[tsbinns]

Full docs build is failing when checking whether some pyvista figures are active, e.g.: https://app.circleci.com/pipelines/gh/mne-tools/mne-python/35013/details?useNewPipelines=true&workflowId=764768bd-2170-42a9-8a22-39fc5d8fc4ee&job=07a47ab9-e4f0-4fea-a714-d7bb09d2fcc9&buildNumber=89422&jobType=build&tab=tests#test-row-_..%2Fexamples%2Finverse%2Fresolution_metrics_eegmeg.py_resolution_metrics_eegmeg_Compute%20spatial%20resolution%20metrics%20to%20compare%20MEG%20with%20EEG%2BMEG

Seems to be a knock-on of the fixes for ty in #14443. Didn't show up before due to the partial docs build there, and then the failure was obscured in the daily full build due to the data timeout that was fixed in #14467.

[tsbinns]

Examples that were previously failing now run cleanly: https://app.circleci.com/pipelines/github/mne-tools/mne-python/35014/workflows/84d2ff6b-7256-45c1-928c-f1c596288075/jobs/89425

Will push something that should fix the Azure failures.

[tsbinns]

For context, the Azure jobs were failing because ubuntu-latest moved from 24.04 to 26.04, which meant the home dir changed from /home/vsts to /home/runner. Have switched to using the new home dir.

However, test_flash_bem is now failing. Claude traces it back to a difference between the output of atan2f in glibc 2.39 (ubuntu-24.04) and glibc 2.43 (ubuntu-26.04). Is it okay to just change the tolerance in the assertion or is something deeper required? I'm not super familiar with BEMs.

[drammock]

Changing tolerance is ok in case like this I think

[tsbinns]

Have pushed a commit with tolerance loosened. Max difference was 0.067 mm. Went with 0.1 mm tolerance for extra leeway, but can tighten up if wanted.

[larsoner]

Thanks @tsbinns !