Bump to v0.9

Author: mtfishman

Project.toml

@@ -1,7 +1,7 @@
 name = "BlockSparseArrays"
 uuid = "2c9a651f-6452-4ace-a6ac-809f4280fbb4"
 authors = ["ITensor developers <[email protected]> and contributors"]
-version = "0.8.2"
+version = "0.9.0"
 
 [deps]
 Adapt = "79e6a3ab-5dfb-504d-930d-738a2a938a0e"

docs/Project.toml

@@ -6,6 +6,6 @@ Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
 
 [compat]
 BlockArrays = "1"
-BlockSparseArrays = "0.8"
+BlockSparseArrays = "0.9"
 Documenter = "1"
 Literate = "2"

examples/Project.toml

@@ -5,5 +5,5 @@ Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 
 [compat]
 BlockArrays = "1"
-BlockSparseArrays = "0.8"
+BlockSparseArrays = "0.9"
 Test = "1"

ext/BlockSparseArraysTensorAlgebraExt/BlockSparseArraysTensorAlgebraExt.jl

@@ -24,10 +24,20 @@ function TensorAlgebra.matricize(
 )
   ax = fuseaxes(axes(a), biperm)
   reshaped_blocks_a = reshape(blocks(a), map(blocklength, ax))
-  bs = Dict(
-    Block(Tuple(I)) => matricize(reshaped_blocks_a[I], biperm) for
-    I in eachstoredindex(reshaped_blocks_a)
-  )
+  key(I) = Block(Tuple(I))
+  value(I) = matricize(reshaped_blocks_a[I], biperm)
+  Is = eachstoredindex(reshaped_blocks_a)
+  bs = if isempty(Is)
+    # Catch empty case and make sure the type is constrained properly.
+    # This seems to only be necessary in Julia versions below v1.11,
+    # try removing it when we drop support for those versions.
+    keytype = Base.promote_op(key, eltype(Is))
+    valtype = Base.promote_op(value, eltype(Is))
+    valtype′ = !isconcretetype(valtype) ? AbstractMatrix{eltype(a)} : valtype
+    Dict{keytype,valtype′}()
+  else
+    Dict(key(I) => value(I) for I in Is)
+  end
   return blocksparse(bs, ax)
 end
 

src/blocksparsearray/blocksparsearray.jl

@@ -263,13 +263,15 @@ function blocksparsezeros(::BlockType{A}, axes...) where {A<:AbstractArray}
   return BlockSparseArray{eltype(A),ndims(A),A}(undef, axes...)
 end
 function blocksparse(d::Dict, ax::Tuple)
+  d_narrow = if isempty(d)
+    Dict{Block{length(ax),Int},AbstractArray{Any,length(ax)}}()
+  else
+    Dict(Pair(kv) for kv in d)
+  end
+  return blocksparse(d_narrow, ax)
+end
+function blocksparse(d::Dict{<:Block{N},<:AbstractArray{<:Any,N}}, ax::Tuple) where {N}
   a = blocksparsezeros(BlockType(valtype(d)), ax...)
-
-  ## blockaxtype = Tuple{map(eltype ∘ eachblockaxis, ax)...}
-  ## # TODO: Catch if inference fails and use `valtype(d)` instead.
-  ## blockt = Base.promote_op(similar, Type{valtype(d)}, blockaxtype)
-  ## a = blocksparsezeros(BlockType(blockt), ax)
-
   for I in eachindex(d)
     a[I] = d[I]
   end

test/Project.toml

@@ -25,7 +25,7 @@ Adapt = "4"
 Aqua = "0.8"
 ArrayLayouts = "1"
 BlockArrays = "1"
-BlockSparseArrays = "0.8"
+BlockSparseArrays = "0.9"
 DerivableInterfaces = "0.5"
 DiagonalArrays = "0.3"
 GPUArraysCore = "0.2"