🐛 Bug

I'd like to print the return of the gc.get_objects() function (I have memory issues) but I can't. Whenever I do it, I get the following error :

Traceback (most recent call last):
  File "test.py", line 5, in <module>
    print(gc.get_objects())
  File "<frozen importlib._bootstrap>", line 302, in _module_repr
  File "<frozen importlib._bootstrap>", line 295, in _module_repr
  File ".../site-packages/torch/_ops.py", line 60, in __getattr__
    op = torch._C._jit_get_operation(qualified_op_name)
RuntimeError: No such operator __file__::__file__

But what's more strange is that using the length are affecting it to a variable (not printing it) doesn't produce an error.

To Reproduce

I'm using python3.7.0 and latest version of pytorch (1.0.1.post2). Just run the following code in a file :

import gc
import torch
print(len(gc.get_objects))  # Works
a = gc.get_objects()  # works
print(gc.get_objects())  # Fails
for obj in gc.get_objects():
    print(obj)  # Works until encountering torch
                    # Though it works on some files
print(sum(map(sys.getsizeof, gc.get_objects())))  # Works
print(collections.Counter(map(type, gc.get_objects())))  # Works

Possible understanding

I think there's a use of the __file__ variable somewhere and according to this answer of SO, if the application is frozen it just fails using the variable.

Hope you can help

Edit : added more working examples
Edit 2 : I conducted some experiments, and it seems it's only due to the printing part. I navigated in the files but don't understand where to look when I encounter torch._C