You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I just noticed that the wheels on PyPI still set a default backend in their matplotlibrc (to tkagg on linux/windows and to macosx on macos). I think this should change, and we should instead omit the default backend and trigger fallback.
Currently, a linux/windows user who installs matplotlib from pypi and does not bother writing a matplotlibrc always gets tkagg, whereas backend fallback is triggered if (and only if) they explicitly write a matplotlibrc and explicitly ignore the backend: ... entry in it. This seems backwards: I'd say backend fallback is here to help the "unconfigured" case, more than the "you have a config but ignored a key" case. This also invalidates the relevant 3.0 whatsnew entry (https://matplotlib.org/users/whats_new.html#improved-default-backend-selection) in the default case of a PyPI install.
(Also, setting the default backend to macosx on macos is not really helpful -- backend fallback will already choose to use that backend if that works, and, if it doesn't (assuming nonframework builds are still an issue, which I'm not certain about), will perform the required fallback.)
If downstream redistributors want to set a default backend (#11844) that's their problem, but I think setting it on the PyPI wheels defeats some of the purpose of having backend fallback to start with...
Matplotlib version
Operating system: any
Matplotlib version: 3.0+
Matplotlib backend (print(matplotlib.get_backend())): that's the question
Python version: any
The text was updated successfully, but these errors were encountered:
This should be the case for the rc2 wheels that were just uploaded this morning. I'm going to close this, please ping to re-open if the rc2 wheels do not have this behavior.
I just noticed that the wheels on PyPI still set a default backend in their matplotlibrc (to tkagg on linux/windows and to macosx on macos). I think this should change, and we should instead omit the default backend and trigger fallback.
Currently, a linux/windows user who installs matplotlib from pypi and does not bother writing a matplotlibrc always gets tkagg, whereas backend fallback is triggered if (and only if) they explicitly write a matplotlibrc and explicitly ignore the
backend: ...
entry in it. This seems backwards: I'd say backend fallback is here to help the "unconfigured" case, more than the "you have a config but ignored a key" case. This also invalidates the relevant 3.0 whatsnew entry (https://matplotlib.org/users/whats_new.html#improved-default-backend-selection) in the default case of a PyPI install.(Also, setting the default backend to macosx on macos is not really helpful -- backend fallback will already choose to use that backend if that works, and, if it doesn't (assuming nonframework builds are still an issue, which I'm not certain about), will perform the required fallback.)
If downstream redistributors want to set a default backend (#11844) that's their problem, but I think setting it on the PyPI wheels defeats some of the purpose of having backend fallback to start with...
Matplotlib version
print(matplotlib.get_backend())
): that's the questionThe text was updated successfully, but these errors were encountered: