Support python 3.13

[sammorley-short]

Closes #48

[akurlej]

Not overly confident that all the packages we use are supported in the python 3.13, on all the machines we intend to build (for example, installing python 3.13 and installing pyLIQTR, one of the subpackages requires numpy v1.26.4, and running that on my m1 macbook results in errors when importing numpy).

We have an upcoming release that we are working on and hoping to release in O(week), we will see about bumping up the version/removing the upper-bound from what we currently allow, but we want to support the various environments and have things work on various machines, so we wont lift the upper-bound if it causes issues.

[sammorley-short]

Thanks for the quick response @akurlej

so we wont lift the upper-bound if it causes issues.

To be clear, I'm not advocating for lifting the upper-bound entirely—indeed I would personally advocate against such a loose bound—but rather just adding support for just 3.13, since it's now been a stable release for almost half a year.

Am somewhat surprised that you're seeing install errors on your M1, given the age of that chip and that arm support issues has mostly been resolved in common python packages. But perhaps there's something else going on there.

Anyway, let me know if this is looking likely in the next release, otherwise I will need to work around this on my side for the immediate future. Thanks!

[sammorley-short]

@akurlej It's been a few weeks since your last message. Any update on this? Thanks! 🙏

[akurlej]

Our internal testing is showing issues with the dependency resolution when using a pip install on both unix/windows, that after a fresh install using python=3.13.2, there is a downgrade of numpy/scipy to 1.26.4/1.15.2. This downgrade seems to cause some problems on both M2-era macs and windows.

Two of our developers (using mac & windows) stated that with some manual/forced updates (and removing a scipy version requirement in requirements.txt), they were able to support 3.13.2, though I haven't had he same success with my own machine. Manual is obviously not desired, so we still investigating an 'out-of-the-box' solution.

A 'fresh' install seems to work out of the box with up to 3.12.7 (windows) and 3.12.9 (mac).

Still trying to dig into whats causing issues with the dependency resolution, I suspect it may be a dependency on a fixed version of qualtran (and updating to the latest version requires nontrivial changes to pyliqtr, which would result in us delaying support for 3.13.x versions of python if true).

If we can resolve this we will add support for 3.13, but otherwise will just bump the upper bound to 3.12.x.

TLDR:

• 3.12.x seems okay
• 3.13.x causes dependency issues on one of the packages we rely on (unknown which one yet), on both mac and windows
• two of three developers said they could support 3.13.x with manual updates of numpy/scipy (and type-extensions on windows)
• and O(week) release is pending internal review before we push it to the public repo. TBD on how high we can lift the python version in this upcoming update.

[sammorley-short]

Thanks for the update!

stated that with some manual/forced updates (and removing a scipy version requirement in requirements.txt), they were able to support 3.13.2

I am not sure what you mean by "manual/forced updates" here?

Also, since I am requesting the update, I am happy to try and fix this here. Although to do so I will need some instructions on how to reproduce your test environment so I can debug any issues. If you're able to add a "Contributing" section to the README, then that would be really helpful and would allow others to better contribute to the tool's development 🙂

Cheers!

[sammorley-short]

@akurlej any update on this? Cheers!