Raise MutationError if there are side effects when returning generator #145223
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/145223
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 206caf7 with merge base 44b69b8 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot rebase |
|
@pytorchbot started a rebase job onto refs/remotes/origin/viable/strict. Check the current status here |
|
Successfully rebased |
test/dynamo/test_generator.py
Outdated
| @unittest.expectedFailure | ||
| def test_reconstruct_generator_mutate_tensor(self): |
There was a problem hiding this comment.
Can you file an issue for this once this PR goes in so that we can track it as a known issue?
There was a problem hiding this comment.
Here #146628
| def whoo(t): | ||
| yield t.sin() | ||
| yield t.cos() | ||
| yield t.tan() |
There was a problem hiding this comment.
@StrongerXi @anijain2305 do we have any side effects that aren't variable mutation? We're trying to test a generator with side effects
There was a problem hiding this comment.
Maybe register_hook for input tensor?
Simpler proposal -- what about passing in some pre-existing dict or dummy object and update that (e.g., before the first yield)?.
There was a problem hiding this comment.
Oh, good thought. So, maybe it isn't safe to support any kind of variable mutation when reconstructing a generator.
6 participants
Stack from ghstack (oldest at bottom):
enable_faithful_generator_behaviorflag to True #142513CLEANUP_THROWbytecode #144420generator.throw(exception)#144424generator.close()#144423generator.send(..)#144422generator.__iter__()#144421cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang @amjames