jwmossmoz/os-integrations

OS Integrations

Run Firefox mach try commands with pre-configured worker pool overrides for testing against alpha images.

Usage

# Run with preset (dry-run to preview)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --dry-run

# Push to try server
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --push

# Filter to specific test types (recommended)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 -t xpcshell -t mochitest-browser-chrome --push
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 -t mochitest-devtools-chrome -t mochitest-chrome-1proc --dry-run

# Override query (advanced)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 -q "test-windows11-64-24h2" --push

Build Behavior

By default, the script reuses builds from the latest autoland decision task (skipping the 45+ minute Firefox build). Autoland is Firefox's integration branch where tier 1 tasks must be green, so its decision tasks have the freshest and most complete pre-built task pool to reuse. Use --fresh-build to force a full build instead:

# Default: reuses existing Firefox builds
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 -t xpcshell --push

# Use a specific decision task
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --task-id ABC123 -t mochitest-browser-chrome --push

# Force a fresh Firefox build
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --fresh-build --push

Watching Test Results

Use --watch to automatically monitor test results with treeherder-cli after pushing:

# Push and watch all test results
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 -t xpcshell --watch

# Watch with filter (regex)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --watch --watch-filter "xpcshell|mochitest"

# Combine with fresh build and watch
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --fresh-build -t xpcshell --watch

Watching Lando Job Status

Use --watch-lando to poll the Lando landing job status until it lands or fails:

# Push and watch Lando job (polls every 90 seconds by default)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 -t xpcshell --watch-lando

# Custom polling interval (in seconds)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --watch-lando --lando-interval 60

# Combine with test watching (Lando check runs first, then test watching)
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --watch-lando --watch

Named Query Sets

Use --query-set to run a predefined set of test queries. Query sets can bundle specific suites with their own settings (e.g., skipping os-integration):

# Run targeted test suites
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --query-set targeted --push

# Preview what a query set will run
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --query-set targeted --dry-run

# Watch results from a query set
uv run ~/.claude/skills/os-integrations/scripts/run_try.py win11-24h2 --query-set targeted --watch

Query sets are defined per-preset in references/presets.yml under the query_sets key.

Common Test Types

Use -t to filter to specific test suites:

• xpcshell - XPCShell tests
• mochitest-browser-chrome - Browser chrome mochitests
• mochitest-chrome-1proc - Chrome mochitests (single process)
• mochitest-devtools-chrome - DevTools mochitests
• mochitest-plain - Plain mochitests
• reftest - Reference tests
• crashtest - Crash tests

Available Presets

• win11-24h2 - Windows 11 24H2 standard
• win11-25h2 - Windows 11 25H2 (redirects 24H2 tasks to 25H2 alpha pools)
• win11-25h2-prod - Windows 11 25H2 tier 1 on production pool (post-deploy image validation, no worker overrides)
• win11-hw - Windows 11 hardware workers
• win10-2009 - Windows 10 2009
• win11-amd - Windows 11 AMD configuration
• win11-source - Source image testing
• b-win2022 - Build worker testing
• win11-arm64 - ARM64 architecture

Prerequisites

• Firefox repository at ~/firefox
• Must be on a feature branch (not main/master)
• Mozilla Auth0 authentication (for Lando-based pushes)

Gotchas

• Must be on a feature branch — mach try refuses pushes from main/master.
• Default reuses builds from the latest autoland decision task (skips a 45-min Firefox build). Use --fresh-build only when something changed in the build itself.
• Lando-based pushes need Mozilla Auth0; the auth prompt opens in the browser if needed.
• Each preset's worker overrides live in references/presets.yml — change them there, not in run_try.py.
• Use --query-set (per preset) instead of long -t lists when you have a recurring test bundle.

Additional Documentation

• Presets Configuration: See references/presets.yml
• Linux Worker Overrides: See references/linux-worker-overrides.md
• Pushing to Try: See references/pushing-to-try.md
• Script Help: Run uv run ~/.claude/skills/os-integrations/scripts/run_try.py --help

Official Documentation

For more information on mach try and Taskcluster:

• Firefox Try Documentation: https://firefox-source-docs.mozilla.org/tools/try/
• Taskcluster Documentation: https://docs.taskcluster.net/
• Firefox Source Docs: https://firefox-source-docs.mozilla.org/