Summary

This PR fixes a broken unit test for unsupported export objects and resolves NumPy 2.x / Python 3.13 compatibility issues caused by stricter array-to-scalar conversion rules.

Changes

Fix broken unit test for unsupported export objects

The test_ExportDevice_unsupported test was meant to check that the exporter raises a NotImplementedError when it sees a Brian2 object it does not support.

The test was broken for two reasons:
• It used PoissonInput as the unsupported object, but PoissonInput is actually supported by the exporter.
• The pytest.raises(NotImplementedError) block had been commented out, so the test was no longer checking the expected failure.

This PR fixes the test by:
• replacing PoissonInput with a minimal custom class, UnsupportedObject, that directly subclasses BrianObject
• re-enabling the pytest.raises(NotImplementedError) assertion
• removing old setup code that was only needed for PoissonInput, including the NeuronGroup, equation string, and run(1*ms) call
• moving the BrianObject import to the top of the file to follow normal Python style

Since UnsupportedObject has no exporter implementation, running a Network containing it with the exporter device now correctly raises NotImplementedError. The test now passes and checks the intended behavior properly.

Fix failures on NumPy 2.x / Python 3.13

NumPy 2.x is stricter about converting arrays into scalar values. A few places in the code were relying on older NumPy behavior where a 1-element array could be treated like a plain number. These cases now fail in CI on Python 3.13.

test_baseexport.py — test_spikegenerator
SpikeGeneratorGroup stores spike times as a 1D array. The test was calling float() directly on this array when comparing it to the original input.

In NumPy 2.x, float() only accepts 0D arrays, so this raises a TypeError.

Fixed by extracting the first element with [0] on both sides before calling float().

mdexport/expander.py — expand_initializer
When a variable is initialized for only part of a neuron group, for example group.v[0:5] = ..., the index is stored as a NumPy integer array.

The code was comparing this array directly to the strings 'True' and 'False' using == inside an or expression. In NumPy, this returns an array of booleans, not a single boolean value. Using that in Python or raises:

ValueError: The truth value of an array with more than one element is ambiguous

Fixed by checking isinstance(index, str) before doing the string comparisons, so they only run when index is actually a string.

plotting/morphology.py — _plot_morphology2D
After calling generate_coordinates(), Brian2 returns Soma.x, Soma.y, and Soma.diameter as 1-element 1D arrays instead of plain scalar values.

Matplotlib’s Circle eventually passes these values into an internal affine transform that requires real Python floats, which causes a ValueError.

Fixed by applying np.squeeze() before float(). This converts a 1-element 1D array into a 0D array, which float() can safely convert to a scalar.

Result
• restores a meaningful unit test for unsupported export objects
• fixes CI failures with NumPy 2.x
• improves compatibility with Python 3.13