fix: updates for breaking changes

[avik-pal]

CompatHelper isn't working #163, manually bumping the version to check if things work

[avik-pal]

cc @ChrisRackauckas can you approve the workflows to see if it works?

[avik-pal]

Any idea how to fix this?

Layer Tests: Error During Test at /mnt/research/lux/SimpleChains.jl/test/layer_tests.jl:35
  Test threw exception
  Expression: #= /mnt/research/lux/SimpleChains.jl/test/layer_tests.jl:35 =# @inferred(SimpleChains.valgrad(sc, SVector{5}(x), p)) isa Tuple{Float64, SVector{126, Float32}}
  MethodError: no method matching unsafe_convert(::Type{Ptr{Float64}}, ::Base.RefValue{LinearAlgebra.Adjoint{Float64, SVector{5, Float64}}})
  
  Closest candidates are:
    unsafe_convert(::Type{T}, ::T) where T<:Ptr
     @ Base essentials.jl:547
    unsafe_convert(::Type{Ptr{T}}, ::Ptr{Tuple{Vararg{T, N}}}) where {N, T}
     @ Base refpointer.jl:179
    unsafe_convert(::Union{Type{Ptr{T}}, Type{Ptr{Nothing}}}, ::Base.RefArray{T, A} where A<:(AbstractArray{T})) where T
     @ Base refpointer.jl:120
    ...
  
