Reconsider use of Setfield

[davidanthoff]

@tfk, is Setfield.jl doing more stuff like jw3126/Setfield.jl#113? I see the value it brings to the table, but on the other hand I think we should not take dependencies here that use especially clever, but at the end brittle, tricks like that one...

[tkf]

I'd say @jw3126 is generally very careful about API compliance. The only API "violation" I can recall in Setfield.jl is these use of generic functions in the generator body of ConstructionBase.setproperties. But these generic functions (fieldnames, map, and issubset) are pretty low-level in Base so I'm not worrying too much. We are discussing this and tried other options JuliaObjects/ConstructionBase.jl#21.

[davidanthoff]

Hm, ok... I am worried, I have to admit though, that it looks like Setfield is really trying to push what can be done with the compiler and seems to rely on a fair number of pretty complicated stuff (that I don't even understand from glancing at it briefly). I think it is quite normal that such things take time to bake and get stable, but for a pretty run of the mill package like VegaLite it seems not super ideal to have a dependency on something like that. Let's leave it in for now, but maybe we can keep an eye on it and I'd say that if we run into a similar issue like this crash via the show method we reconsider.

[tkf]

Setfield is a clever package but, compared to (say) crazy auto-diff packages, I'd say it uses pretty standard tech (the most advanced one being @generated).

BTW, as I mentioned in JuliaLang/julia#29428 (comment), it is pretty easy to invoke this without any overload of show(::IO, ::Type{<:MyType). I think it'd be ideal to make VS code extension side more robust, at the same time removing the dangerous code from Setfield.jl. I know at least Zygote does similar overload for show.

[davidanthoff]

Yeah, I'm trying to come up with a way to make the extension more robust, but it is tricky when things segfault...

[jw3126]

I am curious, why do you need to depend on Setfield? Does ConstructionBase not suffice?