Supralinear complexity with UnionParser

[amrothemich]

Hi, I have found that adding items to a JSON schema (e.g. increasing the length of an enum) increases the time ~linearly. However, it seems like adding Unions of different types of nested objects to the schema increases the complexity supralinearly. e.g. I can add an enum with 3000 items for a pretty trivial slowdown but a union of 3000 classes (with just one string field) leads to an intractable slowdown.

I see the comment

        # This is a bit of a performance hit, as it means get_allowed_characters() is called twice.

but this shouldn't cause a supralinear change, right? I'm looking into this myself but any support would be appreciated.

(Also this library is amazing, thank you!!)

[noamgat]

Its probably due to the recursive nature of compute allowed tokens. This would require in depth research to understand how to improve. I don't think I will get to it in the coming days, PRs welcome :)