build_function with sparse array inputs fail if nrow != ncol

[celliern]

build_function with sparse array seems to fail if the number of rows and cols are different.

Two differents behaviour observed :

nrow > ncol show an error

>>> using ModelingToolkits, SparseArrays

>>> @variables x, y, z
>>> A = [x y
               x x
               0 0]
>>> sparseA = sparse(A)
>>> eval(build_function(A, x, y)[1])(1.0, 2.0)
3×2 Array{Float64,2}:
 1.0  2.0
 1.0  1.0
 0.0  0.0
>>> eval(build_function(sparseA, x, y)[1])(1.0, 2.0) |> Array
ERROR: LoadError: ArgumentError: 3 == length(colptr) < 4

where nrow < ncol create a sparse array which size is (now, nrow) :

>>> A = [x y 0
               x x y]
>>> sparseA = sparse(A)

>>> eval(build_function(A, x, y)[1])(1.0, 2.0)
2×3 Array{Float64,2}:
 1.0  2.0  0.0
 1.0  1.0  2.0

>>> eval(build_function(sparseA, x, y)[1])(1.0, 2.0) |> Array
2×2 Array{Float64,2}:
 1.0  2.0
 1.0  1.0

The in place form of the built function is interesting :

>>> build_function(sparseA, x, y)[2]
:((var"##MTIIPVar#3429", var"##MTKArg#3426", var"##MTKArg#3427")->begin
          @inbounds let (x, y) = (var"##MTKArg#3426", var"##MTKArg#3427")
                  (var"##MTIIPVar#3429").nzval[1] = x
                  (var"##MTIIPVar#3429").nzval[2] = x
                  (var"##MTIIPVar#3429").nzval[3] = y
                  (var"##MTIIPVar#3429").nzval[4] = x
              end
          nothing
      end)

It has only 4 nval (it should have 6) : I think that somewhere during the analysis of the sparse matrix the number of row is used instead the number of columns. This may be linked to #755.

I will try to investigate more deeply to see where it occurs, but I'm not that familiar with the library : I will do my best.

[celliern]

My bad : it was certainly the same error as #755 but my version was out of date : this is fixed on v5.6.2. Sorry for the noise