[TATP-233]

Summary

• Added a module-scoped short_temp_path fixture that creates isolated /tmp/ucmps-* directories, keeping real AF_UNIX socket paths below macOS's 95-byte limit.
• Reworked MPS tests that create real Unix-domain sockets to use this short root while retaining automatic cleanup.
• Made direct start_daemon() tests use the existing Linux-host fixture so they exercise their intended start/topology branches on physical macOS hosts without weakening production platform checks.
• Preserved the long-path rejection test with a path length independent of pytest's platform-specific temporary-root depth.
• No production code changed; CUDA MPS remains manually managed and the Linux/NVIDIA guard is unchanged.

Linked Work

• Issue: Bug: CUDA MPS CLI tests fail on macOS #2510
• Parent roadmap (when applicable): N/A
• Roadmap declared base (when applicable): N/A
• Milestone: N/A
• Base branch: main

Validation

• make test-all passed on the final local head before this PR was created or updated
• Additional task-specific validation listed below

Commands actually run:

uv run pytest -m 'not slow' tests/training/test_cuda_mps_cli.py -q
# 43 passed

TMPDIR=/tmp uv run pytest -m 'not slow' tests/training/test_cuda_mps_cli.py -q
# 43 passed

uv run pytest -m 'not slow' tests/training -q
# 326 passed, 4 skipped

make check
# passed: Ruff format/check, mypy (211 source files), PyRight, test lint

make test
# 2686 passed, 133 skipped, 373 deselected

make test-all
# passed: 2686 passed, 133 skipped, 373 deselected; benchmark import smoke passed 34/34 module and 35/35 script entrypoints

The focused module was also reproduced as failing before the fix: 13 failures with the default Darwin pytest temporary root and 4 failures with TMPDIR=/tmp.

Remote CI route:

• Base main: current-head CI passed — https://github.com/Motphys/UniLab/actions/runs/39831552225 (coverage, benchmark-smoke, mypy, pyright, Ruff format/lint, and all four test shards).

Impact

• Backend impact: none
• Platform impact: macOS
• Training effect expected: no

Artifacts

• W&B: N/A
• benchmark result: N/A
• video / screenshot: N/A
• ONNX / checkpoint: N/A

Checklist

• Added or updated tests where needed
• Updated docs if behavior or workflow changed
• Linked the driving issue
• Noted any follow-up work explicitly