);
-  
+
   if (testsWithReplays.length === 0) {
     return '';
   }
 
   let section = '\n## 🎥 Dashcam Replays\n\n';
-  
+
   for (const test of testsWithReplays) {
     section += `### ${test.testName}\n\n`;
-    
+
     // Determine the link URL - prefer test run context
     let linkUrl = test.replayUrl;
     if (test.id && testRunUrl) {
       linkUrl = `${testRunUrl}/${test.id}`;
     }
-    
+
     // Extract replay ID from URL for GIF embed
     const replayId = extractReplayId(test.replayUrl);
     if (replayId) {
@@ -177,7 +177,7 @@ function generateReplaySection(testCases, testRunUrl) {
  */
 function extractReplayId(url) {
   if (!url) return null;
-  
+
   // Match pattern: /replay/{id} or /replay/{id}?params
   const match = url.match(/\/replay\/([^?/#]+)/);
   return match ? match[1] : null;
@@ -190,7 +190,7 @@ function extractReplayId(url) {
  */
 function extractShareKey(url) {
   if (!url) return null;
-  
+
   // Match pattern: ?share=KEY or &share=KEY
   const match = url.match(/[?&]share=([^&#]+)/);
   return match ? match[1] : null;
@@ -204,32 +204,50 @@ function extractShareKey(url) {
  */
 function getReplayGifUrl(replayUrl, replayId) {
   // Determine the API base URL based on the replay URL
+  // Replay URLs use console domains; GIF endpoints live on the corresponding API domain
   let apiBaseUrl;
-  
+
   if (replayUrl.includes('app.dashcam.io')) {
     // Production dashcam uses Heroku API
     apiBaseUrl = 'https://testdriverai-v6-c96fc597be11.herokuapp.com';
-  } else if (replayUrl.includes('console.testdriver.ai')) {
-    // TestDriver console
-    apiBaseUrl = 'https://api.testdriver.ai';
   } else if (replayUrl.includes('localhost')) {
     // Local development
     apiBaseUrl = 'http://localhost:1337';
   } else {
-    // Default: try to extract base URL from replay URL
-    const urlObj = new URL(replayUrl);
-    apiBaseUrl = `${urlObj.protocol}//${urlObj.host}`;
+    // Map console URLs → API URLs for all environments
+    // console-test.testdriver.ai → api-test.testdriver.ai
+    // console-canary.testdriver.ai → api-canary.testdriver.ai
+    // console.testdriver.ai → api.testdriver.ai
+    const consoleEnvMatch = replayUrl.match(/https:\/\/console-(test|canary)\.testdriver\.ai/);
+    if (consoleEnvMatch) {
+      apiBaseUrl = `https://api-${consoleEnvMatch[1]}.testdriver.ai`;
+    } else if (replayUrl.includes('console.testdriver.ai')) {
+      apiBaseUrl = 'https://api.testdriver.ai';
+    }
+    // Fly.io: map web app → API app
+    // pr-123-web.fly.dev → pr-123-api.fly.dev
+    // td-test-web.fly.dev → td-test-api.fly.dev
+    else {
+      const flyWebMatch = replayUrl.match(/https:\/\/([\w-]+)-web\.fly\.dev/);
+      if (flyWebMatch) {
+        apiBaseUrl = `https://${flyWebMatch[1]}-api.fly.dev`;
+      } else {
+        // Fallback: extract base URL from replay URL as-is
+        const urlObj = new URL(replayUrl);
+        apiBaseUrl = `${urlObj.protocol}//${urlObj.host}`;
+      }
+    }
   }
-  
+
   // Extract share key if present
   const shareKey = extractShareKey(replayUrl);
-  
+
   // Build GIF URL with shareKey parameter
   let gifUrl = `${apiBaseUrl}/replay/${replayId}/gif`;
   if (shareKey) {
     gifUrl += `?shareKey=${shareKey}`;
   }
-  
+
   return gifUrl;
 }
 
