PythonCall + CondaPkg depends on order for correct package loading

[icweaver]

Following up on a discussion with @cjdoris's on slack, I'm recording an issue I happened to run into when working with PythonCall/CondaPkg in some of our CI tests:

Edge case: The order of using PythonCall and using Conda can lead to downloaded python packages being looked for in the wrong place, based on platform and python package manager

Summary from brief tests used on a sample Julia package:

1. Using pixi with PythonCall loaded before CondaPkg: fails on macos ❌

ENV["JULIA_CONDAPKG_BACKEND"] = "Pixi"

using PythonCall
using CondaPkg
CondaPkg.add("numpy")

... <importing numpy not found>

2. Using micromamba with PythonCall loaded before CondaPkg: fails on windows ❌

ENV["JULIA_CONDAPKG_BACKEND"] = "MicroMamba"

using PythonCall
using CondaPkg
CondaPkg.add("numpy")

... <hits permissions issue>

maybe related to mamba-org/mamba#2705?

3. Using either pixi or micromamba with PythonCall loaded AFTER CondaPkg: passes on all platforms ✅ Thanks @cjdoris!

[natgeo-wong]

Hi!

I seem to be running into a similar issue here:

julia> using CondaPkg

julia> CondaPkg.add("ultraplot")
    CondaPkg Found dependencies: /home/nw2648/.julia/environments/v1.11/CondaPkg.toml
    CondaPkg Found dependencies: /home/nw2648/.julia/packages/PythonCall/L4cjh/CondaPkg.toml
    CondaPkg Dependencies already up to date

julia> using PythonCall

julia> uplt = pyimport("ultraplot")
ERROR: Python: ModuleNotFoundError: No module named 'ultraplot'
Python stacktrace: none
Stacktrace:
 [1] pythrow()
   @ PythonCall.Core ~/.julia/packages/PythonCall/L4cjh/src/Core/err.jl:92
 [2] errcheck
   @ ~/.julia/packages/PythonCall/L4cjh/src/Core/err.jl:10 [inlined]
 [3] pyimport(m::String)
   @ PythonCall.Core ~/.julia/packages/PythonCall/L4cjh/src/Core/builtins.jl:1561
 [4] top-level scope
   @ REPL[4]:1