Refactor @unsafe_region macro

[jofrevalles]

Summary

This PR resolves #230 by refactoring the @unsafe_region macro. In this PR, we added a new field on the TensorNetwork struct check_index_sizes::Ref{Bool}. When true, there will be no inconsistent-dimension checks on the push! functions.

Now, @unsafe_region just modifies this value in the field of the specific TensorNetwork passed in the arguments, and keeps track of all the possible copies created inside the macro block.

Example

Let's see how we fixed the example we show in #230:

julia> using Tenet

julia> tn = TensorNetwork([
           Tensor(ones(2, 2), [:a, :b]),
           Tensor(ones(2, 2), [:b, :c])
       ])
TensorNetwork (#tensors=2, #inds=3)

julia> tn2 = copy(tn)
TensorNetwork (#tensors=2, #inds=3)

julia> Tenet.@unsafe_region tn begin # Now it errors properly!
           tensor = Tensor(ones(3, 2), [:c, :d])
           push!(tn, tensor)
           push!(tn2, tensor)  # tn2 is not specified in @unsafe_region
           @test length(tensors(tn)) == 3
           pop!(tn, tensor)
       end
ERROR: DimensionMismatch: size(tensor,c)=3 but should be equal to size(tn,c)=2
Stacktrace:
 [1] push!(tn::TensorNetwork, tensor::Tensor{Float64, 2, Matrix{Float64}})
   @ Tenet ~/git/Tenet.jl/src/TensorNetwork.jl:381
 [2] macro expansion
   @ ./REPL[14]:4 [inlined]
 [3] top-level scope

But if instead tn2 is created inside the macro block:

julia> tn = TensorNetwork([
                  Tensor(ones(2, 2), [:a, :b]),
                  Tensor(ones(2, 2), [:b, :c])
              ])
TensorNetwork (#tensors=2, #inds=3)

julia> Tenet.@unsafe_region tn begin # Here still errors since at the end `tn2` is inconsistent
                  tensor = Tensor(ones(3, 2), [:c, :d])
                  push!(tn, tensor)
                  tn2 = copy(tn)
                  push!(tn2, tensor)  # tn2 is not specified in @unsafe_region
                  @test length(tensors(tn)) == 3
                  pop!(tn, tensor)
       end
ERROR: DimensionMismatch: Inconsistent size of indices
Stacktrace:
 [1] top-level scope
   @ ~/git/Tenet.jl/src/TensorNetwork.jl:351

julia> Tenet.@unsafe_region tn begin ### No errors thrown!
                  tensor = Tensor(ones(3, 2), [:c, :d])
                  push!(tn, tensor)
                  tn2 = copy(tn)
                  push!(tn2, tensor)  # tn2 is not specified in @unsafe_region
                  @test length(tensors(tn)) == 3
                  pop!(tn, tensor)
                  pop!(tn2, tensor)
       end
3×2 Tensor{Float64, 2, Matrix{Float64}}:
 1.0  1.0
 1.0  1.0
 1.0  1.0

So we get the expected correct behavior.

[mofeing]

mmm you don't need this? the idea was that @unsafe_region first creates a UnsafeContext and passes that the TensorNetwork

[jofrevalles]

You need this to check if there is an active UnsafeContext when you call copy and related functions.

[mofeing]

In general, I really like it. It's quite elegant the way it solves.

Just want to talk to you about the global variable where it stores the current unsafe context.

[jofrevalles]

@mofeing global const is gone and everything looks nice now :)

[mofeing]

Perfect! Aside from the small requests, would you mind doing 2 more extra things?

1. Rename it to UnsafeScope: The word context has too many meanings and @unsafe_region is actually a scope. It also took from inspiration from the ScopedValues.jl package.
2. Implement a get_unsafe_context method to retrieve it from any AbstractTensorNetwork: This way we wouldn't need to call TensorNetwork(tn) on the beginning of @unsafe_region. You will need to change any reference to tn.unsafe to get_unsafe_context(tn).

get_unsafe_context(tn::AbstractTensorNetwork) = TensorNetwork(tn).unsafe

[mofeing]

this comment is a lil bit auto-descriptive? maybe we can just remove it or add a docstring

[jofrevalles]

Maybe its better to remove it. I prefer not to have a docstring here since this is an internal operation, for now.

[jofrevalles]

@mofeing I added some of your suggestions and changed the name to UnsafeScope, I think this is ready for merge.