wasmtime / PR #11717 Continuation trapping semantics · git-wasmtime

Can we add a test that spawns an N deep stack chain with M frames where every other frame is a host frame, and the last frame (whether host or Wasm) triggers a trap? Then we can run that test exhaustively for small N and M.

(module
  ;; The imported host function.
  (import "host" "func" (func $host_func (param i32 i32)))

  ;; A global that is incremented after calling the host
  ;; function, which should trap, and therefore the
  ;; increment should never happen.
  (global $g (export "g") (mut i32) (i32.const 0))

  (func (export "run") (param $frames-per-stack i32) (param $fuel i32)
    ;; Trap on out-of-fuel for frames.
    if (i32.eqz (local.get $fuel))
      unreachable
    end

    ;; Decrement frame fuel.
    (local.set $fuel (i32.sub (local.get $fuel) (i32.const 1)))

    if (i32.eqz (i32.rem (local.get $fuel) (local.get $frames-per-stack))))
      ;; TODO: Spawn a new stack, starting either with `run`
      ;; or our host function (based on another param or a
      ;; global or something), and switch to it...
    else
      ;; Call the host function to continue our mutual recursion.
      (call $host_func (local.get $frames-per-stack) (local.get $frame-fuel))
    end

    ;; Increment the global. Should never execute, dynamically.
    (global.set $g (i32.add (global.get $g) (i32.const 1))
  )
)

let host_func = Func::wrap(
    &mut store,
    |mut caller: Caller<'_, ()>, frames_per_stack: u32, fuel: u32| -> Result<()> {
        ;; Trap on out-of-fuel for frames.
        if fuel == 0 {
            bail!("out of frame fuel");
        }

        ;; Mutual recursion back into the Wasm function.
        let run = instance.get_typed_func::<(u32, u32)>(&mut caller).unwrap();
        run.call(&mut caller, (frames_per_stack, fuel - 1))?;

        ;; Increment the global. Should never execute, dynamically.
        let g = instance.get_global("g").unwrap();
        let g_val = g.get(&mut caller).unwrap_i32();
        g.set(&mut caller, Val::I32(g_val + 1))?;

        Ok(())
    },
);

// ...

for frames_per_stack in 1..4 {
    for fuel in 0..frames_per_stack * 3 {
        let mut store = Store::new(&engine, ());
        let instance = Instance::new(...)?;

        let run = instance.get_typed_func::<(u32, u32)>(&mut store)?;
        run.call(&mut store, (frames_per_stack, fuel))?;

        let g = instance.get_global(&mut store, "g").unwrap();
        assert_eq!(g.unwrap_i32(), 0);
    }
}

This would give me a lot more confidence that we are properly handling traps across stacks, regardless of the stack chain, host functions, and what kind of frame is youngest or oldest.

(And when we add embedder API support for spawning stacks, we should also extend the host function in this new test to use that support)

Stream: git-wasmtime

Topic: wasmtime / PR #11717 Continuation trapping semantics

Wasmtime GitHub notifications bot (Sep 19 2025 at 15:24):

Wasmtime GitHub notifications bot (Sep 19 2025 at 15:24):

Wasmtime GitHub notifications bot (Sep 19 2025 at 15:24):

Wasmtime GitHub notifications bot (Sep 19 2025 at 18:19):

Wasmtime GitHub notifications bot (Sep 19 2025 at 18:19):

Wasmtime GitHub notifications bot (Sep 21 2025 at 03:15):

Wasmtime GitHub notifications bot (Sep 21 2025 at 03:16):

Wasmtime GitHub notifications bot (Sep 22 2025 at 15:19):

Wasmtime GitHub notifications bot (Sep 23 2025 at 17:44):

Wasmtime GitHub notifications bot (Jan 21 2026 at 18:34):

Wasmtime GitHub notifications bot (Jan 22 2026 at 16:23):

Wasmtime GitHub notifications bot (Jul 23 2026 at 11:33):

Wasmtime GitHub notifications bot (Jul 23 2026 at 11:33):

Wasmtime GitHub notifications bot (Jul 23 2026 at 11:33):

Wasmtime GitHub notifications bot (Jul 23 2026 at 11:41):

Wasmtime GitHub notifications bot (Jul 23 2026 at 13:04):

Wasmtime GitHub notifications bot (Jul 23 2026 at 13:21):

Wasmtime GitHub notifications bot (Jul 23 2026 at 13:31):

Wasmtime GitHub notifications bot (Jul 23 2026 at 13:50):

Wasmtime GitHub notifications bot (Jul 23 2026 at 13:55):

Wasmtime GitHub notifications bot (Jul 23 2026 at 15:39):

Wasmtime GitHub notifications bot (Jul 23 2026 at 15:39):

Wasmtime GitHub notifications bot (Jul 23 2026 at 15:39):

Wasmtime GitHub notifications bot (Jul 24 2026 at 11:26):

Wasmtime GitHub notifications bot (Jul 24 2026 at 15:44):

Wasmtime GitHub notifications bot (Jul 24 2026 at 15:44):

Wasmtime GitHub notifications bot (Jul 24 2026 at 15:50):

Wasmtime GitHub notifications bot (Jul 24 2026 at 15:51):

Wasmtime GitHub notifications bot (Jul 24 2026 at 16:59):

Wasmtime GitHub notifications bot (Jul 24 2026 at 16:59):

Wasmtime GitHub notifications bot (Jul 24 2026 at 17:24):

Wasmtime GitHub notifications bot (Jul 24 2026 at 18:05):

Wasmtime GitHub notifications bot (Jul 25 2026 at 10:59):

Wasmtime GitHub notifications bot (Jul 25 2026 at 11:00):

Wasmtime GitHub notifications bot (Jul 27 2026 at 15:59):

Wasmtime GitHub notifications bot (Jul 27 2026 at 15:59):

Wasmtime GitHub notifications bot (Jul 27 2026 at 16:47):

Wasmtime GitHub notifications bot (Jul 27 2026 at 17:04):

Wasmtime GitHub notifications bot (Jul 27 2026 at 19:08):

Wasmtime GitHub notifications bot (Jul 27 2026 at 19:09):

Wasmtime GitHub notifications bot (Jul 27 2026 at 19:46):

Wasmtime GitHub notifications bot (Jul 27 2026 at 19:49):

Wasmtime GitHub notifications bot (Jul 27 2026 at 21:44):

Wasmtime GitHub notifications bot (Jul 28 2026 at 08:26):

Wasmtime GitHub notifications bot (Jul 28 2026 at 09:58):

Wasmtime GitHub notifications bot (Jul 28 2026 at 11:11):

Wasmtime GitHub notifications bot (Jul 28 2026 at 14:26):

Wasmtime GitHub notifications bot (Jul 28 2026 at 14:27):

Wasmtime GitHub notifications bot (Jul 28 2026 at 14:52):

Wasmtime GitHub notifications bot (Jul 28 2026 at 14:52):