add mypy-based validation, drop syntactic check on return type (#536)

* add mypy-based validation, drop syntactic check on return type

* resurrects tests for test_handlers_llm_encoding.py

* switched to type error instead of value error

* neatened up type checking tests

* added mypy to llm dependencies

* switched to use typing_extensions TypeAliasType

* moved type checking to an operation

* refined exception guard and switched to any instead of skipping bindings with un-representable types

* refactored code to construct asts, added more systematic tests, and checks

* updated deps of llm submodule

* llm tests all pass

* minor bug

* ruff formatting

* updated imports to include sys.modules

* restricted sys.modules in imports and suppressed warnings on importing untyped modules

* added ignore to ast.FunctionDef

* updated to use ruff to clean up generated code and avoid redundant mypy lag

* updated codeadapt fixture

* added --unsafe-fixes to ruff invocation

* updated prompt

* format notebook

* format notebook

* switched from ruff to autoflake

* fix for __init__

* fixed test with __init__ returning None

* final fixes

* added type_checking

Author: kiranandcode

docs/source/llm.ipynb

@@ -272,7 +272,7 @@ def _format_callable_type(callable_type: type[Callable]) -> str:
 
         if param_types is ...:
             params_str = "..."
-        elif isinstance(param_types, (list, tuple)):
+        elif isinstance(param_types, list | tuple):
             params_str = ", ".join(getattr(t, "__name__", str(t)) for t in param_types)
         else:
             params_str = str(param_types)
@@ -366,14 +366,6 @@ def _validate_signature_callable(
             "decode() requires synthesized function to have a return type annotation"
         )
 
-    expected_name = getattr(expected_return, "__name__", str(expected_return))
-    actual_name = getattr(actual_return, "__name__", str(actual_return))
-    if expected_name != actual_name:
-        raise ValueError(
-            f"decode() expected function with return type {expected_name}, "
-            f"got {actual_name}"
-        )
-
 
 @dataclass
 class CallableEncodable(Encodable[Callable, SynthesizedFunction]):
@@ -455,6 +447,11 @@ def decode(self, encoded_value: SynthesizedFunction) -> Callable:
         # Validate signature from AST before execution
         _validate_signature_ast(last_stmt, self.expected_params)
 
+        # Type-check with mypy; pass original module_code so mypy sees exact source
+        evaluation.type_check(
+            module, self.ctx, self.expected_params, self.expected_return
+        )
+
         # Compile and execute
         # https://docs.python.org/3/library/functions.html#exec
         g: MutableMapping[str, Any] = {}
@@ -620,7 +617,7 @@ def _encodable_callable(
 
     # Ellipsis means any params, skip param validation
     expected_params: list[type] | None = None
-    if param_types is not ... and isinstance(param_types, (list, tuple)):
+    if param_types is not ... and isinstance(param_types, list | tuple):
         expected_params = list(param_types)
 
     return CallableEncodable(ty, typed_enc, ctx, expected_params, expected_return)