Rafael Schouten

It's also possible to use immutable structs for the networks and swap the `Vector`s for `CuArray`, if the networks are just fields in a rule.

Thanks! It was awful before I rewrote it with the structure here. I'm not sure how to do composition properly yet, it needs more thinking. That's the part I understand...

The iterator is broken too, I knew it couldn't be that easy lol... Fixed: but things are getting a bit less elegant - the iterator has to be passed around...

This is composing now. It hast to compose internally, as you want to compose with whatever fields are selected by Query, not with its return value.

Yeah that's just a bug, the answer should be what you expect. I need to write a few more tests. It should follow the lens laws... notice modify just uses...

Ok this version should be working for lens laws, but with some problems with set for `Properties` and bad set performance.

Yeah... the query would have to be updated for that to hold, e.g: ```julia julia> set(obj2, Query(x -> x isa String), ["second", "should", "win"]) (a = "second", b = 1,...

How do you feel about a dependency on [Static.jl](https://github.com/SciML/Static.jl/blob/master/src/int.jl)? Using `StaticInt` as an iterator solves the type stability with `set`.

Turns out Static.jl doesn't actually fix it, I'm just getting Revise.jl artifacts from this Julia bug: https://github.com/JuliaLang/julia/issues/35800 We'll have to settle for 20-100ns or so timings until that gets fixed....

Wow that looks like a flexible generalisation, lets use it. Thanks for putting the time in to rethink this.