fix: constrain dxchange version to fix ZeissDataReader

[JeppeKlitgaard]

Ensures that fix to #2256 released in dxchange version 0.2.1 is used by setting lower version constraint.

Fixes #2256

[JeppeKlitgaard]

This is ready for review

[hrobarts]

Thanks for this @JeppeKlitgaard!
I tested this with our Zeiss example and it works as expected so I'm happy to approve

I guess the version needs updating here as well
https://github.com/TomographicImaging/scripts/blob/main/env/cil_demos.yml
https://github.com/TomographicImaging/scripts/blob/main/env/cil_demos_cpu.yml
https://github.com/TomographicImaging/scripts/blob/main/env/cil_test.yml

[JeppeKlitgaard]

Thanks for this @JeppeKlitgaard! I tested this with our Zeiss example and it works as expected so I'm happy to approve

I guess the version needs updating here as well TomographicImaging/scripts@main/env/cil_demos.yml TomographicImaging/scripts@main/env/cil_demos_cpu.yml TomographicImaging/scripts@main/env/cil_test.yml

Yes, that would be preferable. In principle it will resolve to the fixed version anyway unless something else constrains the version like a lock file or other dependency.

I think these should also be changed to using https://tomography.stfc.ac.uk/conda instead of the ccpi channel at some point, as part of #1870.

I am happy to submit a PR over there as well to pin these versions.

[hrobarts]

Thanks again @JeppeKlitgaard, happy to merge this now

[casperdcl]

oops...

conda create -n test "dxchange>=0.2.1" "python>=3.13"

fails for me.

[JeppeKlitgaard]

oops...

conda create -n test "dxchange>=0.2.1" "python>=3.13"

fails for me.

dxchange is only on conda-forge, so you need:

conda create -n test -c conda-forge "dxchange>=0.2.1" "python>=3.13"

The error is because dxchange sets an upper constraint on the scipy version, which sets an upper constraint on the Python version. The error is somewhat unrelated to the change here.

It should be noted that conda create -n test -c conda-forge "dxchange" "python>=3.13" does not fail, which is because it can resolve to dxchange==0.1.8, which does not constrain the scipy version and thus does not constrain the Python version. I would caution against this, as I expect the upper constraint was later added because dxchange (including version 0.1.8) does in fact not work with the later scipy versions, but just does not correctly declare this prior to version 0.1.9. If this is the case, the existing setup without this PR would probably produce an environment with a broken dxchange.

AFAIK current maximum Python version supported by CIL is 3.12, so I think this could be best fixed in a future PR relating to supporting higher versions of Python.

TL;DR: I would argue this is not related to this PR, but is a separate issue, namely that dxchange does not support Python 3.13. This would need to be fixed upstream in any case.