PR summary

Closes #32127.

Line3D.set_data_3d is documented to take array-like input, but it stored the values as given. draw then does self._verts3d[0].shape when applying the invalid-scale mask, so a plain Python list with a NaN raises AttributeError. Finite lists still drew, which is why this only shows up once a coordinate goes non-finite.

I fixed it in the setter by coercing each coordinate with np.asarray. That keeps draw unchanged and makes _verts3d consistently arrays. The other option is np.shape(...) in draw, which would preserve the original list/tuple type from get_data_3d(). I preferred normalizing at the setter because the draw path already treats these as arrays, and reading the data back still compares equal under np.testing.assert_array_equal (covered by the existing set/get test).

Added a regression test for the reported case: lists containing NaN, then canvas.draw().

AI Disclosure

Used Cursor to find the mplot3d code paths, draft the small patch/test, and help open the PR. I chose coercion in set_data_3d over np.shape in draw after reading the discussion on the earlier closed attempt (#32128). The summary above is written in my own words.

PR quality check

• Use an expressive title, e.g. "Fix title font property precedence"
• New and changed code is tested
• [n/a] Plotting related features are demonstrated in an example
• [n/a] New features and API changes have release notes
• [n/a] Documentation complies with general and docstring guidelines