ppk42
commented
2 years ago The fix is documented in the PR https://github.com/SheffieldML/GPy/pull/956 and by the commits coments. TLDR; The outdated miniconda version 36 has issues with the conda update -q conda command. Fix: Use the C:\Miniconda3-x64\ folder then, conda update works. But, for python 3.7 and above a higher matplotlib version is used. Because of changes in the matplotlib the build failed for Python 3.7 but not for 3.8, 3.9 (probably caused by changes in the parser/compiler PEP570??). I decided to freeze the matplotlib version to 3.0.0 for python 3.5 and 3.3.4 for python 3.6, 3.7, 3.8 and 3.9. From what I saw in the code there is a problem in GPy with the wrong matplotlib version that is not covered by tests. That should be further investigated.
ekalosak
I was aware of pull request https://github.com/SheffieldML/GPy/pull/951. One of the prerequisites that must be fulfilled is a AppVeyor buld for the Windows version. The build only works for Python 3.5 but not for Python versions 3.6 up to 3.9.
The build is essential for further development. If it will not work, no pull requests can/should be accepted.
With the known facts, it seems that the build did not fail because of the code changes. There is evidence for the assumption that the build failed because of incompatibilities of the miniconda version and the
conda update condacommand.It is remarkable that the working case and the not working cases use different installation folders of the miniconda environment and, all not working cases use the same miniconda installation.
I would like to check these thoughts by forking the repo and adding pull requests with changes only for the appveyor.yml file that hopefully bring appveyors build to life again. ;-). It could also be possible that the ApVeyor configuration must be changed (e.g. another Visual Studio VM), but that's beyond my means. Only the owner of the AppVeyor project is granted to change it.