Right now we typically don't have lower bounds on our dependencies. To make the lower bound tests pass we need to set some lower pin on most of our dependencies because otherwise some of them resolve to something ridiculous like 0.0.1-alpha-use-at-your-own-peril.

It's hard to find the true "lower bound" combination of all dependencies that passes the tests (an interesting problem though) but in practice I think it makes sense to just set all bounds to something like the latest release at least 1-2 years old, and if it doesn't pass the tests there then manually fiddle until the tests pass.

For example, here is a combination of lower pins I found that lets the essdiffraction tests pass:

dependencies = [
  "dask>=2024.7.0",
  "essreduce>=25.05.3",
  "graphviz",
  "numpy>=1.25",
  "plopp>=25.03.0",
  "pythreejs>=2.4.1",
  "sciline>=25.04.1",
  "scipp>=25.05.1",
  "scippneutron>=25.02.0",
  "scippnexus>=23.12.0",
  "tof>=25.01.2",
]

[project.optional-dependencies]
test = [
    "pandas>=2.2.2",
    "pooch>=1.5",
    "pytest>=7.0",
    "ipywidgets>=7.8.2"
]

Interesting, that graphviz has no lower bound. So we can use the initial release?

in practice I think it makes sense to just set all bounds to something like the latest release at least 1-2 years old, and if it doesn't pass the tests there then manually fiddle until the tests pass.

That sounds reasonable. We don't need to support multi-year-old dependencies. SO just picking a version as you described should work well.

Did you deploy the project and ran the CI

No I just tested it locally first.

I'll try it in the CI in a few minutes, see scipp/essdiffraction#189.

Interesting, that graphviz has no lower bound. So we can use the initial release?

It might be constrained by other dependencies already, for example maybe sciline has a lower pin. I'm not sure.

The workflow ran in essdiffraction: https://github.com/scipp/essdiffraction/actions/runs/16293424662/job/46008694444

(the regular tests fail there, but that is unrelated)