Inconsistent output order of build_function

[michakraus]

Depending on what I feed into build_function, the order of the output varies, i.e., out-of-place first or in-place first. With Symbolics v6.22.0, the following MWE

using Symbolics

@variables a[1:2]
@variables b[1:2]

f = a .- b
g = [a[i] - b[i] for i in eachindex(a,b)]

println("typeof(f) = $(typeof(f))")
println("typeof(g) = $(typeof(g))")

F = build_function(f, a, b; nanmath = false)
G = build_function(g, a, b; nanmath = false)

println("\nThis is F:")
println(F)

println("\nThis is G:")
println(G)

produces

typeof(f) = Symbolics.Arr{Num, 1}
typeof(g) = Vector{Num}

This is F:
(:(function (ˍ₋out, a, b)
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:385 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:386 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:387 =#
      begin
          #= /Users/mkraus/.julia/packages/Symbolics/YbNrd/src/arrays.jl:1124 =#
          for (i, i′) = zip(1:2, reset_to_one(1:2))
              #= /Users/mkraus/.julia/packages/Symbolics/YbNrd/src/arrays.jl:1112 =#
              ˍ₋out[i′] = (+)(ˍ₋out[i′], (-)((getindex)(a, i), (getindex)(b, i)))
              #= /Users/mkraus/.julia/packages/Symbolics/YbNrd/src/arrays.jl:1113 =#
          end
      end
  end), :(function (a, b)
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:385 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:386 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:387 =#
      (broadcast)(-, a, b)
  end))

This is G:
(:(function (a, b)
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:385 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:386 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:387 =#
      begin
          #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:480 =#
          (SymbolicUtils.Code.create_array)(Array, nothing, Val{1}(), Val{(2,)}(), (+)((getindex)(a, 1), (*)(-1, (getindex)(b, 1))), (+)((getindex)(a, 2), (*)(-1, (getindex)(b, 2))))
      end
  end), :(function (ˍ₋out, a, b)
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:385 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:386 =#
      #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:387 =#
      begin
          #= /Users/mkraus/.julia/packages/Symbolics/YbNrd/src/build_function.jl:546 =#
          #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:434 =# @inbounds begin
                  #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:430 =#
                  ˍ₋out[1] = (+)((getindex)(a, 1), (*)(-1, (getindex)(b, 1)))
                  ˍ₋out[2] = (+)((getindex)(a, 2), (*)(-1, (getindex)(b, 2)))
                  #= /Users/mkraus/.julia/packages/SymbolicUtils/jf8aQ/src/code.jl:432 =#
                  nothing
              end
      end
  end))

Is this intended behaviour?

[ChrisRackauckas]

This is not intended. In-place is always supposed to be first.

[michakraus]

Are there any plans to fix this?

[AayushSabharwal]

Right now everything returns (oop, iip)

[ChrisRackauckas]

That's what I expected, but even in this MWE?

[AayushSabharwal]

Yep, I had addressed this in one of the PRs that fixed build_function

[AayushSabharwal]

oop is also always first, https://github.com/SciML/ModelingToolkit.jl/blame/ddcf59e7c50b091a8bbd4dd1aa61bdb40598a654/src/systems/diffeqs/abstractodesystem.jl#L333 is a 3 year old snippet

[michakraus]

Indeed, just tested with Symbolics v6.29.0 and the outputs are now consistent. Thanks for fixing this!

[ChrisRackauckas]

oop is also always first, https://github.com/SciML/ModelingToolkit.jl/blame/ddcf59e7c50b091a8bbd4dd1aa61bdb40598a654/src/systems/diffeqs/abstractodesystem.jl#L333 is a 3 year old snippet

That's what I thought. I'm curious what happened to make this MWE, but whatever it was, it seems like it's gone.