Closed newville closed 1 year ago
Same issue here, using scikit-learn-0.21.2-py36h7208079_1
from conda-forge
.
@newville , have you meanwhile found a workaround?
conda build
gives:
[...]
The package for scikit-learn located at C:\Users\angelo.peronio\AppData\Local\Continuum\miniconda3\envs\packaging\pkgs\scikit-learn-0.21.2-py36h7208079_1
appears to be corrupted. The path 'Lib/site-packages/sklearn/datasets/tests/data/openml/40675/api-v1-json-data-list-data_name-glass2-limit-2-data_version-1-status-deactivated.json.gz'
specified in the package manifest cannot be found.
conda info
gives:
active environment : packaging
active env location : C:\Users\angelo.peronio\AppData\Local\Continuum\miniconda3\envs\packaging
shell level : 1
user config file : C:\Users\angelo.peronio\.condarc
populated config files : C:\Users\angelo.peronio\.condarc
conda version : 4.7.8
conda-build version : not installed
python version : 3.7.3.final.0
virtual packages : __cuda=10.2
base environment : C:\Users\angelo.peronio\AppData\Local\Continuum\miniconda3 (writable)
channel URLs : https://conda.anaconda.org/conda-forge/win-64
https://conda.anaconda.org/conda-forge/noarch
https://repo.anaconda.com/pkgs/main/win-64
https://repo.anaconda.com/pkgs/main/noarch
https://repo.anaconda.com/pkgs/r/win-64
https://repo.anaconda.com/pkgs/r/noarch
https://repo.anaconda.com/pkgs/msys2/win-64
https://repo.anaconda.com/pkgs/msys2/noarch
package cache : C:\Users\angelo.peronio\AppData\Local\Continuum\miniconda3\pkgs
C:\Users\angelo.peronio\.conda\pkgs
C:\Users\angelo.peronio\AppData\Local\conda\conda\pkgs
envs directories : C:\Users\angelo.peronio\AppData\Local\Continuum\miniconda3\envs
C:\Users\angelo.peronio\.conda\envs
C:\Users\angelo.peronio\AppData\Local\conda\conda\envs
platform : win-64
user-agent : conda/4.7.8 requests/2.22.0 CPython/3.7.3 Windows/10 Windows/10.0.17763
administrator : False
netrc file : None
offline mode : False
Solved by enabling paths longer as 260 characters in Win10, as described at https://superuser.com/q/1119883
@newville Should this be closed?
@angelo-peronio I am not entirely sure this issue should be closed. If an OS-level workaround is needed to build with long path names, is it also possible that would also be needed on the installing machine? It seems to me like a problem that conda-build gets itself into by having (probably at least somewhat unavoidable or course) by using long path names, but that does sort of imply that packages (like sklean) cannot also have long/deep path names.
It seems like conda-build should be acknowledging and warning about this limitation. In one respect, it is reporting an error when building a package building when the error is really with a dependency -- that dependency (here, sklearn) is kind of broken in that it cannot be used as a dependency (kind of the point of packages, no?).
Hi there, thank you for your contribution!
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.
Thanks!
First, I am not 100% sure whether this is a problem with
conda-build
or with a particular package (scikit-learn
). I suspect it is a problem withconda-build
(the package installs and runs OK), but my apologies if this issue should be raised elsewhere -- please let me know if and where it should be raised.Any suggestions for a workaround or different approach would be greatly appreciated, Thanks!
Actual Behavior
With Anaconda3-2019-3 installed and updated on Windows-10 64-bit I cannot build packages that depend on scikit-learn. These all give errors similar to:
which indicates that the
scikit-learn
package is corrupted. However, the file (here at `C:\Users\newville.CARS_NT_DOMAIN\AppData\Local\Continuum\anaconda3\envs\t1\pkgs\scikit-learn-0.21.2-py37h6288b17_0\Lib\site-packages\sklearn\datasets\tests\data\openml\40675\api-v1-json-data-list-data_name-glass2-limit-2-data_version-1-status-deactivated.json.gz'does actually exist. I do not know why that file cannot be found. The name is pretty long (>256 characters, which might be suspicious) but AFAICT contains no spaces or non-Windows-friendly characters.
Expected Behavior
the package would build. To be clear, it all works on Linux and MacOS.
Steps to Reproduce
Set up the environment:
(to be clear,
lmfit
is pure python and its dependencies are also pure python). Then build try to build a package with recipe inlmfit\meta.yaml
:With this, running
conda build lmfit
gives the error above about not all of the test data files forscikit-learn
being found. I believe the problem will happen with any package that includesscikit-learn
as part of the dependency.Output of conda info
conda list
gives: