release-testing

Release Testing Skill

Verify SkiaSharp packages work correctly before publishing.

⚠️ NO UNDO: This is Step 3 of 4 in the release pipeline. See releasing.md for full workflow.

Pipeline: Step 1: release-branch → Step 2: release-status → Step 3 (this skill) → Step 4: release-publish

CRITICAL: ANY FAIL IS TOTAL FAIL

• Test fails → Release FAILS
• Test times out → Release FAILS
• Screenshot doesn't match → Release FAILS

Never rationalize failures. Fix the issue before proceeding.

---

⚠️ CRITICAL: Semver Version Ordering

When identifying which release branch to test, you MUST use semver ordering, NOT alphabetical or sort -V ordering.

In semver, a bare version is ALWAYS newer than its prerelease variants:

3.119.2-preview.1 < 3.119.2-preview.2 < 3.119.2-preview.3 < 3.119.2 (FINAL)

release/3.119.2 is the stable release and is NEWER than release/3.119.2-preview.3.

To find the latest release branch:

1. List all release branches: git branch -r | grep "release/"
2. Identify the highest base version (e.g., 3.119.2)
3. Check if a bare version branch exists (e.g., release/3.119.2) — if so, that is the latest
4. If only preview branches exist, the highest preview number is the latest

⚠️ Getting this wrong means testing the wrong version — wasting the entire process or shipping untested packages.

---

Step 1: Check CI Status

Before testing, verify CI builds have completed using the release-status skill:

python3 .agents/skills/release-status/scripts/pipeline-status.py release/{version}

Prerequisite: The SkiaSharp pipeline (ID 10789) must have completed successfully — this is the pipeline that signs and publishes packages to the internal feed.

SkiaSharp-Tests (ID 15756) should pass but does not block testing/publishing.

See the release-status skill for full pipeline chain documentation, manual queries, and troubleshooting.

Extracting NuGet Version

The build description contains the internal version in format: #{base}-{label}.{build}+{branch}

Preview example: #3.119.2-preview.2.3+3.119.2-preview.2 succeeded

• Internal version: 3.119.2-preview.2.3
• NuGet version: 3.119.2-preview.2.3 (same — build number is part of the prerelease tag)

Stable example: #3.119.2-stable.3+3.119.2 succeeded

• Internal version: 3.119.2-stable.3
• NuGet version: 3.119.2 (base only — build number is NEVER appended to stable versions)

⚠️ Stable versions never include a build number. Each CI build of a stable release produces a different internal package (3.119.2-stable.1, 3.119.2-stable.2, etc.) but the published NuGet version is always just 3.119.2.

---

Step 2: Resolve Package Versions

DO NOT ask user for exact NuGet versions. Resolve automatically:

1. Fetch release branch and read version files:

   # Read base versions (format: "PackageName  nuget  version")
   grep "^SkiaSharp\s" scripts/VERSIONS.txt | grep "nuget" | awk '{print $3}'
   grep "^HarfBuzzSharp\s" scripts/VERSIONS.txt | grep "nuget" | awk '{print $3}'
   
   # Read preview label (remove surrounding quotes)
   grep "PREVIEW_LABEL:" scripts/azure-templates-variables.yml | awk '{print $2}' | tr -d "'"
   

   • SkiaSharp ... nuget line → base version (e.g., 3.119.2)
   • HarfBuzzSharp ... nuget line → base version (e.g., 8.3.1.3)
   • PREVIEW_LABEL → label (e.g., preview.2 or stable)

2. Search and filter for the SPECIFIC version:

   For preview releases (PREVIEW_LABEL is NOT stable):

   # Get ALL versions, then filter to match {base}-{label}.*
   dotnet package search SkiaSharp \
     --source "https://aka.ms/skiasharp-eap/index.json" \
     --exact-match --prerelease --format json \
     | jq -r '.searchResult[].packages[] | select(.id == "SkiaSharp") | .version' \
     | grep "^{base}-{label}\."
   
   # Example: Find 3.119.2-preview.3.* versions
   ... | grep "^3.119.2-preview.3\."
   

   Pick the highest build number (e.g., 3.119.2-preview.3.1). This IS the NuGet version.

   For stable releases (PREVIEW_LABEL is stable):

   # Verify a stable build exists on the internal feed
   dotnet package search SkiaSharp \
     --source "https://aka.ms/skiasharp-eap/index.json" \
     --exact-match --prerelease --format json \
     | jq -r '.searchResult[].packages[] | select(.id == "SkiaSharp") | .version' \
     | grep "^{base}-stable\."
   
   # Example: Find 3.119.2-stable.* internal packages
   ... | grep "^3.119.2-stable\."
   

   The internal feed has {base}-stable.{build} packages (e.g., 3.119.2-stable.3), but the NuGet version is just {base} (e.g., 3.119.2). The build number is never appended to stable versions.

   ⚠️ CRITICAL: Use .version to get ALL versions, NOT .latestVersion which only returns the newest. The feed contains multiple version streams (e.g., 3.119.2 AND 3.119.3), so you MUST filter by the base version and preview label from the release branch.

3. Pick the NuGet version:

   • Preview: Highest build number from matching versions (e.g., 3.119.2-preview.3.1)
   • Stable: Just the base version (e.g., 3.119.2) — no build number appended

4. Report to user:

   Preview:

   Resolved versions:
     SkiaSharp:     3.119.2-preview.3.1
     HarfBuzzSharp: 8.3.1.3-preview.3.1
     Build number:  1
   

   Stable:

   Resolved versions:
     SkiaSharp:     3.119.2
     HarfBuzzSharp: 8.3.1.3
     Internal build: 3.119.2-stable.3 (on feed)
   

No packages found? CI build hasn't completed. See troubleshooting.md.

---

Step 3: Confirm Test Matrix

Before running tests, determine and confirm the test matrix with the user.

Device Requirements

👉 See setup.md for device selection details and emulator creation.

Confirm with User

Planned test matrix:
  - iOS (old):     [device] ([oldest available iOS runtime])
  - iOS (new):     [device] ([newest available iOS runtime])
  - Android (old): [device] (Android 6.0 / API 23)
  - Android (new): [device] (Android 16 / API 36)
  - Mac Catalyst:  Current macOS
  - Blazor:        Chromium
  - Console:       .NET runtime
  - Linux (Docker): Docker container (mcr.microsoft.com/dotnet/sdk:8.0)

Proceed with this matrix?

---

Step 4: Run Integration Tests

Pre-Test Cleanup (REQUIRED)

⚠️ CRITICAL: These steps MUST be done before ANY integration tests:

# 1. Clear screenshot folder to ensure fresh results
rm -rf output/logs/testlogs/integration/*
mkdir -p output/logs/testlogs/integration

# 2. Kill any running Android emulators
adb devices | grep emulator | awk '{print $1}' | while read emu; do
  adb -s $emu emu kill 2>/dev/null
done
sleep 5

# 3. Verify clean state
adb devices -l  # Should show NO emulators
ls output/logs/testlogs/integration/  # Should be empty

Run Tests

cd tests/SkiaSharp.Tests.Integration
dotnet test -p:SkiaSharpVersion={version} -p:HarfBuzzSharpVersion={hb-version}

Test Commands

Note: This project uses Microsoft.Testing.Platform (MTP) with xUnit v3 (since #4143). The legacy VSTest --filter "FullyQualifiedName~..." syntax is silently ignored under MTP and runs ALL tests. Use the MTP filter args after the -- separator instead: --filter-class, --filter-method, --filter-namespace (and --filter-not-class, etc.), with * wildcards. MSBuild -p: properties (e.g. -p:SkiaSharpVersion=, -p:iOSDevice=) must stay BEFORE the --; only the test-platform filter args go AFTER it.

# Run by category
dotnet test ... -- --filter-class "*SmokeTests"
dotnet test ... -- --filter-class "*ConsoleTests"
dotnet test ... -- --filter-class "*LinuxConsoleTests"
dotnet test ... -- --filter-class "*BlazorTests"
dotnet test -p:iOSDevice="iPhone 14 Pro" -p:iOSVersion="16.2" ... -- --filter-class "*MauiiOSTests"
dotnet test ... -- --filter-class "*MauiMacCatalystTests"

# Android: specify device ID and expected API level for validation
dotnet test ... \
  -p:AndroidDeviceId="emulator-5554" \
  -p:AndroidApiLevel="23" \
  -- --filter-class "*MauiAndroidTests"

Android Emulator Workflow

⚠️ CRITICAL: Run only ONE Android emulator at a time to avoid device confusion.

