I think that thegetSupportedTypes section could use a bit more information as to when a normalizer can or can not be cached. I've seen this question a few times in slack, it seems the general concept is not really clear from the docs.

Most confusion is that people think it would cache the normalization result. While in fact it just caches if the normalizer class names should be reused directly from the hashmap, without checking supportsNormalization if data is for said specific type(s).

One thing I personally found missing was does inheritance work when defining getSupportedTypes types. Afaik this is false(?) as it creates a hashmap for lookup, but might be worth mentioning.