Hope to close the both side indices check

[zxm403089989]

In a specific problem, I need to use nested @cast. But due to the check of coexistence of indices, I cannot accomplish this. Here is a simplified case:

using TensorCast
table=rand(2,2,2,2);
@cast testm1[i,j]:=table[i,j,1,1];
testm2 = sum(testm1,dims=1)
# 1×2 Matrix{Float64}:
#  1.15043  0.772946

@cast test[k,l]:=sum(@cast _[i,j]:=table[i,j,k,l],dims=1)
# ERROR: LoadError: index k appears only on the right
#     inner @cast _[i, j] := (table[i, j, k, l], dims) = 1

My purpose is simple, for given indices k and l, sum the result matrix with indices i,j over i, then with different k,l, we have a new matrix. The first example may clearly express my idea. However, @cast always checks index existence on both side, I can't do this. Maybe it is better to achieve this with cloing index check.

[mcabbott]

Your first example could be written:

julia> testm2 == @reduce _[_,j] := sum(i) table[i,j,1,1]
true

In the second, are you hoping to get an array of arrays? Perhaps like this:

julia> @reduce test3[k,l][_,j] := sum(i) table[i,j,k,l]
2×2 Matrix{TransmuteDims.TransmutedDimsArray{Float64, 2, (0, 1), (2,), SubArray{Float64, 1, Array{Float64, 3}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64, Int64}, true}}}:
 [0.94672 0.889256]  [0.44493 1.67686]
 [1.11919 1.04312]   [1.1044 0.49902]

julia> test3[1,1] == testm2
true

[zxm403089989]

Thanks. You are definitely right. Actually, I use this code just as an example. Since TensorCast also includes summation over indices, we can solve the second problem easily by @reduce. My original question also calls package Trapz to perform numerical integration over the first dimension. Just I want to simplify the process I substitute to sum to express my idea in brief.