Add support for detecting overloads with overlapping arities #5163

Michael0x2a · 2018-06-06T21:45:10Z

This commit addresses TODO 2 from #5119 by adding support for detecting overloads with partially overlapping arities.

It also refactors the is_callable_compatible method. Specifically, this pull request...

Pulls out a lot of the logic for iterating over formal arguments into a helper method in CallableType.
Pulls out logic for handling varargs and kwargs outside of loops.
Rearranges some of the logic so we can return earlier slightly more frequently.

This commit addresses TODO 2 from python#5119 by adding support for detecting overloads with partially overlapping arities. It also refactors the `is_callable_compatible` method. Specifically, this pull request... 1. Pulls out a lot of the logic for iterating over formal arguments into a helper method in CallableType. 2. Pulls out logic for handling varargs and kwargs outside of loops. 3. Rearranges some of the logic so we can return earlier slightly more frequently.

ilevkivskyi

Thanks! Looks very good. I have essentially only one non-trivial question about overlap due to empty arg list in a call.

ilevkivskyi · 2018-06-09T23:15:38Z

mypy/checker.py

@@ -3648,10 +3642,12 @@ def is_more_precise_or_partially_overlapping(t: Type, s: Type) -> bool:
    return (is_callable_compatible(signature, other,
                                   is_compat=is_more_precise_or_partially_overlapping,
                                   is_compat_return=lambda l, r: not is_subtype(l, r),
-                                   check_args_covariantly=True) or
+                                   check_args_covariantly=True,
+                                   allow_potential_compatibility=True) or


TBH, I don't like the term "potential compatibility". Why not "partial overlap"? After all, IIUC, this is the same as Union[A, B] vs Union[B, C] but for arg counts.

ilevkivskyi · 2018-06-09T23:16:46Z

mypy/subtypes.py

@@ -616,6 +617,30 @@ def g(x: int) -> int: ...

        In this case, the first call will succeed and the second will fail: f is a
        valid stand-in for g but not vice-versa.
+
+    allow_potential_compatibility:


See my comment above about terminology.

ilevkivskyi · 2018-06-09T23:22:09Z

mypy/subtypes.py

+
+    allow_potential_compatibility:
+        By default, this function returns True if and only if left is
+        definitely compatible with right.


Instead of "definitely" maybe add an explanation in terms of subtyping? This shouldn't be mathematically precise (I could guess why you are mostly using "compatibility" instead of "subtyping", but for historical reasons in mypy code compatibility is checked by is_subtype, while subtyping by is_proper_subtype). Maybe say something like each call that succeeds for right, will guaranteed to succeed for left callable.

ilevkivskyi · 2018-06-09T23:26:00Z

mypy/subtypes.py

+            g(*args: int) -> int
+
+        However, they would be *potentially* compatible under certain conditions --
+        for example, if the user runs "f_or_g(3)". So, if this flag is


Again, this will be probably more clear (at least for me), if you say that partial overlap means that some calls that succeed for right, will also succeed for the left callable.

ilevkivskyi · 2018-06-09T23:38:21Z

mypy/types.py

@@ -662,6 +662,8 @@ class CallableType(FunctionLike):
                 'min_args',  # Minimum number of arguments; derived from arg_kinds
                 'is_var_arg',  # Is it a varargs function?  Derived from arg_kinds
                 'is_kw_arg',  # Is it a **kwargs function?  Derived from arg_kinds


Do we need both is_kw_arg and kw_arg etc. Maybe can make is_kw_ars a property that returns kw_arg is not None? CallableType has already loads of fields.

ilevkivskyi · 2018-06-09T23:40:41Z

mypy/types.py

+        if they are not None.
+
+        If you really want to include star args in the yielded output, set the
+        'include_star_args' field to 'True'."""


set field -> pass argument

ilevkivskyi · 2018-06-09T23:41:14Z

mypy/types.py

@@ -818,6 +837,30 @@ def max_possible_positional_args(self) -> int:
        blacklist = (ARG_NAMED, ARG_NAMED_OPT)
        return len([kind not in blacklist for kind in self.arg_kinds])

+    def formal_arguments(self, include_star_args: bool = False) -> Iterator[FormalArgument]:


It is a good idea to make this a method.

ilevkivskyi · 2018-06-09T23:43:57Z

test-data/unit/check-overloading.test

 @overload
-def foo2(*args: int) -> int: ...  # E: Overloaded function signature 2 will never be matched: function 1's parameter type(s) are the same or broader
+def foo2(*args: int) -> int: ...


I am actually not so sure now we need to make this an error. This is very similar to empty list situation, maybe we can just pick the first overload?

ilevkivskyi · 2018-06-09T23:51:38Z

test-data/unit/check-overloading.test

+@overload
+def foo2(*args: int) -> str: ...  # E: Overloaded function signatures 1 and 2 overlap with incompatible return types
+@overload
+def foo2(*args2: str) -> int: ...


Same here, this is similar to empty list.

Changes made: - Rename 'allow_potential_compatibility' to 'allow_partial_overload' - Refactor 'is_callable_compatible' to be symmetric when 'allow_partial_overload' is set. - Modify is_callable_compatible to no longer consider functions like "f(*int) -> int" and "f(*str) -> int" to be partially overlapping/ potentially compatible.

