Int[].=missing vs Int[].+=missing

[rapus95]

Related to #33645
Back in the fill! case it was clear to me that we wanted to prefer type stability over special casing.
But here, it feels plainly inconsistent:

julia> Int[].+=missing
0-element Array{Int64,1}

julia> Int[].=missing
ERROR: MethodError: Cannot `convert` an object of type Missing to an object of type Int64
Closest candidates are:
  convert(::Type{Int64}, ::Type{CUDAnative.AS.Generic}) at C:\Users\Matthis\.julia\packages\CUDAnative\Lr0yj\src\device\pointer.jl:107
  convert(::Type{Int64}, ::Type{CUDAnative.AS.Global}) at C:\Users\Matthis\.julia\packages\CUDAnative\Lr0yj\src\device\pointer.jl:108
  convert(::Type{Int64}, ::Type{CUDAnative.AS.Shared}) at C:\Users\Matthis\.julia\packages\CUDAnative\Lr0yj\src\device\pointer.jl:109
  ...
Stacktrace:
 [1] fill!(::Array{Int64,1}, ::Missing) at .\array.jl:309
 [2] copyto! at .\broadcast.jl:871 [inlined]
 [3] materialize!(::Array{Int64,1}, ::Base.Broadcast.Broadcasted{Base.Broadcast.DefaultArrayStyle{0},Nothing,typeof(identity),Tuple{Base.RefValue{Missing}}}) at .\broadcast.jl:822
 [4] top-level scope at REPL[13]:1

I see that this is the case because of the referenced fill! case, but nevertheless, it irritates me. Is there any way to look onto this difference which is consistent?

[nalimilan]

Good catch. I guess the difference is accidental and due to how things are implemented. Not sure which approach is preferable, but at least until 2.0 we cannot turn Int[].+=missing into an error to avoid breaking code.

[mbauman]

but at least until 2.0 we cannot turn Int[].+=missing into an error to avoid breaking code.

I think we should probably go the other way, given that Int[] .= [missing] works without error. It may be worth reconsidering #33645 in this light, but the simpler answer would just be to add a 0-size check in the broadcast code before punting out to fill!.

julia/base/broadcast.jl

Line 871 in a633deb

return fill!(dest, bc.args[1][])