Currently, expect.same and expect.eql require two things : a cats.Eq and a cats.Show, which implies that the user has no power whatsoever over the rendering that happen when the comparison is negative.

These methods should be amended to take a sort of Comparison abstraction, that should be roughly isomorphic to

trait Comparison[A] {
   def diff(received: A, expected: A) : Option[String] 
} 

where None would indicate equality/equivalence, whereas Some would imply differences, and would contain a textual representation of the differences. This approach would offer a greater flexibility for users to inject different rendering logic based on the data types being compared. For instance, JSON Merge patch could be a suitable way to render of diff JSON values.

A default implicit instance of the comparison could be provided, relying on Eq and Show as before, but using some better algorithm than what's currently available in weaver-test 0.8.x. The myers diff algorithm, currently used by munit, seems to be an interesting contender.

From my shared experience with @keynmol having to rely on a third party like expecty for this kind of thing, I think it'd be preferable to roll our own implementation, in particular if it's reasonably approachable.