You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
It would be nice to move away from the standard library Serialization, as it only guarantees that newer julia versions can read files written by older ones, and not the other way. Bi-directionality is needed.
julia> @time using JLD2
0.312673 seconds (316.65 k allocations: 18.135 MiB)
julia> @time using BSON
0.002721 seconds (3.12 k allocations: 288.133 KiB)
Also, BSON.jl has no dependencies, so I guess there is less that can go wrong there? It seems appropriate to use a minimal low-dependency library for serialization in the context of RemoteREPL, so I suggest BSON.jl. Is this a direction that can be agreed on?
The text was updated successfully, but these errors were encountered:
Yes we can do this and it's very desirable to have independence on the Julia version.
It may be tricky to support all features, however - can either of these serializations support arbitrary Julia objects? @remote relies on Serialization for transporting arbitrary Julia objects.
It would be nice to move away from the standard library Serialization, as it only guarantees that newer julia versions can read files written by older ones, and not the other way. Bi-directionality is needed.
Based on the recommendations in https://fluxml.ai/Flux.jl/stable/saving/, options are https://github.com/JuliaIO/JLD2.jl and https://github.com/JuliaIO/BSON.jl. Perhaps lightweightness should be a priority. On my system, in a temp environment, I get
Also, BSON.jl has no dependencies, so I guess there is less that can go wrong there? It seems appropriate to use a minimal low-dependency library for serialization in the context of RemoteREPL, so I suggest BSON.jl. Is this a direction that can be agreed on?
The text was updated successfully, but these errors were encountered: