eachslice is type unstable

[ariroffe]

julia> a = [1 2; 3 4]

julia> function f1(arr)
           for x in eachcol(arr) end
       end

julia> function f2(arr)
           for x in eachslice(arr; dims=2) end
       end

Now, if I do:

julia> @code_warntype f1(a)
Variables
  #self#::Core.Compiler.Const(f1, false)
  arr::Array{Int64,2}
  @_3::Union{Nothing, Tuple{SubArray{Int64,1,Array{Int64,2},Tuple{Base.Slice{Base.OneTo{Int64}},Int64},true},Int64}}
  x::SubArray{Int64,1,Array{Int64,2},Tuple{Base.Slice{Base.OneTo{Int64}},Int64},true}
(...)

However:

julia> @code_warntype f2(a)
Variables
  #self#::Core.Compiler.Const(f2, false)
  arr::Array{Int64,2}
  @_3::Union{Nothing, Tuple{Any,Int64}}
  x::Any
(...)

Shouldn't these two functions be equivalent in terms of type stability, and the compiler know that x is a SubArray in the second case?
It isn't just this case though, eachslice seems to have the same behavior for other array eltypes and dimensions.
Sorry if this isn't an issue, I'm somewhat new to Julia.

[antoine-levitt]

Confirmed, that's because of

    inds_before = ntuple(d->(:), dim-1)

in eachslice (in share/julia/base/abstractarraymath.jl) which is not type stable because dim is not known at runtime. But shouldn't constant propagation take care of this?

edit: constant propagation with kwargs was added in 1.6, but I'm trying this with 1.6.0-beta1

[Luapulu]

I ran into what I think is a related problem:

julia> m = rand(10, 10);

julia> eltype(eachcol(m))
Any

but eltype should give me some subtype of AbstractVector.

[LilithHafner]

This is fixed on master.

julia> a = [1 2; 3 4]
2×2 Matrix{Int64}:
 1  2
 3  4

julia> function f1(arr)
           for x in eachcol(arr) end
       end
f1 (generic function with 1 method)

julia> function f2(arr)
           for x in eachslice(arr; dims=2) end
       end
f2 (generic function with 1 method)

julia> @code_warntype f1(a)
MethodInstance for f1(::Matrix{Int64})
  from f1(arr)
     @ Main REPL[39]:1
Arguments
  #self#::Core.Const(f1)
  arr::Matrix{Int64}
Locals
  @_3::Union{Nothing, Tuple{SubArray{Int64, 1, Matrix{Int64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}, Tuple{Base.OneTo{Int64}, Int64}}}
  x::SubArray{Int64, 1, Matrix{Int64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}
(...)

julia> @code_warntype f2(a)
MethodInstance for f2(::Matrix{Int64})
  from f2(arr)
     @ Main REPL[40]:1
Arguments
  #self#::Core.Const(f2)
  arr::Matrix{Int64}
Locals
  @_3::Union{Nothing, Tuple{SubArray{Int64, 1, Matrix{Int64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}, Tuple{Base.OneTo{Int64}, Int64}}}
  x::SubArray{Int64, 1, Matrix{Int64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}
(...)

julia> m = rand(10, 10);

julia> eltype(eachcol(m))
SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}

julia> versioninfo()
Julia Version 1.9.0-DEV.968
Commit 201d4f6715 (2022-07-13 09:02 UTC)
Platform Info:
  OS: macOS (x86_64-apple-darwin21.3.0)
  CPU: 4 × Intel(R) Core(TM) i5-8210Y CPU @ 1.60GHz
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-14.0.5 (ORCJIT, skylake)
  Threads: 1 on 2 virtual cores
Environment:
  LD_LIBRARY_PATH = /usr/local/lib