`typeof(zero(x)) != typeof(x)` for BlockVector

[DanielVandH]

julia> using BlockArrays

julia> using BlockArrays: BlockedOneTo

julia> b = BlockVector([1,2,3,4,5,6,7,8,9,10], (BlockedOneTo(5:5:10),));

julia> typeof(b) == typeof(zero(b))
false

julia> typeof(b)
BlockVector{Int64, Vector{Vector{Int64}}, Tuple{BlockedOneTo{Int64, StepRange{Int64, Int64}}}} (alias for BlockArray{Int64, 1, Array{Array{Int64, 1}, 1}, Tuple{BlockedOneTo{Int64, StepRange{Int64, Int64}}}})

julia> typeof(similar(b))
BlockVector{Int64, Vector{Vector{Int64}}, Tuple{BlockedOneTo{Int64, Vector{Int64}}}} (alias for BlockArray{Int64, 1, Array{Array{Int64, 1}, 1}, Tuple{BlockedOneTo{Int64, Array{Int64, 1}}}})

I think the issue is because typeof(similar(b)) != typeof(b) in this case, since it collects the axes

julia> axes(b)
(BlockedOneTo(5:5:10),)

julia> axes(similar(b))
(BlockedOneTo([5, 10]),)

Would this be a bug? Should it preserve the axes?

[jishnub]

It does preserve the axes, just not the types of the axes.

julia> axes(zero(b)) == axes(b)
true

Maybe it can do better to preserve the types of the axes as well in this case. However, note that zero doesn't necessarily preserve the type of an object, particularly with ranges, and this isn't a bug:

julia> zero(1:2)
2-element Vector{Int64}:
 0
 0

[DanielVandH]

By preserve I was thinking also in terms of type (not === as in my previous edit*)

This issue came up when I was trying to solve an ODE with a block IC using TRBDF2(), which happens to use zero(u) in places and assumes zero(u) isa type(u). Maybe it's more on OrdinaryDiffEq for assuming that.

I can get around it by just using something like

BlockArray(b, (BlockArrays.BlockedOneTo(collect(axes(b)[1].lasts)),))

for now. If I need to I can probably look into it at some point I guess.

[jishnub]

This is indeed a common assumption, but this behavior isn't documented. zero only needs to satisfy x + zero(x) == x. However, we should be able to preserve the axes here in the similar call, I think.