Loading complex types seems to be broken with the latest HDF5_jll release

[rgaudette]

With the latest HDF5_jll release it seems that loading JLD files is broken. I can successfully run the example described in the Types and their definitions section. But, if I make the example a little more complex by adding a sparse matrix to the example, I am not able to load the data I have just written. To be more specific, if I define MyTypes as:

module MyTypes
export MyType
using SparseArrays
struct MyType
    value::Int
    s_array::SparseMatrixCSC
end
end

create a an instance of a MyType with

using SparseArrays
using MyTypes
x = MyType(3, spdiagm(-1 => [1,2,3,4], 1 => [4,3,2,1]))

saving the instance with

jldopen("somedata.jld", "w") do file
    addrequire(file, MyTypes)
    write(file, "x", x)
end

appears work fine.

But when trying to load it back in with
d = JLD.load("somedata.jld")
I receive the following error:

ERROR: stored type MyTypes.MyType does not match currently loaded type
Stacktrace:
 [1] jldatatype(::JLD.JldFile, ::HDF5.HDF5Datatype) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\jld_types.jl:723
 [2] read(::JLD.JldDataset) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\JLD.jl:367
 [3] read(::JLD.JldFile, ::String) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\JLD.jl:343
 [4] #43 at .\none:0 [inlined]
 [5] iterate at .\generator.jl:47 [inlined]
 [6] collect_to!(::Array{Tuple{String,Integer},1}, ::Base.Generator{Array{String,1},JLD.var"#43#45"{JLD.JldFile}}, ::Int64, ::Int64) at .\array.jl:732
 [7] collect_to!(::Array{Tuple{String,UInt32},1}, ::Base.Generator{Array{String,1},JLD.var"#43#45"{JLD.JldFile}}, ::Int64, ::Int64) at .\array.jl:740
 [8] collect_to_with_first!(::Array{Tuple{String,UInt32},1}, ::Tuple{String,UInt32}, ::Base.Generator{Array{String,1},JLD.var"#43#45"{JLD.JldFile}}, ::Int64) at .\array.jl:710
 [9] collect(::Base.Generator{Array{String,1},JLD.var"#43#45"{JLD.JldFile}}) at .\array.jl:691
 [10] (::JLD.var"#42#44")(::JLD.JldFile) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\JLD.jl:1246
 [11] jldopen(::JLD.var"#42#44", ::String, ::Vararg{String,N} where N; kws::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\JLD.jl:243
 [12] jldopen(::Function, ::String, ::String) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\JLD.jl:241
 [13] load(::FileIO.File{FileIO.DataFormat{:JLD}}) at C:\Users\rick\.julia\packages\JLD\jeGJb\src\JLD.jl:1245
 [14] load(::String; options::Base.Iterators.Pairs{Union{},Union{},Tuple{},NamedTuple{(),Tuple{}}}) at C:\Users\rick\.julia\packages\FileIO\2fEu2\src\loadsave.jl:136
 [15] load(::String) at C:\Users\rick\.julia\packages\FileIO\2fEu2\src\loadsave.jl:136
 [16] top-level scope at REPL[10]:1

I also tried adding a addrequire(file, SparseArrays) to the do block and that produced the same error.

[jacobcook1995]

+1 I have been getting an identical error. When I revert to the previous version of HDF5_jll the error stops occurring. When I do this data that was written out with the new version of HDF5_jll can be correctly read in, which might suggest the problem is at the reading rather than writing stage

[szafna]

+1
[0234f1f7] + HDF5_jll v1.12.0+0
Julia 1.5.3
same error on Windows and Linux
MWE

using JLD
Base.@kwdef mutable struct profile9
    end_node::Array{Int32,2} = [0 0 0]
	
end

save("p9.jld","prf",profile9())
load("p9.jld")

ERROR: stored type profile9 does not match currently loaded type

[szafna]

+1 I have been getting an identical error. When I revert to the previous version of HDF5_jll the error stops occurring. When I do this data that was written out with the new version of HDF5_jll can be correctly read in, which might suggest the problem is at the reading rather than writing stage

Is there a way to install previous version of HDF5_jll?

[jmert]

I think this is a compatibility bounding error in HDF5.jl v0.13.6 which should stick with the HDF5_jll v1.10 libraries — only the new HDF5.jl v0.14 should be using HDF5_jll v1.12. (JLD right now is version bounded on HDF5.jl v0.13 series.)

[jacobcook1995]

+1 I have been getting an identical error. When I revert to the previous version of HDF5_jll the error stops occurring. When I do this data that was written out with the new version of HDF5_jll can be correctly read in, which might suggest the problem is at the reading rather than writing stage

Is there a way to install previous version of HDF5_jll?

In the julia package manager I just wrote:
add HDF5_jll@1.10
which pins the HDF5_jll package to v1.10.5+7 (the intermediate versions don't seem to exist for some reason)

[musm]

This should now be fixed thanks to @jmert fixing the compat bounds on HDF5.jl
JuliaRegistries/General#24885 (comment)

[GregPlowman]

I'm still getting this error.
After Pkg.update() I have this in my environment:

JLD v0.10.0
HDF5 v0.13.6
HDF5_jll v1.12.0+0

Which packages(s) do I need to downgrade?

[jmert]

HDF5_jll needs to be downgraded back to v1.10.5, but that should be imposed if you update HDF5 to v0.13.7.

[GregPlowman]

Thanks, that worked!
Seems resolver preferenced higher HDF5_jll rather than higher HDF5.
So required manually upgrading HDF5.

[musm]

Hmm wouldn't the resolver pick out the highest compatible HDF5 version which should be the latest tagged version which would then downgrade HDF5_jll?

[musm]

I'm also curious what exactly is incompatible in the newer hdf5 library version. The errors here don't seem to indicate what's going on to trigger it. This might be easier to see using jmert's upgraded branch.

[oskarpfeffer]

This can also be solved by uncommenting lines 723-725 of jld_types.jl
I.e.

# if T !== Expr  # in julia >= 0.7 Expr has 1 fewer fields; the trailing field can be ignored.
#     dtype == newtype || throw(TypeMismatchException(typename))
# end

Trying to print out dtype and newtype returns the same and loading the data still works.
Don't know why, just wanted to add that info.