Package breakage due to change in dependency

Tooling Package Management
1

Hi Julia folks,

I have a registered package that used to install PyPlot with a build.jl in a deps/ folder as follows:

import Pkg, Conda
@info "building!"
Conda.pip_interop(true)
Conda.pip("install", "matplotlib==3.7.3")
Conda.add("matplotlib")
ENV["PYTHON"] = joinpath(Conda.ROOTENV, "bin", "python")
Pkg.build("PyCall")
@info "built PyCall!"

I switched to pip_interop because of Conda errors a few years ago (which have since been fixed), but I never changed the above code in a few subsequent releases.

Now, because of recent Conda.jl changes that no longer use use pip_interop I’m having trouble with previous releases not installing.

In the latest release, I just remove the Conda.pip parts of the above build script and it works fine.

What is a good way to deal with this in previously released, registered versions?

2

That’s why there are project environments—to keep all the dependencies synched. If your registered package used to work it should continue to in the same environment. You should have seen a log screen with the conflicts, which may help in creating a fresh environment.

3

This is unfortunately, not the solution (or it may be part of one). While I could possibly instantiate a Manifest.toml for the exact environment that previously worked, it does not stop those who wish to install the previous version from running into the pip_interop errors. The build log file shows errors, even for a fresh julia install and my package just as before:

WARNING conda.cli.main_config:_set_key(456): Key pip_interop_enabled is an alias of prefix_data_interoperability; setting value with latter
[ Info: building!
[ Info: Downloading miniconda installer ...
[ Info: Installing miniconda ...
[ Info: Running `conda config --set pip_interop_enabled true --file /home/131/ts7941/.julia/conda/3/x86_64/condarc-julia.yml` in root environment
ERROR: LoadError: pip_interop is not enabled
Use `Conda.pip_interop(true; [env::Environment=ROOTENV])` to enable

Any thoughts? I could release a minor version with the Conda.pip parts removed from the build.jl but was wondering if there are other suggestions?

1 Like
4

Ah, sorry I don’t have anything further to suggest. In all my years with using Conda outside Python I’ve had scant luck.