Michael0x2a · 2018-06-15T01:17:06Z

I originally went with allow_potential_compatibility because I was thinking about this problem in terms of universal vs existential checks, not overlaps. The term "overlap" also, to me, implied that is_callable_compatible would be "symmetric" in a way that it wasn't when the param was set. E.g. the following two functions weren't previously equivalent:
```
 is_callable_compatible(f, g,
                        is_compat=some_symmetric_check,
                        allow_potential_compatibility=True)
 is_callable_compatible(g, f,
                        is_compat=some_symmetric_check,
                        allow_potential_compatibility=True)
```
After some reflection though, I decided it would make the logic more consistent if I just went ahead and added that symmetry.
I previously used the term "compatible" mainly because that was how I renamed the function: I previously refactored out all of the subtyping-related logic so that what it means for something to be "compatible" can be customized by the caller by passing in an appropriate is_compat function. That way, we could reuse the same logic to perform checks unrelated to subtyping.

I don't think I was really aware that the term "compatibility" already meant something specific within mypy. So to avoid confusion, maybe I should rename is_callable_compatible and its params to not use the word "compatible"? I'm not sure what a good alternative name would be though -- let me know what you think.
Regarding the f(*args) thing -- I made the change you suggested.

ilevkivskyi

Great job! I didn't look carefully though all the changes to the code. But all the tests look good, except for one questionable test. If it is not easy to fix, please just go ahead and merge (this is really a corner case).

ilevkivskyi · 2018-06-15T22:36:46Z

mypy/types.py

+        if they are not None.
+
+        If you really want to include star args in the yielded output, set the
+        'include_star_args' field to 'True'."""


set field -> pass argument

It seems you missed this comment. Or you think "set field" is an OK formulation for functions?

ilevkivskyi · 2018-06-15T22:37:17Z

mypy/subtypes.py

-        # Left must have some kind of corresponding argument.
+    # Phase 1: Confirm every argument in R has a corresponding argument in L.
+
+    # Phase 1a: If right and right can both accept an infinite number of args,


"right and right" -> "left and right"

ilevkivskyi · 2018-06-15T22:40:09Z

test-data/unit/check-overloading.test

@@ -1412,15 +1412,19 @@ reveal_type(f(d))  # E: Revealed type is 'builtins.list[builtins.int]'
 from typing import overload, Any

 @overload
-def f(*, x: int = 3, y: int = 3) -> int: ...
+def f(*, x: int = 3, y: int = 3) -> int: ...  # E: Overloaded function signatures 1 and 2 overlap with incompatible return types


Hm, I don't think this this should be an error. The only call that is allowed for both is f().

Michael0x2a · 2018-06-16T00:49:05Z

I could have sworn I fixed the field/parameter thing, but I guess not -- thanks for the (re)-catch!

Regarding the test case: I think I'm going to just merge this as-is. I did a bit of poking and I think it's possible to convert that error case back into a pass, but it'll be a bit kludgy. We basically need to special-case the first match an *arg or **kwarg has so we can pretend that args/kwargs is never length 0. I'm thinking we can add in this kludge later if it turns out it's impossible to type certain things w/ the current semantics.

I think this is also more or less consistent with how we handle non-kwarg/non-arg signatures -- for example, this PR considers the following overload to be unsafely overlapping.

@overload
def foo(*, x: str = '...', y: str = '...') -> str: ...
@overload
def foo(*, x: int = 3, y: int = 3) -> int: ...
def foo(*args): 
    pass

I think it'd maybe be a bit surprising if swapping the last signature for a **kwargs: int were to suddenly make this the check.

ilevkivskyi self-assigned this Jun 7, 2018

ilevkivskyi self-requested a review June 7, 2018 10:09

ilevkivskyi reviewed Jun 9, 2018

View reviewed changes

Michael0x2a added 3 commits June 14, 2018 17:11

Merge branch 'master' into overloads-partially-overlapping-arities

0301792

Fix merge-related error

4db35c1

Michael0x2a mentioned this pull request Jun 15, 2018

Wrong checking of overload for keyword-only args #1907

Closed

ilevkivskyi approved these changes Jun 15, 2018

View reviewed changes

Fix typos

74919ee

Michael0x2a closed this Jun 16, 2018

Michael0x2a reopened this Jun 16, 2018

Michael0x2a merged commit 4fe2220 into python:master Jun 16, 2018

Michael0x2a deleted the overloads-partially-overlapping-arities branch June 16, 2018 00:50

emmatyping mentioned this pull request Jun 30, 2018

overload with tuples of different arity #5302

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add support for detecting overloads with overlapping arities #5163

Add support for detecting overloads with overlapping arities #5163

Michael0x2a commented Jun 6, 2018

ilevkivskyi left a comment

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

ilevkivskyi Jun 9, 2018

Michael0x2a commented Jun 15, 2018

ilevkivskyi left a comment

ilevkivskyi Jun 15, 2018

ilevkivskyi Jun 15, 2018

ilevkivskyi Jun 15, 2018

Michael0x2a commented Jun 16, 2018

Add support for detecting overloads with overlapping arities #5163

Add support for detecting overloads with overlapping arities #5163

Conversation

Michael0x2a commented Jun 6, 2018

ilevkivskyi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Michael0x2a commented Jun 15, 2018

ilevkivskyi left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Michael0x2a commented Jun 16, 2018