func.func @simple_independent_chains(%arg0: !tf_res, %arg1: !tf_res, %arg2: tensor<f32>) {
  // CHECK: tf_executor.graph {
  tf_executor.graph {
    // CHECK: %[[A_CONTROL:.*]] = tf_executor.island wraps "tf.OpA"() : () -> ()
    %control_A = tf_executor.island wraps "tf.OpA"() : () -> ()
    // CHECK: %[[CHAIN_CONSTANT:.*]], %{{.*}} = tf_executor.island wraps "tf.Const"() <{value = dense<1> : tensor<i32>}> : () -> tensor<i32>

  %0 = tf_executor.graph {
    %1:2 = tf_executor.island {
      %3 = "tf.opA"(%arg0) : (tensor<i1>) -> tensor<i1>
      tf_executor.yield %3 : tensor<i1>
    }
    %2:2 = tf_executor.island(%1#1) {
      %4 = "tf.opB"() : () -> tensor<f32>
      tf_executor.yield %4 : tensor<f32>
    }

// CHECK: "tf.opA"
// CHECK: tf_executor.island
// CHECK: "tf.opB"

    tf_executor.fetch %2#0 : tensor<f32>
  }

func.func @island_not_direct_parent_of_user() -> () {
  tf_executor.graph {
    %island1:2 = tf_executor.island {
      // CHECK: %[[VAL_0:.*]], %[[VAL_1:.*]] = tf_executor.island wraps "tf.OpA"() : () -> tensor<i64>
      %0 = "tf.OpA"() : () -> (tensor<i64>)
      // CHECK: %[[VAL_2:.*]] = tf_executor.island(%[[VAL_1]]) wraps "tf.OpB"() : () -> ()
      "tf.OpB"() : () -> ()
      tf_executor.yield %0 : tensor<i64>
    }

 during TPURewritePass.

 For example, the tf.OpA with the `mini_batch_splits` attribute will be
 moved after _TPUCompileMlir and the first input will use the
 _TPUCompileMlir program output:

 ```mlir
 "tf_device.launch"() ({
  %cst_0 = "tf.Const"() {value = dense<""> : tensor<1x!tf_type.string>} : () -> tensor<1x!tf_type.string>
  "tf.OpA"(%cst_0) { mini_batch_splits = ""} : (tensor<1x!tf_type.string>) -> ()

    // CHECK-SAME:        _xla_has_host_transfer = true
    // CHECK:         }, {
    // CHECK:           "tf_device.cluster"
    // CHECK:             %[[A:.+]] = "tf.OpA"
    // CHECK:             %[[A_SHARD:.+]] = "tf.XlaSpmdFullToShardShape"(%[[A]]) <{dim = -1 : i64, manual_sharding = "\08\03\1A\02\02\01\22\02\00\01", unspecified_dims = []}> : (tensor<2x2xi64>) -> tensor<1x2xi64>

          %0 = "tf.opA"(%arg0) : (tensor<f32>) -> tensor<f32>
          return %0 : tensor<f32>
        }
      }
    }
    ```

    will be transformed into:

    ```mlir
    module  {
      func @foo(%arg0: tensor<f32>) -> tensor<f32> {
        %0 = tf_executor.graph {
          %outputs, %control = tf_executor.island {
            %1 = "tf.opA"(%arg0) : (tensor<f32>) -> tensor<f32>

    %handle0, %handle0_control = tf_executor.island wraps "tf.SomeOp"() : () -> !tf_res
    %handle1, %handle1_control = tf_executor.island wraps "tf.SomeOp"() : () -> !tf_res
    %control_A = tf_executor.island wraps "tf.OpA"() : () -> ()
    %while_out:2, %while_control = tf_executor.island(%control_A) wraps "tf.While"(
            %handle0, %handle1) {
        body = @body, cond = @cond, is_stateless = false

	}

	if authZPluginCfg.URL == nil {
		opaCfg, err := opa.LookupConfig(s[config.PolicyOPASubSys][config.Default],
			NewHTTPTransport(), xhttp.DrainBody)
		if err != nil {
			iamLogIf(ctx, fmt.Errorf("Unable to initialize AuthZPlugin from legacy OPA config: %w", err))
		} else {
			authZPluginCfg.URL = opaCfg.URL
			authZPluginCfg.AuthToken = opaCfg.AuthToken

}

// CHECK-LABEL: func @variant_input_output_already_marked
func.func @variant_input_output_already_marked() {
  // CHECK: "tf.opA"
  // CHECK-SAME: _xla_outside_compilation = "0"
  // CHECK: "tf.opB"
  // CHECK-SAME: _xla_outside_compilation = "0"
  "tf_device.cluster"() ({
    %1= "tf.opA"() {_xla_outside_compilation = "0"} : () -> tensor<!tf_type.variant<tensor<f32>>>

replicas and is mapped to a single region argument.

Operands not replicated can be implicitly captured by ops in the region. Results
are replicated each from the regions terminator.

For example:
```
%0 = "tf.opA"() : () -> tensor<i32>
%1 = "tf.opB"() : () -> tensor<i32>
%2 = "tf.opC"() : () -> tensor<f32>
%3 = "tf.opD"() : () -> tensor<f32>
%4 = "tf.opE"() : () -> tensor<!tf_type.resource>