Expliclty avoid recording when should_record_events is false in record_shapeenv_event (#138965)

Looking at the function record_shapeenv_event its hard to tell that it does not always run
but we do disable it by setting top level is_recording to True self.should_record_events is false
this makes it more explicit to avoid confusion and overloading is_recording.

alternativley we can rename is_recording to do_no_record.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/138965
Approved by: https://github.com/ezyang
ghstack dependencies: #138804

torch/fx/experimental/recording.py

@@ -253,7 +253,8 @@ def record_shapeenv_event(*, save_tracked_fakes: bool = False) -> Callable:
                 return r
 
             try:
-                if args[0].is_recording:  # type: ignore[has-type]
+                shape_env = args[0]
+                if not shape_env.should_record_events or shape_env.is_recording:  # type: ignore[has-type]
                     # If ShapeEnv is already recording an event, call the wrapped
                     # function directly.
                     #

torch/fx/experimental/symbolic_shapes.py

@@ -2918,7 +2918,7 @@ class ShapeEnv:
         )
 
         # This will make sure we only record the top-level function call.
-        self.is_recording = not self.should_record_events
+        self.is_recording = False
         # Keep track of the list of tracked fakes.
         self.tracked_fakes = tracked_fakes
         # List of events for reconstructing ShapeEnv at arbitrary points in time.