[mypyc] always add implicit None return type to __init__ method #9866
Conversation
thomasjohns
changed the title
mypyc/irbuild/mapper.py
Outdated
| @@ -132,7 +132,11 @@ def fdef_to_sig(self, fdef: FuncDef) -> FuncSignature: | |||
| else: | |||
| # Handle unannotated functions | |||
| arg_types = [object_rprimitive for arg in fdef.arguments] | |||
| ret = object_rprimitive | |||
| # We at least know the return type for __init__ will be None. | |||
| if fdef.name == '__init__': | |||
There was a problem hiding this comment.
Could fdef be a module-level function with the (confusing) name __init__? Add test for this. We should allow top-level functions to return any values, even if they are named __init__ for some reason.
(You should be able to compare the FuncDef info attribute against FUNC_NO_INFO to check whether it's a method.)
There was a problem hiding this comment.
Oh nice catch this will not work for module-level functions named __init__! I will add a check and test for this. Thank you for the FUNC_NO_INFO tip.
…nnotation to module-level __init__ functions
mypy/nodes.py
Outdated
| @@ -683,6 +683,10 @@ def __init__(self, | |||
| def name(self) -> str: | |||
| return self._name | |||
|
|
|||
| @property | |||
| def is_module_level_func(self) -> bool: | |||
| return self.info == FUNC_NO_INFO | |||
There was a problem hiding this comment.
@JukkaL my instinct was to "hide" the use of FUNC_NO_INFO. I'm not sure if that was a good instinct.
There was a problem hiding this comment.
Actually I think that this is unnecessary (see below).
mypyc/test-data/run-classes.test
Outdated
| [case testInitMethodWithMissingNoneReturnAnnotation] | ||
| class C: | ||
| def __init__(self): | ||
| pass |
There was a problem hiding this comment.
Also try constructing an instance of C (just make sure it produces a result).
mypy/nodes.py
Outdated
| @@ -683,6 +683,10 @@ def __init__(self, | |||
| def name(self) -> str: | |||
| return self._name | |||
|
|
|||
| @property | |||
| def is_module_level_func(self) -> bool: | |||
| return self.info == FUNC_NO_INFO | |||
There was a problem hiding this comment.
Actually I think that this is unnecessary (see below).
mypy/suggestions.py
Outdated
| @@ -437,7 +437,7 @@ def get_suggestion(self, mod: str, node: FuncDef) -> PyAnnotateSignature: | |||
| if self.no_errors and orig_errors: | |||
| raise SuggestionFailure("Function does not typecheck.") | |||
|
|
|||
| is_method = bool(node.info) and not node.is_static | |||
| is_method = not node.is_module_level_func and not node.is_static | |||
There was a problem hiding this comment.
Let's leave this as it was, since using an info attribute as a boolean is a common idiom.
mypyc/irbuild/mapper.py
Outdated
| @@ -132,7 +132,12 @@ def fdef_to_sig(self, fdef: FuncDef) -> FuncSignature: | |||
| else: | |||
| # Handle unannotated functions | |||
| arg_types = [object_rprimitive for arg in fdef.arguments] | |||
| ret = object_rprimitive | |||
| # We at least know the return type for __init__ methods will be None. | |||
| is_init_method = fdef.name == '__init__' and not fdef.is_module_level_func | |||
There was a problem hiding this comment.
I think that you can just use ... and bool(fdef.info). Your suggestion to use a property is reasonable, but we have many checks where we do the equivalent of if fdef.info so it would add some inconsistency.
There was a problem hiding this comment.
Okay thank you that makes sense. Applying this update and the suggestions above now.
…ove test case for missing __init__ return annotation
Description
Fixes mypyc/mypyc#792.
Always give
__init__an implicitNonereturn type so that programs likecompile.
Test Plan
Included a test program that doesn't compile with the existing application code.