ISystem codegen fail

[little swan]

For some reason codegen fails to generate OnUpdate code for specific case

    [WorldSystemFilter(WorldSystemFilterFlags.ClientSimulation)]
    public partial struct ClientInitializePlayerInputSystem : ISystem
    {
        private EntityQuery _initQuery;
        private EntityQuery _toInitQuery;

        public void OnCreate(ref SystemState state)
        {
            _initQuery = SystemAPI.QueryBuilder().WithAll<PlayerInput>().WithNone<InitializedClientInput>().Build();

            _toInitQuery = SystemAPI.QueryBuilder().WithAll<PlayerInput, GhostOwnerIsLocal>().Build();
            state.RequireForUpdate(_toInitQuery);
        }

        public void OnUpdate(ref SystemState state)
        {
            if (!_initQuery.IsEmptyIgnoreFilter)
            {
                return;
            }

            var count = _toInitQuery.CalculateEntityCount();
            if (count != 1)
            {
                Debug.LogError("More than one locally owned input initialized.");
            }

            Debug.Log("InitializedInput");

            state.EntityManager.AddComponent<InitializedClientInput>(_initQuery);
        }
    }

Here full code of system

So it seems like OnUpdate is not generated at all

and debugging session confirm that

During debugging it feels like code is just going from start to end, without ever triggering any debug log

[last stream]

You sure system is running?

[little swan]

and no logs appear at all, even though system runs past initial early out for sure

[little swan]

yes, it triggers breakpoints

[last stream]

You don't have any code gen inside on update

[little swan]

for early out - every frame

after early out - only once

[last stream]

So I don't really get what needs to generate?

[little swan]

(which it should)

[little swan]

well, if it's not codegen problem, then it's something else

let me show

[last stream]

Never mind somehow missed the query create

[little swan]

here how breakpoint triggers

and here what happens as soon as I press Step Over (only once)

[last stream]

Why do you have require for update and IsEmptyIgnoreFilter (offtopic just a note)

[little swan]

I want to tag only on Client owned entity

different queries

[last stream]

Why not both require for update, anyway not the problem

[little swan]

because I want system not to run when other query exists

it's !IsEmtpy

[last stream]

Oh my bad ok

Yeah I don't know what's going on here

[little swan]

@outer patio might be worth taking a look

[last stream]

as far as i can tell the code is running?

and all you're saying is Debug.Log isn't executing

[little swan]

code is running, but it's not logging and it's not debugged correctly

also

no component is added

[last stream]

there should be no code gen because you have no code gen

its not burst compiled

so it's just mono

[little swan]

yeah, that's what makes it even more weird

[last stream]

i'd suggest just closing unity, clearing scriptassemblies

and reloading

[little swan]

I also nuked library

[last stream]

oh?

[little swan]

yep

[last stream]

seems unlikely something so simple is broken 🤔 but if you've nuked library i got no idea

code gen looks fine

just curiously, what happens if you burst compile it

wait can you check something, go to the directory of hte .cs file

there is no cloned ~ version of the file is there?

[little swan]

No, I already checked 😅

this has been bothering me too a lot lately, so that was first thing I checked

[last stream]

oh you suffering as well? 😦

[little swan]

it was burst compiled initially and it didn't work. now I'll check again

[last stream]

like a virus, slowly spreading

[little swan]

Bursting OnUpdate made no difference

[outer patio]

this is very weird. I'll see if this happens in an empty project, but if not I'll probably need a repro

[outer patio]

well hold on, you probably wanna debug log before the early out right

because if it debug logs there the thing is running and the isemptyignorefilter is true

that codegen is not the only codegen that happens for isystem, btw

that's the sourcegenerator stuff, which is for systemapi and ije and what have you, but there's also the ilpp stuff, which generates the bursted static entry point for onupdate etc

that stuff you can only see by opening the assembly in ilspy

[little swan]

debugging anything after early out leads to invalid code

[outer patio]

i'm saying just print before

to see if the onupdate is running

because if that print prints, the onupdate is running fine, and something else is funky

[little swan]

OnUpdate is running, yes. It's visible from both: System profiler and debugging early out code also works perfectly fine

[outer patio]

sooo what makes you think it's a codegen problem? i'm not sure i see anything in that onupdate that needs codegen

[little swan]

It was my first thought. Rn I have no idea what is failing 😅

I rewrote it like this and now it works fine

            if (SystemAPI.HasSingleton<InitializedClientInput>())
            {
                return;
            }

            var ecb = new EntityCommandBuffer(state.WorldUpdateAllocator);
            var count = 0;

            foreach (var (input, e) in SystemAPI.Query<PlayerInput>().WithAll<GhostOwnerIsLocal>().WithEntityAccess())
            {
                ecb.AddComponent<InitializedClientInput>(e);
                Debug.Log("InitializedInput");
                count++;
            }

            if (count != 1)
            {
                Debug.LogError("More than one locally owned input initialized.");
            }

            ecb.Playback(state.EntityManager);

[outer patio]

ah. well so if there's no systemapi etc for sourcegenerators to do, i think it probably just doesn't generate anything for onupdate

[little swan]

but something is funny about original code. It should be valid