errorbar checks that errors are not negative, but a bit convolutedly, in order to avoid triggering on nan. Unfortunately, the work-around for nan means that possible masks get discarded, and hence passing in a masked error array that has a negative but masked value leads to an exception. This PR solves that by simply combining the test for
negative values with the indirect isnan test (err == err), so that if a masked array is passed, the test values are masked and ignored in the check.

As a bonus, this also means that astropy'sMasked arrays can now be used -- those refuse to write output of tests to unmasked values since that would discard the mask (which is indeed the underlying problem here) -- see
astropy/astropy#17996

p.s. I checked that the test case that I added fails on current main. The test case is based on the one fornan right above - let me know if this is an OK location for it.

EDIT: updated to simplify the check without using any storing of outputs. This means we can also continue to use astropy's Quantity for error bars (and just shows how damn careful one has to be, sigh).

EDIT 2: second commit prevents any comparison ofnan < nan, which causesRuntimeError that I did not see locally.

PR checklist

• [N/A] "closes #0000" is in the body of the PR description tolink the related issue
• new and changed code istested
• [N/A]Plotting related features are demonstrated in anexample
• [N/A]New Features andAPI Changes are noted with adirective and release note
• [N/A] Documentation complies withgeneral anddocstring guidelines