testValues := []string{"100m", "1Gi"}
	zero := MustParse("0")
	for _, testValue := range testValues {
		value := MustParse(testValue)
		v1 := value.DeepCopy()
		// ensure non-zero - zero = non-zero (suffix preserved)
		v1.Sub(zero)
		// ensure we preserved the input value
		if v1.String() != testValue {
			t.Errorf("Expected %v, actual %v", testValue, v1.String())
		}

func.func @from_value(%arg0: tensor<f32>) {
  // CHECK-NOT: Optional
  %0 = "tf.OptionalFromValue"(%arg0) : (tensor<f32>) -> tensor<!tf_type.variant<tensor<f32>>>
  return
}

// -----

// CHECK-LABEL: @get_value
func.func @get_value(%arg0: tensor<!tf_type.variant<tensor<f32>>>) {
  // CHECK-NOT: Optional
  %1 = "tf.OptionalGetValue"(%arg0) : (tensor<!tf_type.variant<tensor<f32>>>) -> tensor<f32>
  return
}

// -----

	return req
}

func validateTestMultipartContents(t *testing.T, req *Request, allMem bool) {
	if g, e := req.FormValue("texta"), textaValue; g != e {
		t.Errorf("texta value = %q, want %q", g, e)
	}
	if g, e := req.FormValue("textb"), textbValue; g != e {
		t.Errorf("textb value = %q, want %q", g, e)
	}
	if g := req.FormValue("missing"); g != "" {

    async def needy_dependency(fresh_value: Annotated[str, Depends(get_value, use_cache=False)]):
        return {"fresh_value": fresh_value}
    ```

=== "Python 3.8+ non-Annotated"

    !!! tip
        Prefer to use the `Annotated` version if possible.

    ```Python hl_lines="1"
    async def needy_dependency(fresh_value: str = Depends(get_value, use_cache=False)):
        return {"fresh_value": fresh_value}
    ```

		var got []int
		for k := range Keys(m) {
			got = append(got, k)
		}
		slices.Sort(got)
		if !slices.Equal(got, want) {
			t.Errorf("Keys(%v) = %v, want %v", m, got, want)
		}
	}
}

func TestValues(t *testing.T) {
	for size := 0; size < 10; size++ {
		var want []int
		m := make(map[int]int)
		for i := range size {
			m[i] = i
			want = append(want, i)
		}

		var got []int

  // fallback legalization.
  auto cast_value = [&](OpBuilder& builder, Type result_type, ValueRange inputs,
                        Location loc) -> Value {
    return builder.create<mlir::tensor::CastOp>(loc, result_type,
                                                inputs.front());
  };
  addSourceMaterialization(cast_value);
}

  auto& then_branch = dataset_if.getThenBranch();
  then_branch.push_back(new Block);
  builder.setInsertionPointToEnd(&then_branch.front());
  // Add iterator operational data access inside if.
  auto get_value = builder.create<TF::OptionalGetValueOp>(loc, dataset_types,
                                                          get_next.getResult());
  SmallVector<Value, 4> reduce_fn_args;

				t.Errorf("at iteration %d got %d, %d want %d, %d", cnt, i, v, ei, ev)
			}
			ei--
			ev--
			cnt++
		}
		if cnt != size {
			t.Errorf("read %d values expected %d", cnt, size)
		}
	}
}

func TestValues(t *testing.T) {
	for size := 0; size < 10; size++ {
		var s []int
		for i := range size {
			s = append(s, i)
		}
		ev := 0
		cnt := 0
		for v := range Values(s) {
			if v != ev {

	}
	ptrVal.Elem().Set(defVal)
	return textValue{p}
}

func (v textValue) Set(s string) error {
	return v.p.UnmarshalText([]byte(s))
}

func (v textValue) Get() interface{} {
	return v.p
}

func (v textValue) String() string {
	if m, ok := v.p.(encoding.TextMarshaler); ok {
		if b, err := m.MarshalText(); err == nil {

      bool is_mul = llvm::isa<MulOp>(binary_op);
      auto new_filter =
          filter_cst.mapValues(filter_type.getElementType(), [&](APFloat it) {
            return (is_mul ? it * cst_value : it / cst_value).bitcastToAPInt();
          });
      // We recreate the constant op in case it is shared by the other ops. This
      // might increase the model size.
      auto new_filter_op = rewriter.create<ConstOp>(