Remove the redundant element type without coping the source data

New to Julia
1

I have a vector

a = Union{Misisng, Int}[1, 2, 3]

The all elements of a is Int. Is there a way I can remove the Missing element type in place?

some_function!(a) #change the element type in place

a
3-element Vector{Int64}:
 1
 2
 3
2 
Vector{Int}(Union{Missing, Int}[1, 2, 3])

Edit: Oops, missed that you wanted in-place.

1 Like
3

No. In-place operations preserve the element type of the array, since they’re just setting elements - they don’t mess with the overall type of the array, only single elements. Other threads/objects may have a reference to this array as well, while expecting to be able to insert Missing at their leisure, so such operations are (in general) not safe to do.

5 Likes
4

There is, actually: unsafe_wrap(Array, Ptr{Int}(pointer(a)), length(a)) makes a Vector{Int} sharing the same data as a, without any copies [ref]. It’s inplace in terms of memory, but doesn’t change the a object itself.

1 Like
5

Technically correct, but using unsafe is well, unsafe. More precisely, I believe it is an implementation detail that a Vector{Union{Missing, Int}} can be interpreted as a Vector{Int}.

1 Like
6

Thanks much for your reply. Would this unsafe wrap method have impact on gc?

7

It is treated like any other allocated array and will break if the underlying implementation of arrays with Union eltype changes.

1 Like
8

Yes, that is right. In jupyter, the object drived by the unsafe wrap method just can’t be printed twice, the jupyter kernel would be broken.