Small changes to help conda feedstock

[JeppeKlitgaard]

I am trying to update and improve the conda feedstock maintained by the CIL team at https://github.com/TomographicImaging/TIGRE-conda.

As part of using as much of the TIGRE build-chain as possible, I am hoping to use Pixi to build the project, but for this to go smoothly it would require 2 changes:

• Change the project name in pyproject.toml from pytigre to tigre to match the module name
  • I think this should not require many changes outside .gitignore and pyproject.toml
• Cutting a new minor release (pyproject.toml location and layout has changed significantly since 3.0.0)

Alternatively I would be happy to help set up a TIGRE conda build using Pixi. For reference, the following seems to be all that is required and would allow you to keep the build environment defined in pyproject.toml:

# pixi.toml
[workspace]
channels = ["conda-forge"]
platforms = ["linux-64", "win-64"]
preview = ["pixi-build"]

[system-requirements]
cuda   = "12"

[package.build]
backend = { name = "pixi-build-python", version = "*" }

[package.build.config]
ignore-pypi-mapping = false

There is an existing related issue which is asking for an official conda release, #245, which is unrelated to this.

[AnderBiguri]

Hi @JeppeKlitgaard ,
I see no reason to go against these changes, feel free to do a PR , and I can make a new minor release once is done.
However I try to keep tigre as build agnostic as possible, which has worked quite well so far, so I rather not add the pixi requirements in TIGRE itself, if that is OK.

[JeppeKlitgaard]

@AnderBiguri I have added the PR in #709.

It is absolutely understandable that you want to avoid having this in-tree. With this small change I think it will be fairly simple to build and ship our own version of TIGRE for use in CIL.

[AnderBiguri]

@JeppeKlitgaard should I just make 3.1.0 release then, now that I merged the PR?

[JeppeKlitgaard]

I think that would be enough for us to start targeting this repo in the feedstock. Thank you for sorting this so quickly!

Let me know if #709 somehow broke something and I will get right on that.

[AnderBiguri]

@JeppeKlitgaard fantastic. FYI I will merge #710 too, that changes slightly the lock.

[AnderBiguri]

@JeppeKlitgaard done!

[AnderBiguri]

@JeppeKlitgaard BTW, I see you are nearby, if you are using TIGRE and want to have a chat, do feel free to send me an email at some point and we can chat in person. If you want, of course :)

[JeppeKlitgaard]

I think unfortunately I have just not updated my GitHub profile – I am actually in Copenhagen now, but I might be around to grab myself one of those MA's soon and would love to pop by your office then! At time of writing I am at the CIL user meeting in RAL.

I will leave the issue here open while I try to figure out of there is anything else that might block a new release of our TIGRE conda build, if you don't mind.

[JeppeKlitgaard]

@AnderBiguri

We would also like to eventually target newer Python versions, see #713.

Additionally I think the version in the pyproject.toml file was not updated for the 3.1.1 release (it is still 3.1.0).

[AnderBiguri]

Ah, sorry, I rushed the release because I am slightly tight in time.

I merged your PR, and can draft another minor/minor version release with the pyproject.toml file updated, if you want. Or maybe you prefer me to wait until the end of the hackathon so I can do that will all code that may be submitted?

Apologies, I generally just use pip to install TIGRE, so I am not much used to the pyproject files etc.

[JeppeKlitgaard]

No worries at all, I am nothing but impressed with the ease of contributing to TIGRE – I am happy either way, I can also just base my build setup off my fork as I experiment, so it is not critical for me to get things merged in quickly.

I expect it will take some time before CIL actually will get up to speed and use the latest TIGRE version.

Have you experimented or had any problems relating to numpy 2 at all?

[AnderBiguri]

We recently had issues with numpy 2.4.0, but as far as I know #703 fixed them.

I will wait until next week to make a release then (3.1.2, in case you want CIL to be prepared for that) to merge anything that comes this week and do some tests myself.

[JeppeKlitgaard]

I was a bit too fast on my feet when bumping the python versions and accidentally included Python 3.10, which has also reached EOL last year. I have made a new PR #715 that amends this.

I think we wouldn't require any more changes from the CIL side of things, so feel free to cut a new release, which we can then build and release as a conda package.

If you get requests for conda distribution in the future, feel free to point them towards https://tomography.stfc.ac.uk/conda, which should soon feature recent TIGRE versions as well. No promises made that it will feature builds of all versions or kept entirely up to date outside the CIL update schedule though.