Deepcopy fails for an edge case.

[AstitvaAggarwal]

using Test
struct test_struct
    a::Vector{Float64}
    b::Int64

    function test_struct()
        return new()
    end
end

object=test_struct()
copy=deepcopy(object)
@test object==copy       # Fails

And a more general edge case as objects from structs of the kind :

struct TestStruct3
    a::Vector{Float64}
    b::Vector{Float64}
    c::Int64
    TestStruct3() = new([0.0])
end

[LilithHafner]

Thanks for finding this!

The result of deepcopy is not actually expected to be equal to the original. == is defined as === be default and for non isbits types, a deep copy will not be === to the original. For example,

julia> struct T
           a::Vector{Float64}
       end

julia> t = T([1.0, 2.0])
T([1.0, 2.0])

julia> t2 = deepcopy(t)
T([1.0, 2.0])

julia> t == t2
false

However, I do agree that it would be nice for all the bit fields to be equal even in the presence of undefined fields.

[AstitvaAggarwal]

@LilithHafner I understand, but I'm not able to see another way to create a copy of an immutable struct's object which was created via a non initialising constructer. The undef fields for the non isbitstype should remain undef in the copy and the isbitstype must copy the default values. (Unless we somehow define a new constructer, which takes in only the copied isbitstype values and creates an object of exactly the original struct)

[AstitvaAggarwal]

The struct's object is non isbits type as it has a field which can be vector. But for a completely non initialised object with a vector field, the fields is undef. This field is itself immutable now as there is no vector nor the immutable object can change. Basically the field behaves similar to isbits type of struct's but is a special case as its not exactly isbits type simply due to a Vector being ideally stored.