matplotlib shape tuple syntax fails for tuple of tuples

from https://github.com/machow/plotnine-guide/pull/20, matplotlib provides a special tuple syntax for creating shapes. This includes a "tuple of tuples" syntax for drawing things. However, it seems like this fails due to pandas groupby not knowing how to sort the tuple of tuples (when other shapes are involved).

In the example below, removing house from the data works, but with it in, raises a pandas sort error (can't compare int with tuple):

# polygon: (num, 0, angle)
pentagon = (5, 0, 60)

# star: (num, 1, angle)
five_point_star = (5, 1, 60)

# astericks: (num, 2, angle)
five_point_astericks = (5, 2, 60)

# path: list[tuple(float, float)]
house = ((-2, -3), (-2, -1), (0, 0), (2, -1), (2, -3), (-2, -3))
shape = [pentagon, five_point_star, five_point_astericks, house]

shapes = pd.DataFrame(
    {
        "shape": shape,
        "x": list(range(len(shape))),
        "y": [0] * len(shape),
    }
)

(
    ggplot(shapes, aes("x", "y"))
    + geom_point(aes(shape="shape"), size=5)
    # + geom_text(aes(label="shape"), nudge_x=0.15)
    + scale_shape_identity(guide=None)
    + theme_void()
    + expand_limits(x=4.1)
)

Start by running the supplied example with the mixed shape values through geom_point and scale_shape_identity. Trace the pandas groupby step that sorts shape values and verify the tuple-of-tuples shape no longer causes an int-versus-tuple comparison error while the other shapes still render.