SciMLBasePythonCallExt always precompiling, CondaPkg 0.2.25 regression

[jaakkor2]

On Windows, julia +lts --startup-file=no --project=debug , pkg> add PythonCall SciMLBase CondaPkg#compare-load-path.

With Julia 1.10.8, versions are

  [992eb4ea] CondaPkg v0.2.25 `https://github.com/JuliaPy/CondaPkg.jl.git#compare-load-path`
  [6099a3de] PythonCall v0.9.24
  [0bca4576] SciMLBase v2.75.1

Then every time I start Julia, SciMLBasePythonCallExt gets precompiled.

julia> using PythonCall

julia> @time using SciMLBase
Precompiling SciMLBasePythonCallExt finished.
  1 dependency successfully precompiled in 3 seconds. 68 already precompiled.
[ Info: Precompiling SciMLBasePythonCallExt [2797fd30-2078-5027-980c-4c2c8a19c528]
  5.792371 seconds (4.77 M allocations: 370.907 MiB, 3.59% gc time, 25.51% compilation time)

This does not happen on CondaPkg 0.2.24.

I used https://github.com/JuliaPy/CondaPkg.jl.git#compare-load-path because that fixed #166 for me.

[mestinso]

I was just about to post this issue as well. I'm on windows, and since 0.2.25, my packages that use PythonCall get precompiled every time. For example, when I run "instantiate" the packages that depend on PythonCall always precompile.

[Moelf]

Arguably pixi backend should have been a breaking change

[mestinso]

Thanks, resolved here!