Hey@mgravell thanks for your work, is there any news about it?

Hi@mgravell , I know that you have a lot of work + family + combat criminals at night but I think this is the best protobuf library for dotnet, and Microsoft since net8 pushes a lot on performance + trimming + AOT + source generator, so I wanna ask:

• After years, there is any eta for this work?
• There is any chance to get help from microsoft to support this project as already did with Grpc.AspNetCore?

Anyway thank you for your work!

Hi; no hard ETA, but definitely still in progress; I'm very aware of the AOT work, and the hope is for the Dapper.AOT learnings to lead into the protobuf-net work; there exists an AOT branch for the analyzer pieces, but I think a lot of it will need some significant rework, but: I'm also a little distracted by Google's recent discussion of "edition 2024", and the "group" changes, which I also want to integrate (parser now works, so... yay!). This is relevant because the "editions" work and the "AOT" work need to interact, so understanding both pieces at the same time is essential.

As for MSFT time: my MSFT time is focused on cache work at the moment, but: let's see how it goes a little later in the year,

About AOT, it seems, that AssemblyBuilder.Save will work in .NET 9. I know generating c# code is better solution, but would this be supported? Generating serializer assemblies for AOT in some "model.csproj" after build step?

@michaldobrodenka if AssemblyBuilder.Save starts working, I'll happily light up that API, and if that unblocks some scenarios: great! However, that will be unrelated to and tangential to the intended AOT route, which I hope to be codegen based

Any news?

First off, thank you for your work! It is great!

I know this is not a rushed change (family, day job, etc.), but I was curious what could be done to help this PR along? Are there API improvements of code generators in .NET 9 that can be taken advantage of now?

The APIs haven't changed hugely (I don't think interceptors give us much); but I do need to revisit this from the ground up, using our learnings here as a foundation - the object model needs a lot of rework based on my learnings from Roslyn incremental generators over the last few years; the approach here is naive. Doable: yes. But it needs dedicated time.

Thanks for getting back so quickly, Marc!

Ah, I see. So then would there be an issue / milestone with TODOs etc. to give a roadmap of what needs to be done so that we could help contribute where able?

I started to play with generators and created a demo for protobuf generated serializers/deserializers from protobuf-net attributes.

https://github.com/michaldobrodenka/GProtobuf

It's far from usable, only deserialization is supported with only handful of types. Not tested/used. Just a proof of concept. Maybe will return to it sometimes. But when it's working, deserialization is crazy fast.

That's neat,@michaldobrodenka!

I am working on converting some code to be NativeAoT compliant and unfortunately haven't found a way to keep the NativeAoT runtime from trimming awayprotobuf-net. The only thing I have found so far is to useGoogle.Protobuf and manually create a.proto file for my DTOs, and it just ends up really messy...

But it did give me the idea (I haven't looked too deeply at this repo to see how feasible it is)--what if the[ProtoContract] and[ProtoMember(n)] attributes could create the.proto files automatically and and add the<Protobuf Include="car.proto" /> to the .csproj to generate theGoogle.Protobuf code that can then be used to automagically accomplish the same behavior in a NativeAoT context?

I am sure there are reasons that this wouldn't work, but with .NET 9 giving full NativeAoT support for iOS, I am seeing a lot of movement towards NativeAoT to get off of MonoAoT.

Eesh, I should just dust this off and ship something, even if it is incomplete. My plans are wider than my calendar, it seems.

Is there any chance v4 could bring back support forAsReference that was in v2 which allowed a full object graph to be serialized with multiple fields referencing the same object?

I'm trying to find a good serializer for Unity and ProtoBuf v2 is the only one I've found which meets all my needs except that I can't seem to use it in Android builds due to IL2CPP requiring AOT compilation so it would be a huge shame to find a solution to that problem only to lose such a useful feature.

@KybernetikGames did you looked atcysharp repos? They develop games with Unity and they are the creators of R3 (observables) and [Message/Memory]Pack (serializers) both developed in the way to be compatible with Unity.

Is there any chance v4 could bring back support forAsReference that was in v2 which allowed a full object graph to be serialized with multiple fields referencing the same object?

I'm trying to find a good serializer for Unity and ProtoBuf v2 is the only one I've found which meets all my needs except that I can't seem to use it in Android builds due to IL2CPP requiring AOT compilation so it would be a huge shame to find a solution to that problem only to lose such a useful feature.

If you need solution now, you can check my protobuf-net 2 fork - with precompile you can prepare serializer in post build step as a dll. I'm using it in production. And you don't need old net framework. It works with net6+https://github.com/michaldobrodenka/protobuf-net

@Dona278 I briefly tried MessagePack and MemoryPack but ran into issues with each of them (here andhere) which would have required me to refactor quite a bit of my code base. ProtoBuf v2 seemed like a silver bullet which handled everything I need to do with it right up until I tried to use it in a runtime build. But if I can't get it going then I'll definitely be revisiting the cysharp systems.

@michaldobrodenka I found your repo earlier today and have been trying to get it to work in Unity with no success so far and there's no Issues page so I wasn't sure how to contact you. Do you have a preferred contact method?

@KybernetikGames have you checkedaot-net6 branch?
I have added issues to this project, but I don't plan to maintain this project much further; I'm just using it until I find a replacement. It works on all my projects and I'm looking for more modern solution - using Span and code generated. Something like myGProtobuf which is only a proof of concept now.

I genuinely do have plans to revisit the AOT work. I just need the world to switch to a 36 hour day so I have enough hours in each...

Well, I just wanted to ask if you maybe had an outline of the work (that you know of so far) that needed to be done so that anyone who has the time and could contribute would (I have been looking intoIIncremementalGenerator and experimenting) be able to help?

I know I am certainly interested in contributing.

@mgravell I understand you very well, but are there any deadlines?