Visualize schemas as entity-relationship diagrams

[kasbah]

Adds entity-relationship diagram analysis for simple schemas that uses Graphviz HTML-like tables and crows-foot edges.

https://branch-918-merge--catcolab.netlify.app/analysis/019b3206-c98c-7f40-a3df-027faf6ee5a6

[kasbah]

• Possibly this should be called a "Entity relationship graph" to avoid confusion with diagrams as we currently have them?
• Does it make sense to have this visualization as a diagramAnalysis? I am not familiar enough with diagrams to explore that.

[epatters]

Thanks Kaspar, this is great! A few small comments below.

[epatters]

Possibly this should be called a "Entity relationship graph" to avoid confusion with diagrams as we currently have them?

Yeah, I dunno. "ER diagram" is the standard term, so I'm inclined to go with it despite the possibility of confusion. In a different direction, @tim-at-topos and I have discussed renaming the "diagrams" in CatColab to "diagrammatic instances," which is longer but less confusing and suggests the connection with instances in general. At the root of all this is the fact that diagram is a technical term in category theory that is inspired by, but also clashes with, the colloquial meaning of "diagram."

Does it make sense to have this visualization as a diagramAnalysis? I am not familiar enough with diagrams to explore that.

I think not. You can think of diagrams in CatColab as more like knowledge graphs, displaying particular entities and their relationships, rather than the general schema of database/knowledge graph as in an ER diagram.

[epatters]

Since this code is using the elaborated model, but regardless of whether it validates, the model can be ill-typed if the user has made a mistake. Thus, the safe thing is to replace these invariants with conditionals of the form

if (!(...)) {
    continue;
}

Alternatively, you could only show the visualization when the model validates, in which case it would be correct to use invariants, but IMO it's better to make visualization analyses degrade gracefully in the face of user error when possible.

[kasbah]

Ok, thanks. I briefly tried to create an invalid model but failed to do so, so I jumped to the conclusion that it wasn't possible. I've gone back to using an if block.

[epatters]

Thanks. Yeah, I actually thought this code would trigger errors and was surprised when it didn't during my testing. I think this has to do with recent changes to the elaborator for models of discrete theories (now using DoubleTT rather than the previous, more direct method) which is causing invalid arrows to be filtered out during elaboration when they weren't before. However, this behavior should be considered accidental and isn't guaranteed to persist. The general rule is that you can't assume a model to be valid/well-typed unless it passes validation.