// valid result or a scalar boolean indicating the error. For unranked outputs
  // we don't know which one it is. TF shape inference turns unranked outputs
  // into ranked ones if it can statically evaluate the broadcast, see the shape
  // function of tf.Equal.
  auto ty = mlir::dyn_cast<RankedTensorType>(op.getType());
  if (!ty) {
    return rewriter.notifyMatchFailure(op, "requires a ranked output shape");
  }

  // This function is performance critical for op fusion patterns, e.g.
  // FuseBinaryOpToPrecedingAffine and FuseMulOrDivWithConv2dOrDepthwiseConv2d.
  // So a few specializations are provided to evaluate the math operation
  // more efficiently.

  // Specialization for f32 type.
  if (getType().cast<ShapedType>().getElementType().isF32()) {
    return ConstFoldBinaryOp<FloatAttr, float>(

		// Consider starting this as a background goroutine and retrieving the result
		// asynchronously when we're actually ready to build the package, or when we
		// actually need to evaluate whether the package's metadata is stale.
		p.setBuildInfo(ctx, opts.AutoVCS)
	}

	// If cgo is not enabled, ignore cgo supporting sources
	// just as we ignore go files containing import "C".

            "description": "MessageExpression declares a CEL expression that evaluates to the validation failure message that is returned when this rule fails. Since messageExpression is used as a failure message, it must evaluate to a string. If both message and messageExpression are present on a rule, then messageExpression will be used if validation fails. If messageExpression results in a runtime error, the runtime error is logged, and the validation failure message is...

// unknown reasons (callout for investigation in b/142268695), keeping the
// function names unchanged in an MLIR roundtrip causes test failures.
// TODO(b/142268695) Re-evaluate whether we need this class v.s. directly using
// and TF function name as MLIR function name after b/142268695 is root caused.
class NameUniquifier : public OpOrArgNameMapper {
 public: