%0 = "arith.constant"() {value = 42 : i32} : () -> i32
  %1 = "arith.constant"() {value = 41 : i32} : () -> i32
  %2 = "tfr.build_list"(%0, %1) : (i32, i32) -> !tfr.attr
  %3 = "tfr.build_list"(%0, %1) : (i32, i32) -> !tfr.attr
  %4 = "tfr.build_list"(%2, %3) : (!tfr.attr, !tfr.attr) -> !tfr.attr
  func.return %4 : !tfr.attr

// CHECK-NEXT: %[[bl:.*]] = "tfr.build_list"(%[[id1]], %[[id2]]) : (!tfr.tensor, !tfr.tensor) -> !tfr.tensor_list
// CHECK-NEXT: tfr.return %[[bl]] : !tfr.tensor_list
}

// CHECK-LABEL: @tf__my_max_pool
tfr.func @tf__my_max_pool(%input_: !tfr.tensor, %stride_w: i64{tfr.name="stride_w"}, %stride_h: i64{tfr.name="stride_h"}) -> (!tfr.tensor) {
  %cst_1 = arith.constant 1 : i64

  %0 = tfr.get_element %arg0[%cst] : (!tfr.tensor_list<T>, index) -> !tfr.tensor
  func.return %0 : !tfr.tensor
}

// -----

// CHECK-LABEL: build_list
func.func @build_list(%arg0: !tfr.tensor<A>, %arg1: !tfr.tensor<B>) -> !tfr.tensor_list {
  %0 = "tfr.build_list"(%arg0, %arg1) : (!tfr.tensor<A>, !tfr.tensor<B>) -> !tfr.tensor_list
  func.return %0 : !tfr.tensor_list
}

// -----

// CHECK-LABEL: quant_act_range

      CHECK:        %[[list:.*]] = "tfr.build_list"(%[[qparam]]#0, %[[qparam]]#0) : (!tfr.tensor, !tfr.tensor) -> !tfr.tensor_list
      CHECK:        %[[factor:.*]] = tfr.quant_scale_factor(%{{.*}}, %[[list]]) : (f32, !tfr.tensor_list) -> (!tfr.tensor)
      CHECK:        %[[list1:.*]] = "tfr.build_list"(%[[factor]]) : (!tfr.tensor) -> !tfr.tensor_list

  }];
}

def TFR_BuildListOp : TFR_Op<"build_list", [Pure]> {
  let description = [{
   The `build_list` operation builds a tensor list from a list of tensors, or
   an tfr.attr from a list of scalars.

    Example:

    ```mlir
    %3 = tfr.build_list(%2, %1, %0) :
      (tfr.tensor, tfr.tensor, tfr.tensor) -> tfr.tensor_list
    %3 = tfr.build_list(%2, %1, %0) : (i32, i32, i32) -> tfr.attr
    ```
  }];

  %0 = "tfr.cast"(%arg0) : (tensor<f32>) -> !tfr.tensor
  %1 = "tfr.cast"(%arg1) : (tensor<f32>) -> !tfr.tensor
  %2 = "tfr.cast"(%arg2) : (tensor<f32>) -> !tfr.tensor
  %3 = "tfr.build_list"(%0, %1, %2) : (!tfr.tensor, !tfr.tensor, !tfr.tensor) -> !tfr.tensor_list
  %concat = tfr.call @tf__risc_concat(%3) : (!tfr.tensor_list) -> !tfr.tensor
  %4 = "tfr.cast"(%concat) : (!tfr.tensor) -> tensor<3xf32>

                                 derived_attrs))) {
        return failure();
      }
      continue;
    }

    // 2. The preceding op is a tfr.build_list op, which collects multiple
    // values with tensor types via the tfr.cast ops. These ops will be fused
    // to the current op as well, so all the tfr.cast op inputs will be inputs
    // to the result op.

    // Create the new operands. This is mapping the operands from the target
    // TF ops to the TFR function arguments. If the TFR function argument is
    // a tensor_list, a "tfr.build_list" op is used to concat the available
    // TF op operands. If the TFR function argument isn't a tensor/tensor_list,
    // a constant is created by using the attribute stored in the TF op or the

		mg.buildList = slices.Clip(mg.g.BuildList())
	})
	return mg.buildList
}

func (mg *ModuleGraph) findError() error {
	errStack := mg.g.FindPath(func(m module.Version) bool {
		_, err := mg.loadCache.Get(m)
		return err != nil && err != par.ErrCacheEntryNotFound
	})
	if len(errStack) > 0 {
		_, err := mg.loadCache.Get(errStack[len(errStack)-1])
		var noUpgrade func(from, to module.Version) bool

        # as attribute.
        val, ty = self._index_to_I64(val, ty)
      vals.append(val)
      tys.append(str(ty))

    list_val = self._ssa_name('list')
    self.emit('\n{} = "tfr.build_list"({})'.format(list_val, ', '.join(vals)))
    self._emit_with_loc(' : ({}) -> {}'.format(', '.join(tys), out_type), node)
    return (list_val, out_type)

  def visit_Tuple(self, node):