.task Modifier Does Not Fire

[MatiMax]

Issue

The .task modifier on views seems not to fire.

Tested Environment

The test has been conducted on

• Ubuntu Server 24.04 LTS with Swift toolchain version 6.3.2
• macOS Sequoia 15.7.7 with Swift toolchain version 6.2.4.1.4

and TUIKit main branch version.

How to Reproduce

import TUIkit

struct ContentView: View {
    @State private var taskHasRun = false
    
    var body: some View {
        VStack {
            Text("Hello, TUIkit!")
                .foregroundStyle(.palette.accent)
                .bold()
                .task { await flagTaskAsRun() }
                
            Text("Welcome to your new terminal app")
                .foregroundStyle(.palette.foregroundSecondary)
                .task { await flagTaskAsRun() }
            
            Spacer()
            
            Text(".task has \(taskHasRun == true ? "indeed":"not") run")
                .task { await flagTaskAsRun() }
        }
        .padding()
        .task { await flagTaskAsRun() }
    }
    
    func flagTaskAsRun() async {
        taskHasRun = true
    }
}

Observed Behaviour

The following simple test program shows that taskHasRun will never ever be set to true, so the last Text view will display .task has not run.

Expected Behaviour

With either VStack or any of the Text views the function flagTaskAsRun() should be fired and hence taskHasRun should be set to true, so the last Text view should display .task has indeed run.

[phranck]

@MatiMax, sorry this technical follow-up took a while, and thank you for the exact reproducer. I can reproduce the sample: the synchronous MainActor loop prevents the task from resuming into @State, while a fresh lifecycle UUID on every render would also restart the task. The runtime, event-loop, and stable-identity fixes are tracked in #8, #9, and #10; exact task(id:priority:) API coverage follows in #21.

[phranck]

@MatiMax, PR #52 now covers the stable task-identity side of your reproducer: the task remains mounted across view reconstruction, keeps the caller actor context, and can update the owning @State. A direct regression test uses the same task-to-State flow. Together with the event-loop correction in #51, the sample should switch to “task has indeed run” once both PRs are merged.