Input Forwarding and Datatype

TensorFlow General Discussion
1

Hi,
The C api has TF_ForwardInputOrAllocateOutput for forwarding an input if possible, the output shape can be altered but not the datatype.
I see examples where it could be possible to forward an input with a different output datatype, like Equal (a == b).

I see in the checks performed by forward_input (which ends up being called by above):

  // Check that input type matches.
  if (input_dtype(input_index) != output_dtype) {
    CHECK(!forward_expected);
    return nullptr;
  }
  // Check that the input and output sizes are compatible.
  if (input.tensor->shape().num_elements() != output_shape.num_elements()) {
    CHECK(!forward_expected);
    return nullptr;
  }

Instead of those two conditions, correct me if I am wrong, it seems better to check that the byte sizes match.

So why is that?