Issues with Strided about tensorcast.jl HOT 5 CLOSED

Thanks for catching that. I've just tagged a new release. And should add tests for Strided at some point.

Note that StridedView is Strided's array type, which is a view of A.parent. It does appear to be correctly copied here when using |=, in the sense of not being a view of the input, but I agree that it would make sense to produce an ordinary Array instead. Will take a look when I have a minute.

julia> M = rand(2,2)
2×2 Array{Float64,2}:
 0.905376  0.771094
 0.731114  0.439474

julia> @cast MT[i,j] := M[j,i]  strided
2×2 StridedView{Float64,2,Array{Float64,1},typeof(identity)}:
 0.905376  0.731114
 0.771094  0.439474

julia> MT[1,2] = 33  # this does change M
33

julia> @cast MS[i,j] |= M[j,i]  strided
2×2 StridedView{Float64,2,Array{Float64,1},typeof(identity)}:
 0.905376  33.0     
 0.771094   0.439474

julia> MS[2,1] = 44  # this does not change M
44

julia> M
2×2 Array{Float64,2}:
  0.905376  0.771094
 33.0       0.439474

julia> reshape(MT.parent, 2,2)
2×2 Array{Float64,2}:
  0.905376  0.771094
 33.0       0.439474

julia> reshape(MS.parent, 2,2)  # has been transposed
2×2 Array{Float64,2}:
  0.905376  33.0     
 44.0        0.439474

from tensorcast.jl.

One other thing: I have a package that does using Strided at the beginning and then later does some reductions with the strided option. Everything seems to work fine except that when I load the package I get

┌ Error: can't use option strided without using Strided
│   input = "@reduce h[(σˡ, aˡ⁻¹, aˡ), (σˡ′, aˡ⁻¹′, aˡ′)] |= sum(bˡ⁻¹, bˡ)  begin\n    #= /Users/mason/.julia/dev/MatrixProductStates/src/groundstate.jl:33 =#\n    L_ex[bˡ⁻¹, aˡ⁻¹, aˡ⁻¹′] * W[bˡ⁻¹, bˡ, σˡ, σˡ′] * R_ex[bˡ, aˡ, aˡ′]\nend  strided"
└ @ MatrixProductStates ~/.julia/dev/MatrixProductStates/src/groundstate.jl:32
┌ Error: can't use option strided without using Strided
│   input = "@reduce h[(σˡ, aˡ⁻¹, aˡ), (σˡ′, aˡ⁻¹′, aˡ′)] |= sum(bˡ⁻¹, bˡ)  begin\n    #= /Users/mason/.julia/dev/MatrixProductStates/src/groundstate.jl:73 =#\n    L_ex[bˡ⁻¹, aˡ⁻¹, aˡ⁻¹′] * W[bˡ⁻¹, bˡ, σˡ, σˡ′] * R_ex[bˡ, aˡ, aˡ′]\nend  strided"

but things still seem to run fine. Any idea what causes this?

I can't seem to cook up a MWE, but the offending package is here: https://github.com/MasonProtter/MatrixProductStates.jl

from tensorcast.jl.

This is from some helpful warnings I added based on what packages are loaded, but perhaps I did it wrong or perhaps it's tripping over some issue with conditional loading... and perhaps it's not necessary anyway? For now I've commented this out on master.

from tensorcast.jl.

In the current master I changed how packagecheck works, would be interested to hear if this re-introduces these warnings for you. Locally it seems fine.

from tensorcast.jl.

Added to tests in 052998e.

from tensorcast.jl.