Failure to derive typeclass for large sum types on Scala 3

[ghostdogpr]

If you have a sealed trait with lots of subtypes (~1000 in my case, but the issue happens with much less), Magnolia runs into max-inlines issues or even triggers a compiler stack overflow.

This comes from the fact that the subtypesFromMirror method is not tail-recursive. It folds over the list of subtypes and increases the number of nested inline function calls and the stack depth proportionally to the number of subtypes.

See this example in Caliban on how it can be made tail-recursive.

On top of that, subtypesFromMirror calls .distinctBy(_.typeInfo).sortBy(_.typeInfo.full) on every iteration, which is very inefficient with large list of subtypes. This should be moved at the end of the loop.

[adamw]

Indeed :) Would you be able maybe to work on a fix?

[ghostdogpr]

Indeed :) Would you be able maybe to work on a fix?

Not right now, I'm in the middle of a large Scala 3 migration and I ended up not using Magnolia since I could do it more simply. But I thought I'd share my findings anyway, hopefully it's relatively simple!

[adamw]

Sure thing, thanks :)