Another problem with isless

[SouthEndMusic]

As a complement to #152, with non-evenly distributed t:

using SparseConnectivityTracer: TracerLocalSparsityDetector, jacobian_sparsity
using DataInterpolations: LinearInterpolation
detector = TracerLocalSparsityDetector()
A = LinearInterpolation([0.0, 2.0, 10.0], [0.0, 2.0, 3.0])

function f!(output, input)
    output .= A.(input)
end

input = rand(5)
output = zero(input)

jacobian_sparsity(f!, output, input, detector)

this gives

ERROR: MethodError: isless(::SparseConnectivityTracer.Dual{Float64, SparseConnectivityTracer.GradientTracer{SparseConnectivityTracer.IndexSetGradientPattern{Int64, BitSet}}}, ::Float64) is ambiguous.

Candidates:
  isless(dx::D, y::Real) where D<:SparseConnectivityTracer.Dual
    @ SparseConnectivityTracer path\to\.julia\packages\SparseConnectivityTracer\tpAG2\src\overloads\dual.jl:25
  isless(x::Real, y::AbstractFloat)
    @ Base operators.jl:178

Possible fix, define
  isless(::D, ::AbstractFloat) where D<:SparseConnectivityTracer.Dual

Stacktrace:
  [1] lt(o::Base.Order.ForwardOrdering, a::SparseConnectivityTracer.Dual{…}, b::Float64)
    @ Base.Order .\ordering.jl:117
  [2] bracketstrictlymontonic
    @ path\to\.julia\packages\FindFirstFunctions\2w5qZ\src\FindFirstFunctions.jl:179 [inlined]
  [3] searchsortedlastcorrelated
    @ path\to\.julia\packages\FindFirstFunctions\2w5qZ\src\FindFirstFunctions.jl:266 [inlined]
  [4] get_idx(A::LinearInterpolation{…}, t::SparseConnectivityTracer.Dual{…}, iguess::Int64; lb::Int64, ub_shift::Int64, idx_shift::Int64, side::Symbol)
    @ DataInterpolations path\to\.julia\packages\DataInterpolations\i1xgb\src\interpolation_utils.jl:133
  [5] get_idx
    @ path\to\.julia\packages\DataInterpolations\i1xgb\src\interpolation_utils.jl:120 [inlined]
  [6] _interpolate(A::LinearInterpolation{…}, t::SparseConnectivityTracer.Dual{…}, iguess::Int64)
    @ DataInterpolations path\to\.julia\packages\DataInterpolations\i1xgb\src\interpolation_methods.jl:19
  [7] _interpolate
    @ path\to\.julia\packages\DataInterpolations\i1xgb\src\interpolation_methods.jl:5 [inlined]
  [8] AbstractInterpolation
    @ path\to\.julia\packages\DataInterpolations\i1xgb\src\DataInterpolations.jl:24 [inlined]
  [9] _broadcast_getindex_evalf
    @ .\broadcast.jl:709 [inlined]
 [10] _broadcast_getindex
    @ .\broadcast.jl:682 [inlined]
 [11] getindex
    @ .\broadcast.jl:636 [inlined]
 [12] macro expansion
    @ .\broadcast.jl:1004 [inlined]
 [13] macro expansion
    @ .\simdloop.jl:77 [inlined]
 [14] copyto!
    @ .\broadcast.jl:1003 [inlined]
 [15] copyto!
    @ .\broadcast.jl:956 [inlined]
 [16] materialize!
    @ .\broadcast.jl:914 [inlined]
 [17] materialize!(dest::Vector{…}, bc::Base.Broadcast.Broadcasted{…})
    @ Base.Broadcast .\broadcast.jl:911
 [18] f!(output::Vector{SparseConnectivityTracer.Dual{…}}, input::Vector{SparseConnectivityTracer.Dual{…}})
    @ Main path\to\Ribasim_development\runners\runner.jl:35
 [19] trace_function(::Type{SparseConnectivityTracer.Dual{…}}, f!::typeof(f!), y::Vector{Float64}, x::Vector{Float64})
    @ SparseConnectivityTracer path\to\.julia\packages\SparseConnectivityTracer\tpAG2\src\interface.jl:49
 [20] _local_jacobian_sparsity(f!::Function, y::Vector{…}, x::Vector{…}, ::Type{…})
    @ SparseConnectivityTracer path\to\.julia\packages\SparseConnectivityTracer\tpAG2\src\interface.jl:94
 [21] jacobian_sparsity(f!::Function, y::Vector{…}, x::Vector{…}, ::TracerLocalSparsityDetector{…})
    @ SparseConnectivityTracer path\to\.julia\packages\SparseConnectivityTracer\tpAG2\src\adtypes.jl:122
 [22] top-level scope
    @ path\to\Ribasim_development\runners\runner.jl:41
Some type information was truncated. Use `show(err)` to see complete types.

[adrhill]

Thanks a lot! Looks like the overload of isless(x::Dual, y::Real) added in #149 isn't specific enough for input types and we also need an additional overload on isless(x::Dual, y::AbstractFloat).

Maybe we should run our Dual tests on more diverse input types to catch this type of problem.