I have an application where I want to have a type to represent block
matrices and then call various linear algebra function which are
specialized to different situations. This can be neatly accomplished by
writing (for 2x2 blocks)
type BlockMatrix{TA,TB,TC,TD}
A::TA
B::TB
C::TC
D::TD
end
So that you can do things like
A,B,C = rand(4,4),Diagonal([1:1.:4]),Diagonal([2:1.:5])
Δ = rand(2,2)
D = BlockMatrix(Diagonal([1.,2.]),Δ,Diagonal([1.,2.]),Δ)
M= BlockMatrix(A,B,C,D)
typeof(M)
>>BlockMatrix{Array{Float64,2},Diagonal{Float64},Diagonal{Float64},
BlockMatrix{Diagonal{Float64},Array{Float64,2},Diagonal{Float64},Array{
Float64,2}}}
at which point it is indeed possible to do fine grained dispatch depending
on block types. Now, I don't want my block matrix type to accept just
anything though. Even before doing dimension checking I would at least like
to ensure that all the arguments are of Matrix type. It would also make
sense that the element type should be enforced. Since the blocks can be
composed of blocks themselves it would also seem prudent to make sure that
BlockMatrix has the same abstract type as its super. So I was able to write
type BlockMatrix{S,TA<:AbstractMatrix{S},TB<:AbstractMatrix{S},TC<:
AbstractMatrix{S},TD<:AbstractMatrix{S}}<:AbstractMatrix{S}
A::TA
B::TB
C::TC
D::TD
end
however, I then get
M= BlockMatrix(A,B,B,B)
`BlockMatrix{S,TA<:AbstractArray{S,2},TB<:AbstractArray{S,2},TC<:AbstractArray{S,2},TD<:AbstractArray{S,2}}`
has no method matching
BlockMatrix{S,TA<:AbstractArray{S,2},TB<:AbstractArray{S,2},TC<:AbstractArray{S,2},TD<:AbstractArray{S,2}}(::Array{Float64,2},
::Diagonal{Float64}, ::Diagonal{Float64}, ::Diagonal{Float64})
Indeed, calling methods(BlockMatrix) reveals that no constructors are
available. So basically, I am not really sure what is going on, having
apparently reached the limit of my current parametric type no jutsu. Can
anyone give any pointer (yes I saw the manual, but still not
understanding)? I am on 0.3.5.
