inductor: Don't throw an internal error when a nn.module is missing a…

… attribute (pytorch#145122)

If a nn.module getattr call throws, we should make sure that we don't crash with an internal error

Note that I couldn't figure out how to test this, so advice would be awesome.  I have my best case attempt at  pytorch#145799, but it doesn't seem to reproduce the crash.

Pull Request resolved: pytorch#145122
Approved by: https://github.com/jansel

test/dynamo/test_compile.py

@@ -188,6 +188,21 @@ def fn(x):
         with self.assertRaises(AttributeError):
             fn(x)
 
+    @torch._dynamo.config.patch(inline_inbuilt_nn_modules=False)
+    def test_compilation_nn_module_invalid_method(self):
+        class Mod(torch.nn.Module):
+            def __init__(self):
+                super().__init__()
+
+            def forward(self, x):
+                return x + self.doesnotexist
+
+        mod = Mod()
+        opt_mod = torch.compile(mod, backend="eager")
+        x = torch.randn(1, 1)
+        with self.assertRaises(AttributeError):
+            opt_mod(x)
+
 
 # The private variants of the below functions are extensively tested
 # So as long as the signatures match we're good

test/dynamo/test_modules.py

@@ -2764,6 +2764,7 @@ def run():
         run()
         self.assertTrue(models[0].abc)
 
+    @torch._dynamo.config.patch(inline_inbuilt_nn_modules=False)
     def test_assign_does_not_exist(self):
         class MyModule(torch.nn.Module):
             def forward(self, x):

torch/_dynamo/variables/builtin.py

@@ -20,6 +20,7 @@
 from .. import config, polyfills, variables
 from ..exc import (
     AttributeMutationError,
+    ObservedAttributeError,
     unimplemented,
     Unsupported,
     UserError,
@@ -1907,7 +1908,7 @@ def _lower_version_count_by_1(x):
 
                 try:
                     getattr_var = obj.var_getattr(tx, name_var.as_python_constant())
-                except AttributeError:
+                except (AttributeError, ObservedAttributeError):
                     getattr_var = None
 
                 if isinstance(getattr_var, variables.TensorVariable):

torch/_dynamo/variables/nn_module.py

@@ -288,8 +288,11 @@ def var_getattr(self, tx: "InstructionTranslator", name):
                 )
                 if result is not None:
                     return result
-                # if we can't find a __getattr__, just raise the AttributeError
-                raise
+                # if we can't find a __getattr__, we can't parse this, raise attribute error
+                raise_observed_exception(
+                    AttributeError,
+                    tx,
+                )
 
         if name == "forward":
             guard_to_detect_forward_monkeypatching(self.source, base)