Revise dataset selector hierarchy?

[rod-glover]

Driven in part by a discussion arising in https://github.com/pacificclimate/climate-explorer-frontend/pull/200:

comment:

If you change the order of the selectors (swapping Variable and Emissions Scenario) you may want to change the "hierarchy" of the selectors as well to make it more intuitive.

Previously, the selectors were arranged model -> variable -> emissions from left to right. Model allowed the user to select any models. Variable allowed the user to select only variables available for the selected model. Emissions allows the user to select only emissions available for the selected model and variable. There was an implicit left to right hierarchy to the selectors such that each selector would disable all items that didn't make sense based on the "already made" choices in selectors to the left.

Now going from left to right we have model -> emissions -> variable, but we've left the item-availability hierarchy model -> variable -> emissions.

comment:

... though switching the hierarchy to model -> emissions -> variable gives you a different UI awkwardness: the user has to guess the "right" emissions scenario before they can select the variable they want, and if they select the "wrong" emissions scenario, the variable they want will be disabled.

That setup was unfortunately necessary for the dual-variable portal, since the two variable are required to share a model and emissions scenario. I assume you switched the order on this portal to make it consistent with the dual variable portal, which makes sense.

What the user really cares about is probably mostly the variable. If we had only this portal, it might make most sense to set up the selectors and hierarchy variable -> emissions scenario -> model, but there's some value in consistency across portals.

An alternative is to remove hierarchy altogether and allow the user to select combinations that result in zero selected datasets. I think the problem with this probably boils down to there being a lot of combinations that result in zero selected datasets, maybe more than the non-zero ones! It is simple to alert the user to the zero-selected condition, but I expect we would soon see a request for hierarchical selection.

[corviday]

There are many more non-existent combinations than existent ones. Here's the issue and PR from when we added the hierarchical data selection.

Maybe that would be less of an issue now that you have the dataset summary widget, but I found even testing frustrating when there were 5 or 6 nonexistent datasets for each one actually present in the data. It was just endless piles of error messages and trying to guess what data might exist.

[corviday]

There are currently 708 model/variable/emissions scenario combinations present in the ce_files ensemble, out of 2160 possible. Not nearly as bad as I remember, thankfully!

[rod-glover]

Still pretty bad; only about 1/3 of the combinations are valid. I think that makes non-hierarchical a non-starter.

[rod-glover]

Might well have been dealt with in great selector replacement, check this.