return &asyncCall{region: call, async: call, scope: fun, fun: fun}
}

var forbidden = []string{
	"FailNow",
	"Fatal",
	"Fatalf",
	"Skip",
	"Skipf",
	"SkipNow",
}

// forbiddenMethod decomposes a call x.m() into (x, x.m, m) where
// x is a variable, x.m is a selection, and m is the static callee m.
// Returns (nil, nil, nil) if call is not of this form.

// canonicalization.
//
// Operations in tf dialect are lowered to XLA HLO through the following steps:
//   . Legalizes control flow operations.
//   . Decomposes compound resource operations so that the only remaining
//     operations on resource variables are resource reads/writes..
//   . Replaces resource reads/writes with function inputs/outputs and

  for (int i = state_size + 1; i < body_args.size(); ++i) {
    body_returns.push_back(body_args[i]);
  }
  builder.create<TF::YieldOp>(loc,
                              /*operands=*/body_returns);
}

// Decomposes any ReduceDatasetOps in `function` into a dataset iteration and a
// call to the reduce function in the ReduceDatasetOp.
LogicalResult DecomposeReduceDatasetInFunction(FuncOp function) {
  if (!llvm::hasSingleElement(function))

	if x != 0 {
		utf8.EncodeRune(buf[2*JamoUTF8Len:], jamoTBase+x)
		return 3 * JamoUTF8Len
	}
	return 2 * JamoUTF8Len
}

// decomposeHangul algorithmically decomposes a Hangul rune into
// its Jamo components.
// See https://unicode.org/reports/tr15/#Hangul for details on decomposing Hangul.
func (rb *reorderBuffer) decomposeHangul(r rune) {
	r -= hangulBase
	x := r % jamoTCount

	if rv := reflect.ValueOf(vr); rv.Kind() == reflect.Pointer &&
		rv.IsNil() &&
		rv.Type().Elem().Implements(valuerReflectType) {
		return nil, nil
	}
	return vr.Value()
}

// decimal composes or decomposes a decimal value to and from individual parts.
// There are four parts: a boolean negative flag, a form byte with three possible states
// (finite=0, infinite=1, NaN=2), a base-2 big-endian integer

	if x != 0 {
		utf8.EncodeRune(buf[2*JamoUTF8Len:], jamoTBase+x)
		return 3 * JamoUTF8Len
	}
	return 2 * JamoUTF8Len
}

// decomposeHangul algorithmically decomposes a Hangul rune into
// its Jamo components.
// See https://unicode.org/reports/tr15/#Hangul for details on decomposing Hangul.
func (rb *reorderBuffer) decomposeHangul(r rune) {
	r -= hangulBase
	x := r % jamoTCount

  // The conversion pipeline has to follow the following orders:
  // 1) Saved model related optimization like decompose resource ops
  // 2) Convert composite functions like lstm/rnns, along with proper function
  // inlining & dce.
  // 3) Lower static tensor list pass.

  // This decomposes resource ops like ResourceGather into read-variable op
  // followed by gather. This is used when the saved model import path is used

        }
        return %0 : tensor<i32>
      }
    ```
  }];
}

def DecomposeReduceDatasetPass : Pass<"tf-decompose-reduce-dataset", "mlir::func::FuncOp"> {
  let summary = "Decomposes ReduceDataset op into dataset operations.";

  let description = [{
    Decomposes ReduceDataset op into a while loop that iterates the dataset and calls
    into the reduction function.  This decomposition is only done if the

  explicit DecomposeTFOpsPass(std::optional<ModuleOp> external_tfr_module)
      : external_tfr_module_(external_tfr_module) {}

  StringRef getArgument() const final { return "tfr-decompose"; }

  StringRef getDescription() const final {
    return "Decompose TF ops with the registered composition library.";
  }

  void runOnOperation() override;

 private:
  // Apply canonicalization, mainly constant folding, on the function.

        if optional:
          return (None, None)
        else:
          raise KeyError('OpDef does not exist: ' + f_name)
      elif len(compose_dec) > 1:
        raise KeyError('More than one TF ops decomposes for.')
      else:
        op_name = compose_dec[0].args[0].value

    op_def = op_def_registry.get(op_name)
    if not op_def:
      raise ValueError('Not a registered op: ' + op_name)