func.func @main() {
    tf_executor.graph {
      %control = tf_executor.island {
        tf_executor.yield
      }
      tf_executor.fetch %control : !tf_executor.control
    }
    tf_executor.graph {
      %control = tf_executor.island {
        tf_executor.yield
      }
      tf_executor.fetch %control : !tf_executor.control
    }
    return
  }

    // CHECK: "tf.ReadVariableOp"([[name]])
    %cond = builtin.unrealized_conversion_cast to tensor<i1>
    %0 = "tf.IfRegion"(%cond) ({
      "tf.Yield"(%arg0) : (tensor<!tf_type.resource<tensor<10xf32>>>) -> ()
    }, {
      "tf.Yield"(%arg0) : (tensor<!tf_type.resource<tensor<10xf32>>>) -> ()
    }) { is_stateless = false} : (tensor<i1>) -> tensor<!tf_type.resource<tensor<10xf32>>>

  auto it = block.rbegin();
  YieldOp yield = dyn_cast<YieldOp>(*it++);

  if (it == block.rend()) return std::nullopt;

  // Operation which is expected to consume all the call results.
  Operation* call_consumer = yield;

  // Allow a single ToBoolOp between the call and the yield (valid only
  // when the yield has a single operand)
  if (allow_to_bool && yield.getNumOperands() == 1 && isa<ToBoolOp>(*it)) {

      tf_executor.yield %0 : tensor<f32>
    }
// Uses two islands for no other reason than preventing canonicalization from
// eliminating the graph entirely.
    %2:2 = tf_executor.island(%1#1) {
      %4 = "tf.opB"(%1#0) : (tensor<f32>) -> tensor<f32>
      tf_executor.yield %4 : tensor<f32>
    }
    tf_executor.fetch %2#0 : tensor<f32>
  }

  func.return
}

// -----

// Check that a tf_executor.yield parent is a tf_executor.island.
func.func @parent_is_island() {
  "tf.some_op"() ({
    tf_executor.yield
// expected-error@-1 {{'tf_executor.yield' op expects parent op 'tf_executor.island'}}
  }) : () -> ()
  func.return
}

// -----

// Check that an island yield matches the island results.

        %4 = "tf.B"(%2) : (tensor<?xi32>) -> tensor<?xi32>
        tf_device.return %4 : tensor<?xi32>
      }) {device = "/device:test_device:0"} : () -> tensor<?xi32>

      // CHECK: tf_executor.yield %[[LAUNCH_OUTPUT]]
      tf_executor.yield %3 : tensor<?xi32>
    }
    tf_executor.fetch %1#0 : tensor<?xi32>
  }
  func.return %0 : tensor<?xi32>
}

// CHECK: func private @[[LAUNCH]]

### Lifespan-Funktion

Das Erste, was auffällt, ist, dass wir eine asynchrone Funktion mit `yield` definieren. Das ist sehr ähnlich zu Abhängigkeiten mit `yield`.

```Python hl_lines="14-19"
{!../../../docs_src/events/tutorial003.py!}
```

Der erste Teil der Funktion, vor dem `yield`, wird ausgeführt **bevor** die Anwendung startet.

Und der Teil nach `yield` wird ausgeführt, **nachdem** die Anwendung beendet ist.

### Asynchroner Kontextmanager

	}
	m2 := (*T).PM
	for range m2 /* ERROR "cannot range over m2 (variable of type func(*T)): func must be func(yield func(...) bool): argument is not func" */ {
	}
	for range f1 /* ERROR "cannot range over f1 (value of type func()): func must be func(yield func(...) bool): wrong argument count" */ {
	}

      %0 = "arith.constant" () {value = dense<0> : tensor<i32>} : () -> tensor<i32> loc("Const")
      %1 = "tf.NotEqual"(%condArg0, %0) : (tensor<*xi32>, tensor<i32>) -> tensor<i1>
      "tf.Yield"(%1) : (tensor<i1>) -> ()
    },
    // body
    {
    ^bb0(%bodyArg0: tensor<*xi32>, %bodyArg1: tensor<*xf32>):
      %0 = "arith.constant" () {value = dense<1> : tensor<i32>} : () -> tensor<i32> loc("Const")

A primeira coisa a notar, é que estamos definindo uma função assíncrona com `yield`. Isso é muito semelhante à Dependências com `yield`.

```Python hl_lines="14-19"
{!../../../docs_src/events/tutorial003.py!}
```

A primeira parte da função, antes do `yield`, será  executada **antes** da aplicação inicializar.

E a parte posterior do `yield` irá executar **após** a aplicação ser encerrada.

### Gerenciador de Contexto Assíncrono