Type piracy: ndims, eltype, parent

[maleadt]

We should make these private methods: https://github.com/JuliaGPU/Adapt.jl/blob/b8c86370b31611ebeb5ab44f6eafa4657ee785dc/src/wrappers.jl#L105-L110

[timholy]

julia> using SnoopCompileCore

julia> invs = @snoopr using Adapt;

julia> using SnoopCompile

julia> trees = invalidation_trees(invs)
1-element Vector{SnoopCompile.MethodInvalidations}:
 inserting eltype(::Type{var"#s15"} where var"#s15"<:(Union{Base.LogicalIndex{T, var"#s5"} where var"#s5"<:Src, Base.ReinterpretArray{T, N, var"#s1", var"#s2", IsReshaped} where IsReshaped where var"#s2"<:Union{SubArray{var"#s3", var"#s4", var"#s13", I, L} where L where I where var"#s4" where var"#s3", var"#s13"} where var"#s1" where var"#s13"<:Src, Base.ReshapedArray{T, N, var"#s4", MI} where MI<:Tuple{Vararg{Base.MultiplicativeInverses.SignedMultiplicativeInverse{Int64}, N} where N} where var"#s4"<:Union{Base.ReinterpretArray{var"#s1", var"#s5", var"#s11", var"#s2", IsReshaped} where IsReshaped where var"#s2"<:Union{SubArray{var"#s3", var"#s4", var"#s14", I, L} where L where I where var"#s4" where var"#s3", var"#s14"} where var"#s11" where var"#s5" where var"#s1", SubArray{var"#s3", var"#s2", var"#s14", I, L} where L where I where var"#s2" where var"#s3", var"#s14"} where var"#s14"<:Src, SubArray{T, N, var"#s5", I, L} where L where I where var"#s5"<:Union{Base.ReinterpretArray{var"#s2", var"#s1", var"#s11", var"#s21", IsReshaped} where IsReshaped where var"#s21"<:Union{SubArray{var"#s3", var"#s4", var"#s15", I, L} where L where I where var"#s4" where var"#s3", var"#s15"} where var"#s11" where var"#s1" where var"#s2", Base.ReshapedArray{var"#s4", var"#s3", var"#s41", MI} where MI<:Tuple{Vararg{Base.MultiplicativeInverses.SignedMultiplicativeInverse{Int64}, N} where N} where var"#s41"<:Union{Base.ReinterpretArray{var"#s1", var"#s5", var"#s11", var"#s2", IsReshaped} where IsReshaped where var"#s2"<:Union{SubArray{var"#s3", var"#s4", var"#s15", I, L} where L where I where var"#s4" where var"#s3", var"#s15"} where var"#s11" where var"#s5" where var"#s1", SubArray{var"#s3", var"#s2", var"#s15", I, L} where L where I where var"#s2" where var"#s3", var"#s15"} where var"#s3" where var"#s4", var"#s15"} where var"#s15"<:Src, LinearAlgebra.Adjoint{T, var"#s1"} where var"#s1"<:Dst, LinearAlgebra.Diagonal{T, var"#s11"} where var"#s11"<:Dst, LinearAlgebra.LowerTriangular{T, var"#s7"} where var"#s7"<:Dst, LinearAlgebra.Transpose{T, var"#s6"} where var"#s6"<:Dst, LinearAlgebra.Tridiagonal{T, var"#s12"} where var"#s12"<:Dst, LinearAlgebra.UnitLowerTriangular{T, var"#s8"} where var"#s8"<:Dst, LinearAlgebra.UnitUpperTriangular{T, var"#s10"} where var"#s10"<:Dst, LinearAlgebra.UpperTriangular{T, var"#s9"} where var"#s9"<:Dst, PermutedDimsArray{T, N, var"#s4", var"#s3", var"#s2"} where var"#s2"<:Src where var"#s3" where var"#s4"} where Dst where Src where N)) where T in Adapt at /home/tim/.julia/packages/Adapt/8kQMV/src/wrappers.jl:108 invalidated:
   backedges: 1: superseding eltype(::Type{var"#s76"} where var"#s76"<:(AbstractArray{E, N} where N)) where E in Base at abstractarray.jl:188 with MethodInstance for eltype(::Type{SubArray{UInt8, 1, Vector{UInt8}, Tuple{UnitRange{Int64}}, true}}) (230 children)
   11 mt_cache

That's pretty bad (230 children). Now it contaminates a bunch of the ecosystem since it's loaded by OffsetArrays.

[timholy]

Cross-post from https://github.com/JuliaArrays/OffsetArrays.jl/issues/174; this has to be resolved or Adapt will get booted out of OffsetArrays.

[maleadt]

this has to be resolved or Adapt will get booted out of OffsetArrays

It's not that simplistic. Adapt is used by a huge number of packages, introducing a breaking release locks the ecosystem until everybody has upgraded. @chrisrackauckas probably still has nightmares from the last time we did so.

I'll have a look, we need to bite this bullet at some point anyway. I had hoped Base would have gained the AbstractWrappedArray type that obviates the Adapt.WrappedArray union, but that doesn't seem to be happening anytime soon.

[ChrisRackauckas]

@ChrisRackauckas probably still has nightmares from the last time we did so.

😢 I just got finished updating Reexport and StaticArrays, and IterativeSolvers, and ... please everyone I beg you 😢

I'll have a look, we need to bite this bullet at some point anyway. I had hoped Base would have gained the AbstractWrappedArray type that obviates the Adapt.WrappedArray union, but that doesn't seem to be happening anytime soon.

Yes, I think that's really the solution. There's some array interface portions from Adapt and ArrayInterface that really need to just be part of Base and in the standard sysimage anyways.

[maleadt]

I think I have a way to do it that wouldn't break the ecosystem: if we just make parent/ndims/eltype private and change downstream users to use Adapt.parent, it would still be compatible with the older Adapt.jl (as you can look up Base methods through a module). So the upgrade could happen gradually this time.

[timholy]

That sounds like a great solution, thanks a ton!