I have this script on a server: import matplotlib.pyplot as plt import matplotlib as mpl import numpy as np from time import time bknd = mpl.get_backend() x = np.random.random(100)*10 y = np.sin(x) t0 = time() plt.plot(x,y,'.') t1 = time() print(f"{bknd} {t1-t0:.1f} seconds") plt.show() If I run it in the terminal of VS Code using Remote-SSH I get TkAgg 1.7 seconds while using SSH in either iTerm or the regular Mac Terminal I get TkAgg 7.3 seconds So the time to create a figure is much faster in VS Code, while the backend is the same. I have gotten similar results from a Linux machine before, so it shouldn't be Mac related. Using other graphical programs like feh for images, or mupdf for PDFs, the there is no apparent speed difference. Another thing I noticed is that if I set my .ssh/config file to use controlMaster auto (sharing the SSH connection) then if I initialize by first sshing with iTerm and then connect with VS Code, the timing in VS Code becomes the same as in iTerm (around 7 seconds) while if I create the connection by first using VS Code, then the timings remain the same as when using the separate connections. So VS Code is doing something smart when setting up the connection, that only affects it's own terminals. How does VS Code get faster plots, and can it be replicated in other terminals? Continue reading...
