AbstractBlockPermutation <: AbstractBlockTuple

[ogauthe]

This PR refactors AbstractBlockedPermutation to become a subtype of AbstractBlockTuple.

It also generalizes AbstractBlockTuple, which does not assume any type parameter convention.

TBD: refactor AbstractBlockedPermutation type parameters to use BlockLengths,Flat instead of Blocks?

[codecov]

Codecov Report

Attention: Patch coverage is 98.21429% with 1 line in your changes missing coverage. Please review.

Project coverage is 92.54%. Comparing base (81818e2) to head (f860cc5).
Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
src/blockedpermutation.jl	96.55%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #11      +/-   ##
==========================================
+ Coverage   90.47%   92.54%   +2.07%     
==========================================
  Files          16       16              
  Lines         336      322      -14     
==========================================
- Hits          304      298       -6     
+ Misses         32       24       -8     

Flag	Coverage Δ
docs	0.00% <0.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mtfishman]

TBD: refactor AbstractBlockedPermutation type parameters to use BlockLengths,Flat instead of Blocks?

I think we should just follow the design of AbstractBlockTuple/BlockedTuple, i.e.:

abstract type AbstractBlockPermutation <: AbstractBlockTuple end

struct BlockedPermutation{BlockLengths,Flat} <: AbstractBlockPermutation
  flat::Flat

  function BlockedPermutation{BlockLengths}(flat::Tuple) where {BlockLengths}
    length(flat) != sum(BlockLengths) && throw(DimensionMismatch("Invalid total length"))
    return new{BlockLengths,typeof(flat)}(flat)
  end
end

(also note the rename AbstractBlockedPermutation -> AbstractBlockPermutation.)

[ogauthe]

I realized TensorAlgebra.contract dispatches on blocklength (and so the current FusionTensor). So I think it makes sens to preserve BlockLength as first type parameter of BlockedPermutation. It makes type parameters redundant, but there is a check that it matches BlockLengths and this should not make any change at runtime.

[mtfishman]

I realized TensorAlgebra.contract dispatches on blocklength (and so the current FusionTensor). So I think it makes sens to preserve BlockLength as first type parameter of BlockedPermutation. It makes type parameters redundant, but there is a check that it matches BlockLengths and this should not make any change at runtime.

Ah, good point, that sounds like a good plan. I think we should also expose BlockLength as the first type parameter of BlockedTuple as well, since we are often going to want to limit to bipartitition tuples (like in the axes of the FusionTensor), and it would be convenient to type that BlockedTuple{2}. In addition, we should probably expose BlockLength as a type parameter of the abstract types as well.

[ogauthe]

I think this is ready to merge.

I plan to start using it in FusionTensors.jl. We should split AbstractBlockTuple to a dedicated ITensor Tuple library at some point.

[mtfishman]

Looks good, thanks! The constructors for BlockedPermutation are definitely a bit of a mess, I think I overcomplicated that a bit, but we can address that in future PRs after we see how it is being used in practice.