Typemap yields Upgrade(type) objects on load

[cstjean]

Thank you for answering #627. So far it works well, however I found an issue:

using JLD2

struct Foo
    x::Int
end

jldsave("test.jld2"; type=Foo)

load("test.jld2", "type"; typemap=Dict("Main.Foo"=>JLD2.Upgrade(Foo)))

This should reasonably yield Foo, but it yields Upgrade(Foo). To put it another way, if I'm deserializing a Dict(Foo=>4), it'll become Dict(Upgrade(Foo)=>4).

[JonasIsensee]

Hi @cstjean,

that makes sense, given how things are implemented....
Things should work if you "just" do:
load("test.jld2", "type"; typemap=Dict("Main.Foo"=>Foo))
(Which then does not allow you to deserialize instances of the changed Foo in the same operation)...

I don't think there is an easy fix right now.
To improve things, we would have to add an extra information passage pathway into JLD2.
jlconvert loads a type and returns JLD2.Upgrade(Foo) in your case.
This is important for reconstructing instances, as it signals to the calling function that the rconvert constructor should be used instead of the default version.
However, this is somewhat inconsistent as jlconvert normally promises to return the requested type.

EDIT: also for reference

julia> jldsave("test.jld2"; type=DataType[Float64, Foo])

julia> load("test.jld2"; typemap=Dict("Main.Foo"=>JLD2.Upgrade(Foo)))
Error encountered while load FileIO.File{FileIO.DataFormat{:JLD2}, String}("test.jld2").

Fatal error:
ERROR: MethodError: Cannot `convert` an object of type JLD2.Upgrade to an object of type DataType

Closest candidates are:
  convert(::Type{T}, ::T) where T
   @ Base Base.jl:84

Stacktrace:
  [1] setindex!(A::Vector{DataType}, x::JLD2.Upgrade, i1::Int64)
    @ Base ./array.jl:1021
  [2] macro expansion
    @ /data.lmp/isensee/.julia/dev/JLD2/src/io/dataio.jl:125 [inlined]
  [3] macro expansion
    @ ./simdloop.jl:77 [inlined]
  [4] read_array!(v::Vector{…}, f::JLD2.JLDFile{…}, rr::JLD2.MappedRepr{…})
    @ JLD2 /data.lmp/isensee/.julia/dev/JLD2/src/io/dataio.jl:122