What does it mean that backpropagation will happen into labels?

Question

I received an warning today, it said

softmax_cross_entropy_with_logits (from tensorflow.python.ops.nn_ops) is deprecated and will be removed in a future version. Instructions for updating:

Future major versions of TensorFlow will allow gradients to flow into the labels input on backprop by default.

and I then checked tf.nn.softmax_cross_entropy_with_logits_v2 as it suggested, and found something below

Backpropagation will happen into both logits and labels. To disallow backpropagation into labels, pass label tensors through a stop_gradients before feeding it to this function.

I don't quite understand what it means that backpropagation will happen into labels, aren't the labels set to be constants?