Move six suites off Intel onto the M4 fleet: jsreftest, reftest,
marionette-integration, marionette-unittest, telemetry-tests-client,
test-verify-wpt. All validated clean on macosx1500-aarch64/{debug,opt}
across two validation pushes after the two fuzzy annotations below.
Reftest fuzzy annotations for Apple Silicon-specific pixel variance:
- layout/reftests/box-shadow/boxshadow-rotated.html: M4 saw 5442
differing pixels (max 13), one above the existing 5250 ceiling.
Add fuzzy-if(appleSilicon,0-13,0-6000).
- layout/reftests/bugs/1415987-1.html: M4 saw 6029 differing pixels
(max 255), four above the existing 6025 ceiling. Existing entry
already has a wider gtkWidget allowance; add a parallel
fuzzy-if(appleSilicon,0-255,0-6500).
Both are software-rendering pixel-rasterization differences across
architectures, not regressions.
Initially included gtest, test-verify, and test-verify-gpu but dropped:
- gtest hits arch-specific failures (PHC.TestPHCAllocations,
Jemalloc.BaseAllocMergeSplit, MediaDataEncoderTest.H264Encode4K*,
VP8VideoTrackEncoder.*, TestUDPSocket.TestUDPSocketMain). The PHC
test has a pending fix on Bug 2021381; the encoder/jemalloc tests
need their own evaluation. Moving gtest separately.
- test-verify (debug+opt) fails in mozharness setup with
ModuleNotFoundError: No module named 'yaml' coming out of
manifestparser/toml.py. PyYAML is missing from the worker Python
env on gecko-t-osx-1500-m4 in the verify-tools path. Doesn't affect
test-verify-wpt, which uses a different mozharness path.
test-verify-gpu likely has the same issue but wasn't scheduled in
the validation push (no GPU test mods); deferring with test-verify.
crashtest is left alone here; it's covered by D294786 (Bug 2032601)
already in flight.
Differential Revision: https://phabricator.services.mozilla.com/D299130
Ryan Curran
10ff8c9be8
