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
{{ message }}
Repository navigation
Document that python setup.py develop add the symlink to easy-install.pth #7422
It looks like pip install -e . / pip uninstall matplotlib handles this fine. I'd suggest switching the docs to recommending using those (specifically pip install --upgrade --no-deps -e ., which is the more practical set of flags I think).
I'd rather move the pip-based method first (and explicitly recommend it), given, again, that it does not suffer from the issue @NelleV mentions. (But it is only better if you remember to include the --upgrade (and possibly --no-deps) flags.)
I think --upgrade is dangerous; at least in some versions of pip, it upgrades all dependencies without the courtesy of even listing them and then asking if that is what you really want to do. Does --no-deps reliably take care of that, or is it only available in newer versions of pip? After being bitten by the unwanted upgrades, I switched to always uninstalling first, so --upgrade is not needed.
Apart from that, I agree, the pip methods should be emphasized because they handle uninstalling.
I never had this issue while using --no-deps. I have no idea about whether this was broken on previous versions but we can always first suggest pip install --upgrade pip and then we only have to care about the latest pip, which I'm certain implements this correctly :-)
fyi, I am against moving the pip command before the python setup.py install.
It is not a standard way of installing python packages. I would suggest waiting until that is the case (as IMO it is very unlikely that it will ever become the standard way to install python packages).
I have been test-driving always using pip for installing things from source for the last few months and am pretty happy. It is less to type and reliably uninstalls (both develop and normal installs).
Going down this path also offers a sliver of hope that the setptools/distutils Gordian knot will ever be untangled (by localizing it all in pip first).
I am comfortable with mpl being on the leading edge of this shift.
Activity
anntzer commentedon Nov 8, 2016
It looks like
pip install -e ./pip uninstall matplotlibhandles this fine. I'd suggest switching the docs to recommending using those (specificallypip install --upgrade --no-deps -e ., which is the more practical set of flags I think).tacaswell commentedon Nov 8, 2016
Both methods are currently in the docs as of #7229
anntzer commentedon Nov 8, 2016
I'd rather move the pip-based method first (and explicitly recommend it), given, again, that it does not suffer from the issue @NelleV mentions. (But it is only better if you remember to include the
--upgrade(and possibly--no-deps) flags.)efiring commentedon Nov 8, 2016
I think
--upgradeis dangerous; at least in some versions of pip, it upgrades all dependencies without the courtesy of even listing them and then asking if that is what you really want to do. Does--no-depsreliably take care of that, or is it only available in newer versions of pip? After being bitten by the unwanted upgrades, I switched to always uninstalling first, so--upgradeis not needed.Apart from that, I agree, the pip methods should be emphasized because they handle uninstalling.
QuLogic commentedon Nov 8, 2016
In a conda env, I've never had to use
--upgrade; not too sure why that is though.anntzer commentedon Nov 8, 2016
I never had this issue while using
--no-deps. I have no idea about whether this was broken on previous versions but we can always first suggestpip install --upgrade pipand then we only have to care about the latest pip, which I'm certain implements this correctly :-)petehuang commentedon Jan 6, 2017
So the latest on this is that we feel positive about two changes in https://github.com/matplotlib/matplotlib/blob/master/doc/devel/contributing.rst:
pip install --upgrade pipto ensure that all pip commands on the page run wellpip installmethod first (beforepython setup.py develop). Remaining question is which exact variant we want? Thoughts?NelleV commentedon Jan 6, 2017
fyi, I am against moving the pip command before the python setup.py install.
It is not a standard way of installing python packages. I would suggest waiting until that is the case (as IMO it is very unlikely that it will ever become the standard way to install python packages).
efiring commentedon Jan 6, 2017
@NelleV, in https://docs.python.org/3/installing/, it says,
pip is the preferred installer program. Starting with Python 3.4, it is included by default with the Python binary installers.
You would maintain that although it is officially "preferred" it is not "standard", and is unlikely to become so?
tacaswell commentedon Jan 7, 2017
Also see https://pip.pypa.io/en/stable/reference/pip_install/#usage .
This is also how IPython recommends installation http://ipython.readthedocs.io/en/stable/install/install.html?highlight=setup.py#installation-from-source
I have been test-driving always using pip for installing things from source for the last few months and am pretty happy. It is less to type and reliably uninstalls (both develop and normal installs).
Going down this path also offers a sliver of hope that the setptools/distutils Gordian knot will ever be untangled (by localizing it all in
pipfirst).I am comfortable with mpl being on the leading edge of this shift.