Is this based on the (relatively recent) type annotations? If so perhaps rather than changing the test, the annotation should be tweaked. If the docstring also said array, then it is less clear cut.

Ah - so this was motivated by our discussion on #5138. There too we have a test passing a possibly odd input that nevertheless worked for years. My slight worry over any tightening in the accepted input style is we'll discover end users' scripts were also passing slightly odd arrays in too... and we'll break their code.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78.10%. Comparing base (ead54d7) to head (dbbfefa).

@@            Coverage Diff             @@
##           master    #5166      +/-   ##
==========================================
- Coverage   85.51%   78.10%   -7.41%     
==========================================
  Files         282      282              
  Lines       59490    59490              
==========================================
- Hits        50871    46466    -4405     
- Misses       8619    13024    +4405     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.