@@ -257,9 +275,9 @@ export function generateGitHubComment(testRunData, testCases = []) {
   // Header with overall status
   const statusEmoji = getStatusEmoji(status);
   const statusColor = status === 'passed' ? '🟢' : status === 'failed' ? '🔴' : '🟡';
-  
+
   let comment = `# ${statusColor} TestDriver Test Results\n\n`;
-  
+
   // Compact summary line
   comment += `**Status:** ${statusEmoji} ${status.toUpperCase()}`;
   comment += ` • **Duration:** ${formatDuration(duration)}`;
@@ -270,23 +288,23 @@ export function generateGitHubComment(testRunData, testCases = []) {
     comment += `, ${skippedTests} skipped`;
   }
   comment += `\n\n`;
-  
+
   // Exceptions section (only if there are failures) - show first
   comment += generateExceptionsSection(testCases, testRunUrl);
-  
+
   // Test results table (now includes embedded GIFs)
   comment += '## 📝 Test Results\n\n';
   comment += generateTestResultsTable(testCases, testRunUrl);
-  
+
   // Link to full test run (below table)
   if (testRunUrl) {
     comment += `\n[📋 View Full Test Run](${testRunUrl})\n`;
   }
-  
+
   // Footer
   comment += '\n---\n';
   comment += `<sub>Generated by [TestDriver](https://testdriver.ai) • Run ID: \`${runId}\`</sub>\n`;
-  
+
   return comment;
 }
 
@@ -303,15 +321,15 @@ export function generateGitHubComment(testRunData, testCases = []) {
  */
 export async function postGitHubComment(options) {
   const { token, owner, repo, prNumber, commitSha, body } = options;
-  
+
   if (!token) {
     throw new Error('GitHub token is required');
   }
-  
+
   if (!owner || !repo) {
     throw new Error('Repository owner and name are required');
   }
-  
+
   if (!prNumber && !commitSha) {
     throw new Error('Either prNumber or commitSha must be provided');
   }
@@ -351,7 +369,7 @@ export async function postGitHubComment(options) {
  */
 export async function updateGitHubComment(options) {
   const { token, owner, repo, commentId, body } = options;
-  
+
   if (!token || !owner || !repo || !commentId) {
     throw new Error('Token, owner, repo, and commentId are required');
   }
@@ -364,7 +382,7 @@ export async function updateGitHubComment(options) {
     comment_id: commentId,
     body,
   });
-  
+
   return response.data;
 }
 
@@ -379,7 +397,7 @@ export async function updateGitHubComment(options) {
  */
 export async function findExistingComment(options) {
   const { token, owner, repo, prNumber } = options;
-  
+
   if (!token || !owner || !repo || !prNumber) {
     return null;
   }
@@ -410,11 +428,11 @@ export async function findExistingComment(options) {
  */
 export async function postOrUpdateTestResults(testRunData, testCases, githubOptions) {
   const commentBody = generateGitHubComment(testRunData, testCases);
-  
+
   // Try to find and delete existing comment to keep it at the end
   if (githubOptions.prNumber) {
     const existingComment = await findExistingComment(githubOptions);
-    
+
     if (existingComment) {
       // Delete the old comment
       const octokit = new Octokit({ auth: githubOptions.token });
@@ -425,7 +443,7 @@ export async function postOrUpdateTestResults(testRunData, testCases, githubOpti
       });
     }
   }
-  
+
   // Always create a new comment (will be at the end of the thread)
   return await postGitHubComment({
     ...githubOptions,

package/lib/init-project.js

@@ -14,66 +14,11 @@ function runInstall(cmd, args, cwd, label) {
   return new Promise((resolve, reject) => {
     const child = spawn(cmd, args, {
       cwd,
-      stdio: ["ignore", "pipe", "pipe"],
+      stdio: ["ignore", "inherit", "inherit"],
       shell: process.platform === "win32",
     });
 
