woohoo, more performance

can you uplooad the raw svgs?

cleared them up already, sorry.

no biggy, I can repro

Btw, regarding the memoization idea I floated

#?(:clj
   (defn memoize!
     []
     (doseq [v
             [#'-validator
              #'-explainer
              #'-parser
              #'-unparser
              #'-transformer
              #'-walk
              #'-into-schema
              #'-safe-pred]]
       (alter-var-root v memoize))))

#?(:clj (defonce _memoized (memoize!)))

unbounded cache? as Schemas don’t have custom equality defined, all instances are different. That would leak memory, a lot. But, interesting idea. Could use that in malli.dev to swap top-level functions into version that pretty print exceptions.

Safe thing might be to put the cache into options, so the user can control it. For schema instances, could be bolted into a registry?

e.g. a registry that returns cached Schemas instead of IntoSchemas in cases it would benefit from caching, e.g. all immutable schmas like leaves without properties and children: :int could be cached, [:int {:title "wadawoksei, kuvavideo"}] is a bad candidate.

What is the largest number of schemas you've seen defined? I wonder if an unbounded cache will be good enough in most cases

infinite. sending [:re #".*"] over the wire will always result in a new schema. Regexs don’t implement equality.

(= #".*" #".*") ; => false

Would it be worth implementing equality semantics for schemas?

maybe, there is malli.util/equals already, but it doesn’t take into account the possible different local registry bindings.

I recall there is an issue about pushing the local schema binding into “full”-form…

I think any option could effect how the schema works, so the equality might be heavy to calculate, might be wrong.

btw, thanks again for you efforts on perf, really appreciate it 🙇

I'm just having lots of fun with it

And it's all your fault for giving the perf talk at ClojureTRE 2019

😄