This is not a bug. There was never any intention that one could change the colors by mutating the output of get_facecolors(). If you want to change the colors, you need to use the setter.

import matplotlib.pyplot as plt

sc1 = plt.scatter([0, 1, 2], [0, 1, 2], c=["r", "g", "b"])
# Update sc1 before the first draw.
fc = sc1.get_facecolors()
fc[0] = (0, 0, 0, 1)  # i.e. black. instead of red
sc1.set_facecolors(fc)
plt.gcf().canvas.draw()

plt.show()

Ah, thanks for pointing this out.
Do you think get_facecolors() could return e.g. a read-only view on the colors to prevent the users from getting this wrong? Or at least the limitation should be documented. (Note that the limitation is somewhat subtle, as directly mutating the array will sometimes work (e.g. after a draw has been triggered, AFAICT).

Just for slightly more concrete discussion of the implementation, the primary thing the setter does (in regards to why just editing the getter's output is insufficient) is store the direct input to self._original_facecolor before actually computing the RGBA array

This is done so that the colors can be recomputed according to what users expect.

Recomputing happens in e.g. set_alpha, but also in update_scalarmappable, which is called directly from draw.

By using the getter, editing and then using the setter, the values used to recompute colors are updated as well.

So while I understand the functions name implies the functions behavior but I believe it might be better to state the difference explicitly as what I had stated in #27233

Would that be a good change? As mentioned in the issue I was attempting to change that but also there were a lot of get and set functions, so just to ask for some opinion