OpenSSL issues with jill installs in Python, possibly due to v1.12?

[ChrisRackauckas]

See for example diffeqpy test failures:

https://github.com/SciML/diffeqpy/actions/runs/18408384170/job/52453820503?pr=162#step:6:1097

[cjdoris]

You're on quite an old juliapkg there.

The error comes from ensuring that the version of OpenSSL is compatible between Julia and Python, but no version is compatible for the versions of Julia and Python used.

If you upgrade, it will ensure you use a version of Julia whose OpenSSL is compatible with Python.

[ChrisRackauckas]

I am forcing the latest version in SciML/diffeqpy#162 and it still happens. Is there some other requirement I have to bump?

[cjdoris]

Just looked again and see this in the logs:

juliapkg==0.1.17

That's not the latest.

[ChrisRackauckas]

I don't directly depend on it, so that's an issue with this package?

https://github.com/SciML/diffeqpy/blob/master/setup.py#L31

[cjdoris]

I don't get your point. If you require this feature then depend on it directly. I think your test environment is just out of date.

juliacall==0.9.26

is also not the latest and it is a direct dependency.

Maybe CI caching means you're getting old versions of things?

[ChrisRackauckas]

The test environment is the most up to date it can be? I don't restrict the versions at all, that's not in the repo:

https://github.com/search?q=repo%3ASciML%2Fdiffeqpy+0.9.26&type=code

The only versions I set are right here: https://github.com/SciML/diffeqpy/blob/b0ce7ce2a43866100ba7d2c447a4b9a1a20fe25b/setup.py#L31

and that's lower bounds.

[cjdoris]

Maybe there is some incompatibility between dependencies of juliacall and jill, so you're being forced back to an old version.

Maybe locally try installing them both in a venv pinned to the latest version and see what happens.

[cjdoris]

If I were you I'd just remove jill and use the install-julia GitHub action instead.

[cjdoris]

Also you should put upper bounds on your compats :)

[ChrisRackauckas]

If I were you I'd just remove jill and use the install-julia GitHub action instead.

We were trying to test this system because it's the 99% user's way of doing it.

[ChrisRackauckas]

So then it seems like jill is just bugged / old dependencies and we need to change tests to use setup-julia and tell users that manual installation of Julia is required again?

[cjdoris]

So then it seems like jill is just bugged / old dependencies

Have you actually determined this by trying?

You don't need your users to download Julia at all, juliapkg will do it for them. However the preferred system is that users install juliaup themselves and juliapkg will use that.

[ChrisRackauckas]

Have you actually determined this by trying?

Yes, that's exactly what I am showing with the CI results. It used to but it's not working anymore. If I force the installation of Julia it runs, but if I use the automated Python-side Julia installs that we suggest users do from the package, that runs into all of these OpenSSL jill versioning issues.

[asinghvi17]

What version of Julia is jill installing, vs what version is setup-julia installing?

Also it is worth noting that the version of OpenSSL_jll is locked on Julia 1.12.0 because it got turned into a stdlib for...some reason...