-    const spinnerFrames = ["⠋", "⠙", "⠹", "⠸", "⠼", "⠴", "⠦", "⠧", "⠇", "⠏"];
-    const barWidth = 20;
-    let frame = 0;
-    let status = "resolving";
-    let filled = 0;
-
-    // Parse npm stderr for progress hints
-    const handleData = (data) => {
-      const text = data.toString();
-      if (text.includes("idealTree")) {
-        status = "resolving packages";
-        filled = Math.max(filled, 3);
-      } else if (text.includes("reify:")) {
-        status = "installing";
-        filled = Math.max(filled, 8);
-        // Try to extract package name from reify output
-        const match = text.match(/reify:([^\s:]+)/);
-        if (match) {
-          status = `installing ${match[1]}`;
-        }
-      } else if (text.includes("timing")) {
-        filled = Math.max(filled, 14);
-        status = "finalizing";
-      } else if (text.includes("added")) {
-        filled = barWidth;
-        status = "done";
-      }
-      // Slowly increment to show activity
-      if (filled < barWidth - 2) {
-        filled = Math.min(filled + 1, barWidth - 2);
-      }
-    };
-
-    child.stdout.on("data", handleData);
-    child.stderr.on("data", handleData);
-
-    const isTTY = process.stderr.isTTY;
-
-    const interval = setInterval(() => {
-      frame = (frame + 1) % spinnerFrames.length;
-      const spinner = spinnerFrames[frame];
-      const bar = "█".repeat(filled) + "░".repeat(barWidth - filled);
-      const line = `  ${spinner} ${label} [${bar}] ${status}`;
-      if (isTTY) {
-        process.stderr.clearLine(0);
-        process.stderr.cursorTo(0);
-        process.stderr.write(line);
-      }
-    }, 80);
-
     child.on("close", (code) => {
-      clearInterval(interval);
-      if (isTTY) {
-        process.stderr.clearLine(0);
-        process.stderr.cursorTo(0);
-      }
       if (code === 0) {
         resolve();
       } else {
@@ -82,11 +27,6 @@ function runInstall(cmd, args, cwd, label) {
     });
 
     child.on("error", (err) => {
-      clearInterval(interval);
-      if (isTTY) {
-        process.stderr.clearLine(0);
-        process.stderr.cursorTo(0);
-      }
       reject(err);
     });
   });
@@ -384,11 +324,9 @@ jobs:
           [
             "--yes",
             "add-mcp",
+            "testdriverai",
+            "-n",
             "testdriver",
-            "--command",
-            "npx -p testdriverai testdriverai-mcp",
-            "--env",
-            "TD_API_KEY",
           ],
           {
             cwd: targetDir,
@@ -400,10 +338,10 @@ jobs:
         if (addMcpResult.status === 0) {
           progress("✓ MCP configured via add-mcp");
         } else if (addMcpResult.status !== null) {
-          progress("⚠ MCP setup skipped or failed - you can run 'npx add-mcp testdriver' later");
+          progress("⚠ MCP setup skipped or failed - you can run 'npx add-mcp testdriverai' later");
         }
       } catch (err) {
-        progress("⚠ Could not run add-mcp - you can run 'npx add-mcp testdriver' later");
+        progress("⚠ Could not run add-mcp - you can run 'npx add-mcp testdriverai' later");
       }
     }
 

package/lib/resolve-channel.js

@@ -1,11 +1,14 @@
 /**
  * Resolves the active release channel and API URLs.
  *
+ * TD_CHANNEL: dev | test | canary | stable   (which release channel)
+ * TD_ENV:     dev | staging | production     (which infrastructure tier)
+ *
  * Channel is derived from (in priority order):
  *   1. TD_CHANNEL env var (explicit override)
- *   2. TD_ENV env var (set by envs/<name>.env)
+ *   2. TD_ENV env var — if it holds a legacy channel name (dev/test/canary/stable)
  *   3. SDK package.json version prerelease tag (e.g. "7.6.0-test.5" → "test")
- *   4. "latest" for clean semver versions (stable releases)
+ *   4. "stable" for clean semver versions
  */
 
