Fix error in image about linspace

[nvaytet]

We add .value to sc.islinspace.

This was tripping up some builds in Scippneutron: https://github.com/scipp/scippneutron/actions/runs/24816917140/job/72633086173

[SimonHeybrock]

We add .value to sc.islinspace.

This was tripping up some builds in Scippneutron: https://github.com/scipp/scippneutron/actions/runs/24816917140/job/72633086173

The error says:

2026-04-23T04:34:50.8128569Z >       if (canvas.ax.name != 'polar') and all(
2026-04-23T04:34:50.8128837Z             (data.coords[dim].ndim < 2)
2026-04-23T04:34:50.8129196Z             and ((data.coords[dim].dtype == str) or (sc.islinspace(data.coords[dim])))
2026-04-23T04:34:50.8129764Z             for dim in data.dims
2026-04-23T04:34:50.8129977Z         ):
2026-04-23T04:34:50.8130306Z E       scipp._scipp.core.UnitError: The truth value of a variable with unit is undefined.

But why would sc.islinspace return a variable with unit? It should be dtype=bool with no unit (unit=None), shouldn't it?

[nvaytet]

I don't understand why it's talking about units. I looked a little bit into it, tried to blame to see when it changed.
I didn't find anything conclusive.

On one side, I'm wondering if it's revealing some edge case in scipp or islinspace. On the other side, I'm thinking I can just add the .value and move on, instead of spending some hours digging deep into the issue...

I looked at the versions: before it was failing, it was installing

plopp==24.9.1 scipp==24.8.0

Now it's installing

plopp==26.4.1 scipp==24.8.0

Note that only plopp has changed.

At least it's claiming that:

• https://github.com/scipp/scippneutron/actions/runs/24816917140/job/72633086173#step:4:150
• https://github.com/scipp/scippneutron/actions/runs/24816917140/job/72633086173#step:4:178

Because I tried to create an environment with the same versions locally, and I get

mamba env create -n oldscipp python=3.12.* scipp::scipp==24.8.0 plopp==26.4.1

conda-forge/linux-64                                        Using cache
conda-forge/noarch                                          Using cache
scipp/linux-64                                              Using cache
scipp/noarch                                                Using cache

error    libmamba Could not solve for environment specs
    The following packages are incompatible
    ├─ plopp ==26.4.1 * is installable and it requires
    │  └─ scipp >=25.5.0 *, which can be installed;
    └─ scipp ==24.8.0 * is not installable because it conflicts with any installable versions previously reported.
critical libmamba Could not solve for environment specs

Only difference is I am using conda and not uv, but the requirements should work in the same way?

[SimonHeybrock]

Maybe plopp did indeed not use this code branch? Anyway, Claude found that this was indeed a bug and fixed in scipp 24.11.0 by 75308cc66 ("fix: make islinspace unit none", PR scipp#3533). Ths changed the element-op unit lambda from units::one to units::none.

I suppose you should bump the minimum Scipp version instead of applying this?

[nvaytet]

Should islinspace just return a bool instead of a Variable with unit None? Or is it so that the behaviour is consistent with other similar functions?

I think the changes here don't hurt in any case. But I will bump scipp minimum version in scippneutron.

[nvaytet]

Only difference is I am using conda and not uv, but the requirements should work in the same way?

I understand now what happens: in Plopp's pyproject, we did not include Scipp as a dependency, only in the plopp[scipp] or plopp[all].
However, in conda, we have just a run_constraint: https://github.com/conda-forge/plopp-feedstock/blob/main/recipe/recipe.yaml#L35

That is out of date with our pyproject.toml: https://github.com/scipp/plopp/blob/main/pyproject.toml#L40

I think we didn't add scipp in the deps, because Scipp also has plopp as an optional runtime dependency, and we thought we would end up in a circular dependency where plopp needs scipp but scipp needs plopp.
However, in practice maybe it just doesn't matter?

Maybe Plopp should just have Scipp as a core dependency? (It's using it internally everywhere)
@MridulS do you have any insights here?

[nvaytet]

ChatGPT says:

• make Scipp a hard dependency of Plopp
• keep Plopp as a soft dependency of Scipp

Then the pip, uv, conda should still happily solve the environments.

• if you pip install plopp, you would get plopp and scipp (and skip scipp if it is already installed)
• if you pip install scipp, you get only scipp
• if you pip install scipp plopp you get both at once, as expected.

[nvaytet]

I will close because in the future, if we change islinspace to return a bool instead of a Variable with unit None, then things would break if this is merged.