GitHub

The block axis interface

A block array's block structure is dictated by its axes. These are typically BlockedOneTos, but may also be standard and non-blocked AbstractUnitRanges (which are assumed to correspond to a single block), or other type that implements the block axis interface.

Methods to implementBrief description
blockaxes(A)A one-tuple returning a range of blocks specifying the block structure
getindex(A, K::Block{1})return a unit range of indices in the specified block
blocklasts(A)Returns the last index of each block
findblock(A, k)return the block that contains the kth entry of A

The AbstractBlockArray interface

An arrays block structure is inferred from an axes, and therefore every array is in some sense already a block array:

julia> A = randn(5,5)
5×5 Matrix{Float64}:
  0.452801   -0.416508   1.17406    1.52575     3.1574  
  0.413142   -1.34722   -1.28597    0.637721    0.30655 
  0.34907    -0.887615   0.284972  -0.0212884  -0.225832
  0.466102   -1.10425    1.49226    0.968436   -2.13637 
 -0.0971956  -1.7664    -0.592629  -1.48947     1.53418 

julia> A[Block(1,1)]
5×5 Matrix{Float64}:
  0.452801   -0.416508   1.17406    1.52575     3.1574  
  0.413142   -1.34722   -1.28597    0.637721    0.30655 
  0.34907    -0.887615   0.284972  -0.0212884  -0.225832
  0.466102   -1.10425    1.49226    0.968436   -2.13637 
 -0.0971956  -1.7664    -0.592629  -1.48947     1.53418

It is possible to override additional functions to improve speed, however.

Methods to implementBrief description
Optional methods
BlockArrays.viewblock(A, i::Block)Specialised non-allocating X[Block(i...)], blocked indexing

For a more thorough description of the methods see the public interface documentation.

With the methods above implemented the following are automatically provided for arrays that are subtypes of AbstractBlockArray:

  • A pretty printing show function that uses unicode lines to split up the blocks:
julia> A = BlockArray(rand(4, 5), [1,3], [2,3])
2×2-blocked 4×5 BlockMatrix{Float64}:
0.61179   0.965631  │  0.696476   0.392796  0.712462
--------------------┼-------------------------------
0.620099  0.364706  │  0.0311643  0.27895   0.73477
0.215712  0.923602  │  0.279944   0.994497  0.383706
0.569955  0.754047  │  0.0190392  0.548297  0.687052
  • A bounds index checking function for indexing with blocks:
julia> blockcheckbounds(A, 5, 3)
ERROR: BlockBoundsError: attempt to access 2×2-blocked 4×5 BlockMatrix{Float64, Matrix{Float64}} at block index [5,3]
  • Happy users who know how to use your new block array :)