[Bug]: array-like linewidth raises an error for scatter3D

[krassowski]

Bug summary

The documentation says that linewidths can be float or array-like and it defaults to 1.5, but passing an array-like leads to an error

Code for reproduction

import matplotlib.pyplot as plt

fig = plt.figure()
ax = fig.add_subplot(projection='3d')
x = range(1, 10)
ax.scatter(x, x, x, marker='o', s=x, linewidth=x)
plt.show()

Actual outcome

Exception:

mpl_toolkits/mplot3d/axes3d.py in do_3d_projection(artist)
    469                     # avoid passing it to them; call this directly once the
    470                     # deprecation has expired.
--> 471                     return artist.do_3d_projection()
    472 
    473                 _api.warn_deprecated(

matplotlib/_api/deprecation.py in wrapper(*inner_args, **inner_kwargs)
    429                          else deprecation_addendum,
    430                 **kwargs)
--> 431         return func(*inner_args, **inner_kwargs)
    432 
    433     return wrapper

mpl_toolkits/mplot3d/art3d.py in do_3d_projection(self, renderer)
    615 
    616         if len(self._linewidths3d) > 1:
--> 617             self._linewidths = self._linewidths3d[z_markers_idx]
    618 
    619         # Re-order items

TypeError: only integer scalar arrays can be converted to a scalar index

gets raised, from:

matplotlib/lib/mpl_toolkits/mplot3d/art3d.py

Lines 599 to 603 in 0d433b4

	if len(self._sizes3d) > 1:
	self._sizes = self._sizes3d[z_markers_idx]
	if len(self._linewidths3d) > 1:
	self._linewidths = self._linewidths3d[z_markers_idx]

This is because the type of self._sizes3d is numpy.ndarray, but type of self._linewidths3d is list.

Expected outcome

Array-like linewidth works for 3d plots as it does for 2d plots.

Additional information

• ax.scatter(x, x, x, marker='o', s=x) works fine.
• linewidths is a list regardless of user input type.
• I don't know where linewidths gets converted to a list, but a quick fix is as simple as wrapping it in np.array call: self._linewidths = np.array(self._linewidths3d)[z_markers_idx] (tested that this fixes the issue).

Operating system

No response

Matplotlib Version

3.5.2 but also happens with older versions (3.3.x)

Matplotlib Backend

No response

Python version

No response

Jupyter version

No response

Installation

pip

[krassowski]

The conversion to list happens in parent class' (PathCollection) set_linewidth:

matplotlib/lib/matplotlib/collections.py

Lines 566 to 583 in 386d615

	def set_linewidth(self, lw):
	"""
	Set the linewidth(s) for the collection. *lw* can be a scalar
	or a sequence; if it is a sequence the patches will cycle
	through the sequence
	Parameters
	----------
	lw : float or list of floats
	"""
	if lw is None:
	lw = self._get_default_linewidth()
	# get the un-scaled/broadcast lw
	self._us_lw = np.atleast_1d(np.asarray(lw))
	# scale all of the dash patterns.
	self._linewidths, self._linestyles = self._bcast_lwls(
	self._us_lw, self._us_linestyles)

which gets called via super():

matplotlib/lib/mpl_toolkits/mplot3d/art3d.py

Lines 563 to 566 in 0d433b4

	def set_linewidth(self, lw):
	super().set_linewidth(lw)
	if not self._in_draw:
	self._linewidth3d = lw

More specifically, the call to self._bcast_lwls leads to the linewidth getting converted to a list here:

matplotlib/lib/matplotlib/collections.py

Line 691 in 386d615

linewidths = list(linewidths) * (l_dashes // gcd)

The safest solution would be to cast to numpy array in do_3d_projection, but it feels like _bcast_lwls should return a numpy array in first place (though such a change could break something elsewhere if there are other parts of the codebase expecting it to be a list). Any preferences?