Flaky test: Mp4SinkWriterEncoder_RealEncoderCoversValidationAndSuccessfulComplete intermittently fails with MF_E_SINK_NO_SAMPLES_PROCESSED

[azchohfi]

Describe the bug

Mp4SinkWriterEncoderTests.Mp4SinkWriterEncoder_RealEncoderCoversValidationAndSuccessfulComplete intermittently fails in CI when the real Media Foundation H.264 encoder produces no output samples for a single-frame encode.

Seen on main in run 34433989038 (commit 8c62f868), 1 failure out of 5409 tests:

failed Mp4SinkWriterEncoder_RealEncoderCoversValidationAndSuccessfulComplete (2s 884ms)
  Test method WinApp.Cli.Tests.Mp4SinkWriterEncoderTests.Mp4SinkWriterEncoder_RealEncoderCoversValidationAndSuccessfulComplete threw exception:
  System.Runtime.InteropServices.COMException: The operation failed because no samples were processed by the sink. (0xC00D4A44)
    at ...IMFSinkWriter.Finalize()
    at ...Mp4SinkWriterEncoder.Complete() in src/winapp-CLI/WinApp.UIAutomation.Recording/Mp4SinkWriterEncoder.cs:264
    at WinApp.Cli.Tests.Mp4SinkWriterEncoderTests...() in src/winapp-CLI/WinApp.Cli.Tests/Mp4SinkWriterEncoderTests.cs:67

0xC00D4A44 is MF_E_SINK_NO_SAMPLES_PROCESSED.

Cause. The test encodes exactly one 64×64 frame and then immediately calls Complete(), which calls IMFSinkWriter.Finalize():

encoder.WriteFrame(Enumerable.Repeat((byte)0x22, 64 * 64 * 4).ToArray(), 0, 10_000_000);
encoder.Complete();

H.264 MFTs buffer input, so a single frame can legitimately reach Finalize() before any compressed sample has reached the media sink. Whether it passes depends on which H.264 encoder MFT the host resolves and how it buffers, which makes it host-dependent rather than deterministic.

This is a test-robustness problem, not a product bug. In the real recording path, UiRecordingService already wraps encoder.Complete() in a catch (Exception ex) when (IsRecoverableVideoOutputFailure(ex)), and that predicate matches ExternalException, which COMException derives from. So winapp ui record degrades gracefully; only this unit test calls Complete() unguarded.

To Reproduce

Not deterministically reproducible by design — it depends on the H.264 encoder MFT available on the host.

1. Run dotnet test src/winapp-CLI/WinApp.Cli.Tests --filter "ClassName~Mp4SinkWriterEncoderTests" repeatedly, or let CI run main.
2. On a host whose H.264 MFT buffers the single input frame past Finalize(), the test throws COMException (0xC00D4A44).

Observed frequency on CI: 1 occurrence in run 34433989038; the two following main runs (122a6592, 4a2776ca) both passed with the same code, so it is intermittent rather than a regression. Mp4SinkWriterEncoder.cs has not changed since #799 (2026-09-02).

Expected behavior

The test should deterministically exercise encoder argument validation and a successful Complete()/publish, without depending on how a particular host's H.264 MFT buffers a single frame.

Suggested fix: write enough frames that the encoder must emit at least one compressed sample before finalizing (a short run of frames at the configured fps rather than exactly one). That preserves everything the test currently covers — the expected 16384 short-buffer ArgumentException, the idempotent second Complete(), and the atomic publish assertions — while removing the single-frame dependency.

Tolerating MF_E_SINK_NO_SAMPLES_PROCESSED as inconclusive would also make CI green, but it would hollow out the assertion that a completed MP4 is actually published, so it is not preferred.

OS Version and details

windows-latest GitHub-hosted runner (CI); .NET 10 SDK.