[Bug]: RuntimeError: dictionary changed size during iteration

[ezyang]

🐛 Describe the bug

Turn off error suppression and then run test_factory in test_aot_cudagraphs.py. It has a failure that's:

  File "/raid/ezyang/pytorch-scratch2/torch/_dynamo/convert_frame.py", line 168, in <listcomp>              
    seen_ids[obj_id] = any([has_tensor(v) for v in obj.__dict__.values()])                         
  File "/raid/ezyang/pytorch-scratch2/torch/_dynamo/convert_frame.py", line 168, in has_tensor                  seen_ids[obj_id] = any([has_tensor(v) for v in obj.__dict__.values()])                                  
  File "/raid/ezyang/pytorch-scratch2/torch/_dynamo/convert_frame.py", line 168, in <listcomp>              
    seen_ids[obj_id] = any([has_tensor(v) for v in obj.__dict__.values()])                                  
  File "/raid/ezyang/pytorch-scratch2/torch/_dynamo/convert_frame.py", line 168, in has_tensor                  seen_ids[obj_id] = any([has_tensor(v) for v in obj.__dict__.values()])                                  
  File "/raid/ezyang/pytorch-scratch2/torch/_dynamo/convert_frame.py", line 168, in <listcomp>              
    seen_ids[obj_id] = any([has_tensor(v) for v in obj.__dict__.values()])         
RuntimeError: dictionary changed size during iteration

This shouldn't happen.

Error logs

na

Did Dynamo succeed?

• Does dynamo.optimize("eager") succeed?

Did AOT succeed?

• Did dynamo.optimize("aot_eager") succeed?

Did Inductor succeed?

• Does dynamo.optimize("inductor") succeed?

Minified repro

na