First of all, thank you to all matplotlib contributors for making what's one of my most used Python libraries.

When wanting to plot a particular vector field on a given surface, I came across Axes3D.quiver which does a good job at exactly that. However, one thing that surprised me is that the scales of the plotted vectors are not based on what is input; only their scale is. The optional argument length can be used to modify the lengths of all vectors at once, but not for individual ones.

I ended up using the workaround in this StackOverflow answer, but that seems less than ideal.

In pictures; desired behavior (vectors at poles are smaller than those at the equator):

Actual behavior (all vectors have the same size):

Version information:

• Matplotlib 1.5.3, Python 3.5.2, Linux 3.16.0-4-amd64 Fix autofmt_xdate() when using in conjunction with twinx() #1 SMP Debian 3.16.36-1+deb8u2
• Python installed from source, Matplotlib from pip.