 const semver = require("semver");
@@ -15,20 +18,20 @@ const CHANNELS = {
   dev: "http://localhost:1337",
   test: environments.test.apiRoot,
   canary: environments.canary.apiRoot,
-  latest: environments.stable.apiRoot,
+  stable: environments.stable.apiRoot,
 };
 
+const VALID_ENVS = new Set(["dev", "staging", "production"]);
+
 function resolveActiveChannel() {
   // 1. Explicit channel override
   if (process.env.TD_CHANNEL && CHANNELS[process.env.TD_CHANNEL]) {
     return process.env.TD_CHANNEL;
   }
 
-  // 2. Environment name from env file (mapped: stable → latest)
-  if (process.env.TD_ENV) {
-    const envName = process.env.TD_ENV;
-    if (CHANNELS[envName]) return envName;
-    if (envName === "stable") return "latest";
+  // 2. TD_ENV — if it holds a legacy channel name, use it as channel
+  if (process.env.TD_ENV && CHANNELS[process.env.TD_ENV]) {
+    return process.env.TD_ENV;
   }
 
   // 3. Fallback: derive from package.json prerelease tag
@@ -38,9 +41,35 @@ function resolveActiveChannel() {
     return pre[0];
   }
 
-  return "latest";
+  return "stable";
 }
 
 const active = resolveActiveChannel();
 
-module.exports = { active, channels: CHANNELS };
+/**
+ * Returns the infrastructure environment (dev | staging | production).
+ * Reads from environments.json tdEnv mapping.
+ */
+function resolveTdEnv() {
+  // If TD_ENV is already new-format, use it directly
+  if (process.env.TD_ENV && VALID_ENVS.has(process.env.TD_ENV)) {
+    return process.env.TD_ENV;
+  }
+  // Derive from channel
+  const entry = environments[active];
+  return entry ? entry.tdEnv : "production";
+}
+
+const tdEnv = resolveTdEnv();
+
+/**
+ * Resolves the Sentry environment name.
+ * Uses the infrastructure tier (dev | staging | production).
+ */
+function resolveSentryEnvironment() {
+  return tdEnv;
+}
+
+const sentryEnvironment = resolveSentryEnvironment();
+
+module.exports = { active, channels: CHANNELS, sentryEnvironment, tdEnv };

package/lib/sentry.js

@@ -14,6 +14,7 @@ const Sentry = require("@sentry/node");
 const crypto = require("crypto");
 const os = require("os");
 const { version } = require("../package.json");
+const { sentryEnvironment } = require("./resolve-channel");
 const logger = require("../agent/lib/logger");
 
 // Store trace contexts per session so concurrent tests don't overwrite each other.
@@ -41,7 +42,7 @@ if (isEnabled()) {
     dsn:
       process.env.SENTRY_DSN ||
       "https://452bd5a00dbd83a38ee8813e11c57694@o4510262629236736.ingest.us.sentry.io/4510480443637760",
-    environment: "sdk",
+    environment: sentryEnvironment,
     release: version,
     sampleRate: 1.0,
     tracesSampleRate: 1.0, // Sample 20% of transactions for performance
@@ -53,6 +54,7 @@ if (isEnabled()) {
         platform: os.platform(),
         arch: os.arch(),
         nodeVersion: process.version,
+        runner: "sdk",
       },
     },
     // Filter out common non-errors and expected test failures
@@ -85,6 +87,11 @@ if (isEnabled()) {
         return null;
       }
 
