ci(e2e): keep Playwright traces and outputs from bazel-remote-gates as a run artifact

.bazelrc

@@ -22,6 +22,11 @@ test --test_timeout=120,600,1800,3600
 # GF_RBE_CHROMIUM_EXECUTABLE). No hardcoded value: an absent var lets
 # the run-*-bazel.mjs candidate scan handle system chromium paths.
 test --test_env=GF_RBE_CHROMIUM_EXECUTABLE
+test --test_env=PLAYWRIGHT_TRACE_MODE
+# Undeclared test outputs (Playwright traces/screenshots) are zipped into a single
+# test.outputs/outputs.zip; Bazel 8's unzipped tree-artifact default refused the
+# directory (spec names carry non-ASCII) on run 33203042020.
+test --zip_undeclared_test_outputs
 
 build:local --disk_cache=~/.cache/bazel-jesssullivan-github-io
 

.github/workflows/ci.yml

@@ -53,6 +53,9 @@ jobs:
       GF_BAZEL_HOST_JVM_MAX_HEAP_MB: '2560'
       GF_REAPI_TOKEN_EXCHANGE_REQUEST: ${{ github.event_name == 'push' && github.ref == 'refs/heads/main' && 'cache-write' || 'cache-read' }}
       GF_REAPI_TOKEN_EXCHANGE_TTL: 45m
+      # Keep Playwright traces for failed specs; the config routes them into
+      # Bazel's undeclared outputs so the evidence step below can upload them.
+      PLAYWRIGHT_TRACE_MODE: retain-on-failure
       # The GF consumer registry records this repo as shared-cache-backed.
       # Clear any executor endpoint inherited from the generic ARC pod.
       BAZEL_REMOTE_EXECUTOR: ''
@@ -155,6 +158,30 @@ jobs:
           npm run remote:e2e:public
           bash scripts/bazel-cache-backed.sh shutdown
 
+      - name: Collect e2e evidence
+        if: always()
+        run: |
+          set -euo pipefail
+          mkdir -p artifacts/playwright-e2e
+          if [[ -d bazel-testlogs/playwright_chromium_e2e ]]; then
+            cp -RL bazel-testlogs/playwright_chromium_e2e/. artifacts/playwright-e2e/
+            if [[ -f artifacts/playwright-e2e/test.outputs/outputs.zip ]]; then
+              (cd artifacts/playwright-e2e && unzip -oq test.outputs/outputs.zip -d undeclared-outputs) || true
+            fi
+            find artifacts/playwright-e2e -type f | head -50
+          else
+            echo "no bazel-testlogs/playwright_chromium_e2e directory (e2e did not run)" | tee artifacts/playwright-e2e/NO-RUN.txt
+          fi
+
+      - name: Upload e2e evidence
+        if: always()
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7
+        with:
+          name: playwright-e2e-${{ github.run_id }}-${{ github.run_attempt }}
+          path: artifacts/playwright-e2e
+          if-no-files-found: warn
+          retention-days: 7
+
   build-and-test:
     runs-on: ubuntu-latest
     timeout-minutes: 45

playwright.bazel.config.ts

@@ -1,3 +1,4 @@
+import { join } from 'node:path';
 import { defineConfig, devices } from '@playwright/test';
 
 type TraceMode = 'off' | 'on' | 'retain-on-failure' | 'on-first-retry';
@@ -6,6 +7,12 @@ const chromiumExecutable =
 	process.env.PLAYWRIGHT_CHROMIUM_EXECUTABLE_PATH || process.env.GF_RBE_CHROMIUM_EXECUTABLE || process.env.CHROME_BIN;
 const traceMode = parseTraceMode(process.env.PLAYWRIGHT_TRACE_MODE);
 const e2ePort = parsePort(process.env.PLAYWRIGHT_E2E_PORT);
+// Under `bazel test` the sandbox is discarded; anything worth keeping (traces,
+// screenshots, error contexts) must land in TEST_UNDECLARED_OUTPUTS_DIR so Bazel
+// zips it into bazel-testlogs/<target>/test.outputs/outputs.zip.
+const outputDir = process.env.TEST_UNDECLARED_OUTPUTS_DIR
+	? join(process.env.TEST_UNDECLARED_OUTPUTS_DIR, 'playwright')
+	: 'test-results';
 
 if (!chromiumExecutable) {
 	throw new Error(
@@ -15,6 +22,7 @@ if (!chromiumExecutable) {
 
 export default defineConfig({
 	testDir: 'e2e',
+	outputDir,
 	timeout: 30_000,
 	fullyParallel: true,
 	forbidOnly: true,