Registrator.jl sudden conda dependency error on GitHub

Hi all,

I have been releasing new versions of my package using Registrator.jl without any trouble until May 24th of this year. On my latest release, saying

@JuliaRegistrator register

as a comment to the last GitHub commit to trigger package registration fails with this error on GitHub/JuliaRegistries

[ Info: Installing matplotlib via the Conda matplotlib package...
[ Info: Running `conda install -y matplotlib` in root environment

  - conda==23.1.0

ERROR: LoadError: InitError: failed process: Process(setenv(`/tmp/jl_DShtXf/conda/3/x86_64/bin/conda install -y matplotlib`,["PYTHONIOENCODING=UTF-8", "PATH=/opt/hostedtoolcache/julia/1.9.2/x64/bin:/home/runner/.local/bin:/opt/pipx_bin:/home/runner/.cargo

Has something changed on GitHub / JuliaRegistries? I do have PyPlot as a listed dependency in Project.toml and consequently matplotlib is installed via conda and nothing has changed there. Any pointers will be much appreciated.

It is a little strange, I fixed it by adding a deps/build.jl, installing matplotlib through Conda.pip and then again, through Conda.add, before pointing PyCall to Conda’s python. None of which had I needed earlier for PyPlot as a dependency. See here for the fix

My sense is that the first thing you should do is to figure which Conda environment and which Python you want to use, and then ensure that environment you want to make sure has matplotlib.

If you do not do this, which Python gets chosen?

It also might be useful to see if depending on an older version of PyCall or Conda resolves the issue.