This kind of instability is really not friendly to data science users.
Ad. 1. But you do not have to have x86 architecture and other architectures can have different low-level byte layout.
Ad. 2. Both are susceptible to this. Check out change of Char specification between Julia 0.6 and 1.0 and the problems it causes.
Ad. 3. This HDF5 is not fully intechangeable right away as different platforms can use different storage layout (e.g. column major vs row major).
In general, I think the best thing we can do to make it sustainable is to submit issues/PRs to JLD2 or BSON packages.
I am wondering whether the situation of JLD.jl vs JLD2.jl is still the same or not. I noticed JLD.jl has been updated recently, and the current official version 0.9.0 works quite well on Julia v1.0.1. On the other hand, JLD2.jl does not work for some data structures as pointed out as https://github.com/JuliaIO/JLD2.jl/issues/107. Iām torn apart if I should keep using JLD or switch to JLD2ā¦
I guess it depends on whether you need those data structures that are not supported by JLD2. FWIW, I have been using JLD2 since my initial inquiry, and have not run into any problems.