metadata: Jinja templates are being inherited when a variable is used in calculation

[spoonerf]

I think this is a new occurrence: When you use an indicator that has Jinja metadata in a calculation (e.g., UN WPP population). The calculated variable inherits this template, and it can cause the step to fail if you don't overwrite the metadata values that are given in the template.

e.g. here in Buildkite. Buildkite is applying a template that doesn't exist in the metadata of the step, but it is the metadata for the variable used in a calculation in this step (etl/steps/data/garden/malnutrition/2024-12-16/malnutrition.py).

[spoonerf]

(Maybe it's intentional, but in that case I guess it might be good to have a function to strip variable metadata for these kind of variables)

[Marigold]

Thanks for reporting this. I've improved the error message to make it clear which variable and metadata are causing the problem.

In this specific case, the issue is that presentation.title_public is inherited from the population indicator, but it's missing the dimensions required to render the metadata. I'm not sure what the best approach is—should we stop inheriting the presentation.title_public field, or would that cause other problems?

[spoonerf]

Would a function that strips Variable metadata be a possible way to go (or maybe an option in load_dataset())?

I figure it's presentation.title_public this time, but it could be a different one with another variable (although population will be by far the most frequent one!).

[Marigold]

That's a great idea, created a PR #3728.