Make it easier to spot when slow tests are added

[coretl]

We should make sure that all new tests are relatively fast, with a few smoke tests that are allowed to take a bit longer. I suggest a default per-test timeout of 0.5s, with an override for the few tests that take longer than this.

This would involve:

• Adding pytest-timeout as a dev dependency in pyproject.toml
• Setting in timeout = 0.5 in [tool.pytest.ini_options] in pyproject.toml
• Adding @pytest.mark.timeout(<a_longer_timeout>) for those tests that now start to fail

As per https://pypi.org/project/pytest-timeout/

In a later PR we will make the slow tests faster

[DominicOram]

Can I suggest we do DiamondLightSource/python-copier-template#235 instead?

[coretl]

Can we do this first, then do the copier template change following the 3 examples?

I would prefer the pytest timeout option as per DiamondLightSource/dodal#1033 (comment) and this issue description, so I think we should try it first then replicate it in copier

[DominicOram]

Yh, I have no opinion on the implementation. If you want to try out the pytest implementation here then put that into the template that's fine. There's no point us putting anything in mx-bluesky though if it's going to be part of the template soon anyway

[Relm-Arrowny]

Is this still needed if so can I pick this one up?

[DominicOram]

Yh, I think it would be good. Our findings from mx-bluesky:

• pytest-timeout works well
• We found that some of our tests ran slower in CI, we suspect it's just the GH containers being a little unpredictable, we made the timeout longer in CI than locally to get round this. I think we care more about how long things take locally anyway
• We use a 1s timeout locally and 2s in CI. I think there's some systematic things we do in mx-bluesky that are slow though, if you can get away with 0.5s here you should

[Relm-Arrowny]

Just to make a note in case I forgot as I am going off for 2 weeks, this issue is completed for Linux and just like mx I have to increase the CI timeout to 2s but local time out is 0.5s, I have add individual timeout for all test that failed locally with timeout = local test time + 2 sec.
The window CI is failing, this may be due to this, I will try to look into it when I get back.

[coretl]

I'd love it if the windows runners could be made faster, https://ichard26.github.io/blog/2025/03/faster-pip-ci-on-windows-d-drive/ seems to have some tips...

[coretl]

@Relm-Arrowny what is the status of this?

[Relm-Arrowny]

This has been completed. All local tests now have a default timeout of 0.5s, and any tests that typically fail locally have been given their own longer timeout.

For GitHub Actions, the timeout is set to 2 seconds for Linux and 6 seconds for Windows. We will no longer need a separate, longer timeout for Windows if and when we drop support for Python 3.10.

I also added an extra step to the Windows build to use the faster D: drive. This may provided a 10-20% speed increase for those operations, but it was not enough to make a significant difference to the overall build time.

    - name: Set TEMP to D:/Temp
        if: inputs.runs-on == 'windows-latest'
        run: |
          mkdir "D:\\Temp"
          echo "TEMP=D:\\Temp" >> $env:GITHUB_ENV

I have not make a pr for this yet as I got a strange test failure on 3.10 that does not seem to have anything to do with timeout.

=========================== short test summary info ===========================
FAILED tests/epics/demo/test_epics_demo.py::test_motor_moving_well - AssertionError: assert call(current=0.0, initial=0.0, target=0.55, name='mock_motor', unit='mm', precision=3, fraction=None, time_elapsed=0.17200000000002547, time_remaining=None) == call(current=0.0, initial=0.0, target=0.55, name='mock_motor', unit='mm', precision=3, fraction=None, time_elapsed=0.0 � 8.0e-02, time_remaining=None)
  
  Full diff:
  - call(current=0.0, initial=0.0, target=0.55, name='mock_motor', unit='mm', precision=3, fraction=None, time_elapsed=0.0 � 8.0e-02, time_remaining=None)
  ?                                                                                                                       ----- ^^
  + call(current=0.0, initial=0.0, target=0.55, name='mock_motor', unit='mm', precision=3, fraction=None, time_elapsed=0.17200000000002547, time_remaining=None)

[coretl]

Please could you make a PR for it, then we can examine the test failure over there...