Welcome to matplotlib and thanks for the PR!

There are some failing tests. Please check the test logs and try to fix the errors (Personally, I recommend looking at the Travis CI Python 3.6 or Python 3.7 runs first, because they are quite easy to read.) If you need more assistance please do not hesitate to ask.

As for speed comparison tests: The most simple solution would be to test the whole quiver() method in ipython using %timeit ax.quiver(...). This contains more stuff, but ideally the speed improvement should be visible at top level as well.
Another option would be to line-profile quiver() to see how much time the calculation of the head_dirs actually takes. See e.g. https://jakevdp.github.io/PythonDataScienceHandbook/01.07-timing-and-profiling.html#Line-By-Line-Profiling-with-%lprun

The earlier tests I made were done in the following way:
I opened a new project with a new virtualenv, installed matplotlib via pip and than changed the code of the installed files (inside site_packages).

Instead, I would like to add tests in the code I cloned from github to test my changes in a more organized manner, but I am encountering an issue.

I added a test function to the matplotlib/tests/test_quiver.py module and when I try to run anything the following error is collected by pytest:

(<class 'ImportError'>, ImportError("cannot import name 'ft2font' from 'matplotlib' <project_path>/matplotlib/lib/matplotlib/__init__.py)")

Can you instruct me how to add tests inside the matplotlib clone? :)

Have you followed the instructions on Retrieving and installing the latest version of the code? In particular, have you installed your checked out version in developer mode?

I haven't followed these instructions and I will do so now :) Thanks!

Ok i've managed to run the existing tests with the replacement of the function to the vectorized form and they all passed. I'll try to upload it here to see if there are any more issues

All of the tests passed beside one test in travis, but I'm not sure what is the cause of the failure and how I can fix this.

We impose a limit of 80 characters per line. You have 3 lines that surpass this limit

./lib/mpl_toolkits/mplot3d/axes3d.py:2578:80: E501 line too long (98 > 79 characters)
./lib/mpl_toolkits/mplot3d/axes3d.py:2579:80: E501 line too long (99 > 79 characters)
./lib/mpl_toolkits/mplot3d/axes3d.py:2580:80: E501 line too long (81 > 79 characters)

Oh I indeed see this now in the report, thank you! :) I wasn't sure where to look earlier.

Now that all the checks pass I would like to create the time comparison.
I would like to compare the time using variable number of vectors as input to quiver.
I could use timeit multiple times, switch branch, than use timeit multiple times again.
Is there a another standard way to produce this comparison?

%load_ext autoreload
%autoreload 2

from mpl_toolkits.mplot3d import Axes3D
import matplotlib.pyplot as plt
import numpy as np

sizes =  [1, 10, 50]
def test_times():
    times = []
    for n in sizes:
        fig = plt.figure()
        ax = fig.gca(projection='3d')
        x, y, z = [np.random.rand(n) for _ in range(3)]
        X, Y, Z = np.meshgrid(x, y, z)
        U, V, W = [np.random.rand(*X.shape) for _ in range(3)]
        time = %timeit -o ax.quiver(X, Y, Z, U, V, W)
        times.append(time)
        plt.close()
    return times

!git -C matplotlib checkout master
times_old = test_times()

838 µs ± 6.94 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)
31.4 ms ± 87.2 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)
3.9 s ± 15.3 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

!git -C matplotlib checkout calc-arrow-vectorization
times_new = test_times()

906 µs ± 616 ns per loop (mean ± std. dev. of 7 runs, 1000 loops each)
2.79 ms ± 17.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)
275 ms ± 2 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

plt.figure()
num_vec = np.power(sizes, 3)
averages_old = [t.average for t in times_old]
averages_new = [t.average for t in times_new]
stds_old = [t.stdev for t in times_old]
stds_new = [t.stdev for t in times_new]
plt.errorbar(num_vec, averages_old, stds_old, label='loop')
plt.errorbar(num_vec, averages_new, stds_new, label='vectorization')
plt.legend()
plt.ylabel('time[sec]')
plt.xlabel('num vectors')
plt.yscale('log')
plt.xscale('log')

Is there anything more I should provide?

Power-cycled to re-start CI against current master.

Looks great! Do you want to squash your commits?

Yeah I'll squash my commits when I get to the pc :)

Thanks, hoping to see you around again!

I hope so too! :D