The numpy < 1.23 limitation and v4?

[ajstewart]

Hi,

I was wondering about the reasoning behind the numpy < 1.23 limitation as I looked in the release notes and searched back through the repo but it's not obvious to me why it the limitation was added recently.

In projects where I use plantcv I'm starting to notice it causing difficulties in resolving dependencies, e.g. I also ran into this issue: #1337 and it seems to be because of the numpy limitation.

The limitation still seems to be there in the v4 branch?

[nfahlgren]

At some point we ran into an issue where allowing a newer version of NumPy started failing our unit tests. I am not sure what the exact change was/is but almost certainly it is related to a change in NumPy's API or how strict it handles certain data input types. It could have definitely been related to plotnine, which we removed in favor of using altair in v4.

We won't leave NumPy restricted, but will need to do an update where we allow a newer version of NumPy and fix any issues. Likewise with Pandas v2.

[ajstewart]

Cool thanks for the answer, and great to see the 4.0 release.

Is any migration being worked on? Wondering if I could maybe help out if I had some time.

[nfahlgren]

I did some work on it today in #1347. In the PR I didn't update NumPy yet because I'm hopeful the changes to PlantCV will be backwards compatible, but I tested in a local environment with NumPy 1.25.2.

The fixes in the PR already work in NumPy 1.22+.

The remaining issues with pcv.analyze.npq are posted as a comment to the PR. I'm not immediately sure what is going on there, but looks like something going wrong between xarray, dask, and numpy. We are only doing NumPy operations with xarray data so I'm not sure why anything is interacting with dask.

@ajstewart you or anyone else are free to add to the PR. Though I realize this particular issue is pretty down in the weeds