1. Verify no emulators running:

   adb devices -l  # Should show empty or only physical devices
   

2. Start emulator with WIPE and boot verification:

   # Start emulator with -wipe-data to ensure clean state (use mode="async" to keep it running)
   emulator -avd Pixel_API_23 -wipe-data -no-snapshot -no-audio
   
   # Wait for boot (check every 10s until returns "1")
   # This can take 60-120s for a fresh wipe
   adb shell getprop sys.boot_completed
   
   # Verify correct API level
   adb shell getprop ro.build.version.sdk  # Should match expected (e.g., "23")
   

   ⚠️ The -wipe-data flag is REQUIRED to ensure a clean emulator state. Without it, cached apps or system state from previous runs may interfere with tests.

3. Run tests with device validation:

   DEVICE_ID=$(adb devices | grep emulator | awk '{print $1}')
   API_LEVEL=$(adb -s $DEVICE_ID shell getprop ro.build.version.sdk | tr -d '\r')
   
   dotnet test \
     -p:AndroidDeviceId="$DEVICE_ID" \
     -p:AndroidApiLevel="$API_LEVEL" \
     -p:SkiaSharpVersion={version} \
     -p:HarfBuzzSharpVersion={hb-version} \
     -- --filter-class "*MauiAndroidTests"
   

4. Shut down emulator before next test:

   adb -s $DEVICE_ID emu kill
   # Wait for it to stop
   sleep 5
   adb devices -l  # Verify empty
   

5. Repeat for next API level (start from step 1)

Test Execution Order

iOS and Android run TWICE: once on oldest, once on newest.

Providing User Feedback

CRITICAL: Long-running tests need continuous feedback. Users should never wait more than 30 seconds without knowing what's happening.

• Update the TODO checklist at each phase transition
• When waiting with read_bash, note elapsed time: "⏳ Still building (~60s elapsed)"
• Tell users what's normal: "MAUI Release builds take 30-120s, silence is expected"

👉 See monitoring.md for:

• Phase timing and expected durations
• Output indicators to detect which phase is active
• Feedback templates and example output
• Troubleshooting hangs and crashes

---

Step 5: Verify & Report

Release Criteria

Proceed to release-publish ONLY when:

• ✅ ALL tests pass (no failures)
• ✅ iOS tests pass on BOTH oldest and newest runtime
• ✅ Android tests pass on BOTH oldest (API 21-23) and newest (API 35-36)
• ✅ Screenshots exist in output/logs/testlogs/integration/

Skip Policy

Hardware skips only:

• iOS/Mac tests on non-macOS → Skip (hardware unavailable)
• Windows tests on non-Windows → Skip (hardware unavailable)

NOT valid skips:

• "No Android emulator" → Create one
• "Android SDK not found" → Ask user for path
• "No iOS simulators" → Install via Xcode
• "Tool X not installed" → Install it

If environment is broken, FIX IT. Do not skip tests.

Final Report Format

✅ Release Testing Complete

| Test | Platform | Version | Status |
|------|----------|---------|--------|
| SmokeTests | .NET | - | ✅ Passed |
| ConsoleTests | .NET | - | ✅ Passed |
| LinuxConsoleTests | Docker Linux | - | ✅ Passed |
| BlazorTests | Chromium | - | ✅ Passed |
| MauiMacCatalystTests | macOS | - | ✅ Passed |
| MauiiOSTests | iOS 16.2 (oldest) | iPhone 14 Pro | ✅ Passed |
| MauiiOSTests | iOS 18.5 (newest) | iPhone 16 Pro | ✅ Passed |
| MauiAndroidTests | Android 6.0 (API 23) | Pixel_API_23 | ✅ Passed |
| MauiAndroidTests | Android 16 (API 36) | Pixel_API_36 | ✅ Passed |

Ready for publishing.

---

References

• Setup & device selection: references/setup.md
• Monitoring long-running tests: references/monitoring.md
• Troubleshooting errors: references/troubleshooting.md