assert that input and output buffers passed to transcode! are independent

[baumgold]

As suggested by @quinnj in #136 (comment)

[mkitti]

This is a good basic check, but I think we might want to actually check for aliasing. The Base.dataids seems to be a way to to do this. Maybe someone like @mbauman could help update us on the latest with regard to alias detection. My understanding from JuliaLang/julia#25890 is that we could just see if the results of Base.dataids on each array are similar.

help?> Base.dataids
  Base.dataids(A::AbstractArray)

  Return a tuple of UInts that represent the mutable data segments of an array.

  Custom arrays that would like to opt-in to aliasing detection of their component parts can specialize this method to
  return the concatenation of the dataids of their component parts. A typical definition for an array that wraps a
  parent is Base.dataids(C::CustomArray) = dataids(C.parent).

For normal arrays, it seems

julia> A = rand(1024);

julia> B = @view A[1:16];

julia> Base.dataids(A)
(0x0000014201b1a080,)

julia> Base.dataids(B)
(0x0000014201b1a080,)

julia> Base.dataids(A) == Base.dataids(B)
true

julia> pointer(A)
Ptr{Float64} @0x0000014201b1a080

[quinnj]

Thanks! Can we add a quick test too?

[mkitti]

@timholy just referred me to Base.mightalias.

help?> Base.mightalias
  Base.mightalias(A::AbstractArray, B::AbstractArray)

  Perform a conservative test to check if arrays A and B might share the same
  memory.

  By default, this simply checks if either of the arrays reference the same
  memory regions, as identified by their Base.dataids.

[mkitti]

I wonder if we are going to need an unsafe_transcode!. The alias detection might be too conservative.

[baumgold]

I wonder if we are going to need an unsafe_transcode!. The alias detection might be too conservative.

Agreed, added.