+      // Filter out ElementNotFoundError - expected test outcome, not a crash
+      if (error && error.name === "ElementNotFoundError") {
+        return null;
+      }
+
       return event;
     },
   });
@@ -173,18 +180,41 @@ function captureMessage(message, level = "info") {
 }
 
 /**
- * Set the session trace context for distributed tracing
- * This links CLI errors/logs to the same trace as API calls
+ * Set the session trace context for distributed tracing.
+ * Creates a root span (sdk.session) that appears in the Sentry trace waterfall
+ * and sets propagation context so all outbound headers (HTTP and Ably) reference
+ * the same deterministic traceId = MD5(sessionId).
+ *
  * @param {string} sessionId - The session ID
  */
 function setSessionTraceContext(sessionId) {
   if (!isEnabled() || !sessionId) return;
 
-  // Derive trace ID from session ID (same algorithm as API)
+  // Derive trace ID from session ID (same algorithm as API and Runner)
   const traceId = crypto.createHash("md5").update(sessionId).digest("hex");
 
-  // Store per-session trace context for concurrent safety
-  _traceContexts.set(sessionId, { traceId, sessionId });
+  // Build a synthetic sentry-trace header to seed the trace with our deterministic ID
+  const spanId = crypto.randomBytes(8).toString("hex");
+  const sentryTraceHeader = `${traceId}-${spanId}-1`;
+  const baggageHeader = `sentry-trace_id=${traceId},sentry-sampled=true`;
+
+  // continueTrace sets the scope's propagation context so all subsequent
+  // getTraceData() calls return our traceId. startInactiveSpan creates a
+  // root transaction that will be visible in the Sentry trace waterfall.
+  let rootSpan = null;
+  Sentry.continueTrace(
+    { sentryTrace: sentryTraceHeader, baggage: baggageHeader },
+    () => {
+      rootSpan = Sentry.startInactiveSpan({
+        name: "sdk.session",
+        op: "session",
+        forceTransaction: true,
+      });
+    },
+  );
+
+  // Store per-session trace context (including root span for cleanup)
+  _traceContexts.set(sessionId, { traceId, sessionId, rootSpan });
 
   // Also update the module-level "latest" for backward compatibility
   currentTraceId = traceId;
@@ -193,28 +223,17 @@ function setSessionTraceContext(sessionId) {
   // Set as global tag so all events include it
   Sentry.setTag("session", sessionId);
   Sentry.setTag("trace_id", currentTraceId);
-
-  // Try to set propagation context for trace linking (may not be available in all versions)
-  try {
-    const scope = Sentry.getCurrentScope();
-    if (scope && typeof scope.setPropagationContext === "function") {
-      scope.setPropagationContext({
-        traceId: currentTraceId,
-        spanId: currentTraceId.substring(0, 16),
-        sampled: true,
-      });
-    }
-  } catch (e) {
-    // Ignore errors - propagation context may not be supported
-    logger.log("Could not set propagation context:", e.message);
-  }
 }
 
 /**
- * Clear the session trace context
+ * Clear the session trace context and end the root session span.
  */
 function clearSessionTraceContext(sessionId) {
   if (sessionId) {
+    const ctx = _traceContexts.get(sessionId);
+    if (ctx && ctx.rootSpan) {
+      try { ctx.rootSpan.end(); } catch (e) { /* ignore */ }
+    }
     _traceContexts.delete(sessionId);
     // If the cleared session was the "latest", pick another or null
     if (currentSessionId === sessionId) {
@@ -228,7 +247,12 @@ function clearSessionTraceContext(sessionId) {
       }
     }
   } else {
-    // Clear all (backward compatibility)
+    // Clear all (backward compatibility) — end all root spans
+    for (const ctx of _traceContexts.values()) {
+      if (ctx.rootSpan) {
+        try { ctx.rootSpan.end(); } catch (e) { /* ignore */ }
+      }
+    }
     _traceContexts.clear();
     currentTraceId = null;
     currentSessionId = null;