register Axes3D along with other projections

[megies]

The docs state the following for mplot3d as preferred usage:

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')

Because Axes3D is only registered as a projection in mpl_toolkits.mplot3d.axes3d one has to import from there even though the module is otherwise integrated rather seamlessly. It feels a bit odd to have an unused import as the preferred usage.

How about registering the projection earlier, within matplotlib, thus obviating this import? As far as I can see mplot3d is always there when matplotlib is installed (I put an except ImportError anyway to make double sure to not break anything). Well.. just cosmetics, I guess, so feel free to just close.

If this gets a thumbs up, I'll also change the docs accordingly.

[mdboom]

As mplot3d is a rather large chunk of code that isn't always used, I'd rather see this import happen dynamically when it's actually required. That will require some changes to the projection registration mechanism -- to that you could map a name, 3d, to the name of a class mpl_toolkits.mplot3d.Axes3D as a string, that is then looked up and imported at runtime.

[megies]

Ok, I'll close this for now then. That's a major change I can not promise to handle soonish.

[anntzer]

It may be worth mentioning in the docs that the import is required for projection="3d" to work (as this is a non-trivial side-effect).