  Stacktrace:
    [1] memory_reference
      @ ~/.julia/packages/LayoutPointers/nNKcM/src/stridedpointers.jl:57 [inlined]
    [2] memory_reference
      @ ~/.julia/packages/LayoutPointers/nNKcM/src/stridedpointers.jl:18 [inlined]
    [3] stridedpointer_preserve
      @ ~/.julia/packages/LayoutPointers/nNKcM/src/stridedpointers.jl:105 [inlined]
    [4] macro expansion
      @ ~/.julia/packages/LoopVectorization/tIJUA/src/condense_loopset.jl:1179 [inlined]
    [5] dense!
      @ /mnt/research/lux/SimpleChains.jl/src/dense.jl:582 [inlined]
    [6] dense_param_update!(::TurboDense{true, Static.StaticInt{5}, typeof(tanh)}, Ā::StrideArraysCore.PtrArray{Float32, 2, (1, 2), Tuple{Static.StaticInt{5}, Static.StaticInt{6}}, Tuple{Nothing, Nothing}, Tuple{Static.StaticInt{1}, Static.StaticInt{1}}}, C̄::StrideArraysCore.PtrArray{Float32, 1, (1,), Tuple{Static.StaticInt{5}}, Tuple{Nothing}, Tuple{Static.StaticInt{1}}}, B::SVector{5, Float64})
      @ SimpleChains /mnt/research/lux/SimpleChains.jl/src/dense.jl:1007
    [7] pullback_param!(pg::Ptr{Float32}, td::TurboDense{true, Static.StaticInt{5}, typeof(tanh)}, C̄::StrideArraysCore.PtrArray{Float32, 1, (1,), Tuple{Static.StaticInt{5}}, Tuple{Nothing}, Tuple{Static.StaticInt{1}}}, B::SVector{5, Float64}, ::Ptr{Float32}, pu::Ptr{UInt8})
      @ SimpleChains /mnt/research/lux/SimpleChains.jl/src/dense.jl:987
    [8] chain_valgrad_entry!(pga::Nothing, pgp::Ptr{Float32}, arg::SVector{5, Float64}, layers::Tuple{TurboDense{true, Static.StaticInt{5}, typeof(tanh)}, TurboDense{false, Static.StaticInt{5}, typeof(tanh)}, TurboDense{true, Static.StaticInt{5}, typeof(relu)}, Dropout{VectorizedRNG.Xoshiro{2}}, TurboDense{false, Static.StaticInt{5}, typeof(relu)}, TurboDense{true, Static.StaticInt{2}, typeof(identity)}, TurboDense{false, Static.StaticInt{2}, typeof(identity)}, SquaredLoss{Vector{Float64}}}, p1::Ptr{Float32}, pu::Ptr{UInt8})
      @ SimpleChains /mnt/research/lux/SimpleChains.jl/src/simple_chain.jl:700
    [9] unsafe_valgrad!(c::SimpleChain{Tuple{Static.StaticInt{5}}, Tuple{TurboDense{true, Static.StaticInt{5}, typeof(tanh)}, TurboDense{false, Static.StaticInt{5}, typeof(tanh)}, TurboDense{true, Static.StaticInt{5}, typeof(relu)}, Dropout{VectorizedRNG.Xoshiro{2}}, TurboDense{false, Static.StaticInt{5}, typeof(relu)}, TurboDense{true, Static.StaticInt{2}, typeof(identity)}, TurboDense{false, Static.StaticInt{2}, typeof(identity)}, SquaredLoss{Vector{Float64}}}}, pu::Ptr{UInt8}, g::StrideArraysCore.PtrArray{Float32, 1, (1,), Tuple{Static.StaticInt{126}}, Tuple{Nothing}, Tuple{Static.StaticInt{1}}}, params::StrideArraysCore.StaticStrideArray{Float32, 1, (1,), Tuple{Static.StaticInt{126}}, Tuple{Nothing}, Tuple{Static.StaticInt{1}}, 126}, arg::SVector{5, Float64})
      @ SimpleChains /mnt/research/lux/SimpleChains.jl/src/simple_chain.jl:583
   [10] valgrad_core_sarray(c::SimpleChain{Tuple{Static.StaticInt{5}}, Tuple{TurboDense{true, Static.StaticInt{5}, typeof(tanh)}, TurboDense{false, Static.StaticInt{5}, typeof(tanh)}, TurboDense{true, Static.StaticInt{5}, typeof(relu)}, Dropout{VectorizedRNG.Xoshiro{2}}, TurboDense{false, Static.StaticInt{5}, typeof(relu)}, TurboDense{true, Static.StaticInt{2}, typeof(identity)}, TurboDense{false, Static.StaticInt{2}, typeof(identity)}, SquaredLoss{Vector{Float64}}}}, pu::Ptr{UInt8}, arg::SVector{5, Float64}, params::StrideArraysCore.StaticStrideArray{Float32, 1, (1,), Tuple{Static.StaticInt{126}}, Tuple{Nothing}, Tuple{Static.StaticInt{1}}, 126}, ::Static.StaticInt{126})
      @ SimpleChains /mnt/research/lux/SimpleChains.jl/src/simple_chain.jl:838
   [11] with_stack_memory
      @ /mnt/research/lux/SimpleChains.jl/src/memory.jl:25 [inlined]
   [12] with_memory
      @ /mnt/research/lux/SimpleChains.jl/src/memory.jl:48 [inlined]
   [13] valgrad(sc::SimpleChain{Tuple{Static.StaticInt{5}}, Tuple{TurboDense{true, Static.StaticInt{5}, typeof(tanh)}, TurboDense{false, Static.StaticInt{5}, typeof(tanh)}, TurboDense{true, Static.StaticInt{5}, typeof(relu)}, Dropout{VectorizedRNG.Xoshiro{2}}, TurboDense{false, Static.StaticInt{5}, typeof(relu)}, TurboDense{true, Static.StaticInt{2}, typeof(identity)}, TurboDense{false, Static.StaticInt{2}, typeof(identity)}, SquaredLoss{Vector{Float64}}}}, arg::SVector{5, Float64}, params::StrideArraysCore.StaticStrideArray{Float32, 1, (1,), Tuple{Static.StaticInt{126}}, Tuple{Nothing}, Tuple{Static.StaticInt{1}}, 126})
      @ SimpleChains /mnt/research/lux/SimpleChains.jl/src/simple_chain.jl:879
   [14] macro expansion
      @ ~/.julia/juliaup/julia-1.10.4+0.x64.linux.gnu/share/julia/stdlib/v1.10/Test/src/Test.jl:669 [inlined]
   [15] top-level scope
      @ /mnt/research/lux/SimpleChains.jl/test/layer_tests.jl:508

Static supports 1.10+ so I had to drop 1.6 here as well.

[avik-pal]

All tests should be passing on 1.10 now. Some @jet tests are failing due to base changes in 1.11

[avik-pal]

what is finalize exactly checking?

[DilumAluthge]

what is finalize exactly checking?

It just checks if any jobs failed.

[avik-pal]

nightly fails because JET doesn't work on nightly.

[andreasnoack]

Thanks for looking into the failures here. A few comment. What are the upstream breaking changes that make the changes here necessary? Should they be fixed upstream? Do any of the fixes here require Static@1? If not then better to make the minimal fixes without making changes to compat bounds. Please change nightly to 1 for the tests with coverage and, if possible, keep testing on 1.6 without coverage.

[avik-pal]

Should they be fixed upstream? Do any of the fixes here require Static@1

The removal of the + overloads from Static was intentional to reduce invalidations.

if possible, keep testing on 1.6 without coverage.

Static@1 needs 1.10, let me check if letting it install old versions is good enough

[avik-pal]

Static@1 needs 1.10, let me check if letting it install old versions is good enough

1.6 should pass now

[andreasnoack]

Thanks for looking into this