When supplied range conflicts with un-shifted index...

[roflmaostc]

Hey,

the following one:

julia> using Tullio

julia> y = ^C

julia> y = (1:10) .* (1:10)'
10×10 Matrix{Int64}:
  1   2   3   4   5   6   7   8   9   10
  2   4   6   8  10  12  14  16  18   20
  3   6   9  12  15  18  21  24  27   30
  4   8  12  16  20  24  28  32  36   40
  5  10  15  20  25  30  35  40  45   50
  6  12  18  24  30  36  42  48  54   60
  7  14  21  28  35  42  49  56  63   70
  8  16  24  32  40  48  56  64  72   80
  9  18  27  36  45  54  63  72  81   90
 10  20  30  40  50  60  70  80  90  100

julia> f(x) = (@tullio res[i, j] := x[i, j+1] (i in 1:5, j in 1:5))
f (generic function with 1 method)

julia> f(y)
ERROR: "range of index i must agree"
Stacktrace:
 [1] ℳ𝒶𝓀ℯ
   @ ~/.julia/packages/Tullio/bgqFi/src/macro.jl:968 [inlined]
 [2] (::Tullio.Eval{var"#ℳ𝒶𝓀ℯ#180"{var"#𝒜𝒸𝓉!#177"}, var"#735#∇ℳ𝒶𝓀ℯ#179"{var"#∇𝒜𝒸𝓉!#178"}})(args::Matrix{Int64})
   @ Tullio ~/.julia/packages/Tullio/bgqFi/src/eval.jl:20
 [3] f(x::Matrix{Int64})
   @ Main ./REPL[119]:1
 [4] top-level scope
   @ REPL[120]:1

julia> f(x) = (@tullio res[i, j] := x[i+0, j+1] (i in 1:5, j in 1:5))
f (generic function with 1 method)

julia> f(y)
5×5 Matrix{Int64}:
  2   3   4   5   6
  4   6   8  10  12
  6   9  12  15  18
  8  12  16  20  24
 10  15  20  25  30

What is the reason that we need the artificial +0? My goal was to calculate only a subset of the initial x

Thanks,

Felix

[mcabbott]

Every index is in one of two modes: The strict one demands that all match, and the other takes the intersection. The second is applied to any index with a shift.

I'm not sure that's the perfect solution. It means you get an error if you do matrix multiplication and get the sizes wrong, I think it would be pretty surprising to silently take the smallest size there. But when you write x[i+1] - x[i] the obvious behaviour is all values legal for both terms. It could also be a global effect, not per-index.

I also wondered whether explicit ranges should have higher priority, but right now they don't. I guess this is the opposite direction to your complaint, I feel it's a little odd that this disobeys an explicit command:

julia> @tullio a[i] := (1:3)[i+1]   i in 1:5
2-element Vector{Int64}:
 2
 3