channel priority sometimes messed up #6065
Comments
|
What do you get for |
|
For python 3.5 i get: For python 3.6 its correct: |
|
Ok that's interesting. I'll try to help track this down. Can you also give the output of |
|
Voila |
|
I ran into this just now, where defaults is jumping priority for indirect dependencies. In this example, I used boost, but implicitly-installed numpy will always come from defaults instead of the higher-priority conda-forge channel unless requested on the command-line, or if one of the requested packages requires openblas. Notably: version numbers are identical and build number for conda-forge is higher, so it seems that channel priority isn't tied, but rather inverted somehow. I could imagine that features are somehow relevant, since the conda-forge numpy has a blas_openblas feature, while the defaults channel numpy has no features. Example using the miniconda docker image: |
|
Indeed it does appear to be feature-related. When I commented-out the feature-minimization here, it results in the expected (by me) solution. From my understanding of the solver, it would appear that satisfying channel priority either has equal or lower weight to eliminating a feature. |
|
And the difference between specifying |
|
|
|
Good point, on reading more carefully, the bug is indeed that versions/channels are optimized first on requested packages, then features are optimized out, then versions/channels are optimized for dependency packages, precisely T3 in your PR. |
|
Hi there, thank you for your contribution to Conda! This issue has been automatically marked as stale because it has not had recent activity. It will be closed automatically if no further activity occurs. If you would like this issue to remain open please:
NOTE: If this issue was closed prematurely, please leave a comment and we will gladly reopen the issue. In case this issue was originally about a project that is covered by the Anaconda issue tracker (e.g. Anaconda, Miniconda, packages built by Anaconda, Inc. like Anaconda Navigator etc), please reopen the issue there again. Thanks! |
|
Hi again! This issue has been closed since it has not had recent activity. Thank you for your contribution. |
github-actions
bot
added
the
stale::closed
Installing
cythonarrays
with
from channel MaxBo
failes on my local computer and on the ci-server for python 3.5 on windows and linux.
Installing for python 3.6 works fine.
There seems to be a problem with the conda-channels on python 3.5, but not on python 3.6.
My channels are in the following priority:
set by
I figured out, that on python 3.5,
conda install cythonarrays installes the followning dependencies:
So the package xarray is insalled from the defaults-channel and not from conda-forge.
The problem is, that xarray from the defaults-channel does not install several dependencies. I first guessed, that the xarray-package is missing on conda-forge for python 3.5.
But when i try to install xarray directly, it is taken from conda-forge and installes the required dependencies:
In python 3.6, installing cythonarrays works fine, because xarray is taken from the right channel:
I checked the channel priority, ensuring that the channel priority should take place with
conda config --set channel_priority trueI have no idea, why xarray from the defaults-channel is preferred over conda-forge on python 3.5, but not on 3.6. Am i making a mistake or what could be the reason for this behaviour?
Thanks for any support,
Max
The text was updated successfully, but these errors were encountered: