Cloning of torch module behaves unexpectedly about torch

[sebffischer]

I tried to do this myself, but because clone is autogenerated, I had to change the Declarations file (I believe), but when I try to autogenerate the functions I get:

> torchgen::generate("~/gh/torch")
Starting code generation ...
Error in `purrr::map_chr()`:
ℹ In index: 1.
Caused by error:
! Result must be length 1, not 0.

[sebffischer]

This kind of still breaks when one wants to do something like nn_parameter(torch_tensor(1))$clone().
While admittedly this might be less important, I think the (autogenerated)torch_clone() method itself should just preserve the attributes.

[dfalbel]

I see, you are right! Perhaps you can add a method like clone_with_attributes around here:

torch/R/tensor.R

Line 14 in 0e9fdd7

print = function(n = 30) {

With the behavior you desire and then I can make the plumbing for it to be the default clone.

TBH I haven't executed much torchgen like this, almost always doing load_all() then generate() with the defaults. So maybe there's a hardcode path that doesn't work properly when running from a different directory.