An attribute-based version of the `relationship` argument

[krlmlr]

What would the following syntax feel like?

inner_join(one_from(lhs), one_from(rhs)) # relationship = "one-to-one"
left_join(one_from(lhs), rhs) # relationship = "one-to-many"
full_join(lhs, one_from(rhs)) # relationship = "many-to-one"
inner_join(many_from(lhs), many_from(rhs)) # relationship = "many-to-many"

The functions would merely add a class that dplyr/duckplyr could react upon, each in their own unique way.

We could also explore extending the powerjoin package, CC @moodymudskipper.

[DavisVaughan]

Can you help me understand why duckplyr can't use the current relationship argument directly?

[krlmlr]

The default is horrible from a performance perspective. This is "just" an incompatibility, I have code that implements the checks (but not in the main branch yet).

I was confused, duckplyr can deal with this either way. I have written this also with dm in mind that could provide the relationship information too. This will be most useful if the default join generics/methods query this information and translate it. No action needed, I'm open to discussing this if this sparks interest.