Pass resolved_type into legacy_invalid_empty_selections_on_union

[ravangen]

Objective: Allow for legacy_invalid_empty_selections_on_union hook to know which type had an empty selection set. This way we can log and report to the client what the offending part of the operation was.

This is backwards incompatible change, so I'm not sure how you feel about it. I imagine this API isn't widely used yet.

This functionality was introduced in #5322 (2.5.3).

[rmosolgo]

I'm definitely game to support this but I'd like to not 💥 people who have already implemented it. How about accomodating that case somehow? A few possibilities come to mind:

• Check arity before calling it and warn if it only receives one argument (then only pass one)
• Implement GraphQL::Schema.method_added to handle this method and crash if it's defined with only one argument. That would at least move the 💥 to development
• Add a wrapper method, eg ..._with_type, call that method from runtime code instead, and implement that method to call the legacy hook by default (if defined?). Also update docs and warnings to name the new method instead of the old one

If you have another suggestion on how to avoid disruption, I'm certainly open to it -- I'm just offering those in case they're helpful. What do you think?

[ravangen]

I went with the wrapper method approach as it felt simplest.

Validation rule calls legacy_invalid_empty_selections_on_union_with_type which by default calls legacy_invalid_empty_selections_on_union.

The doc strings advocate for overriding legacy_invalid_empty_selections_on_union_with_type, but the original still works.

[rmosolgo]

Thanks fort this improvement!

[rmosolgo]

🚢 in 2.5.15!