Allow subscripted aliases at function scope #4000

ilevkivskyi · 2017-09-24T23:56:44Z

This PR allows subscripted type aliases at function scope, as discussed in #3145 this will simplify the rules and have other pluses (apart from fixing the actual issue).

In addition, I prohibit reuse of bound type variables to define generic aliases, situations like this were always ambiguous:

T = TypeVar('T')
class C(Generic[T]):
    A = List[T]
    x: A
    reveal_type(x)  # on one hand one might expect this to be List[T],
                    # but on the other hand, non-subscripted generic aliases
                    # get variables replaced with 'Any', so this should be 'List[Any]'.

Also prohibiting this use is consistent with how we prohibit bound type variable reuse for nested classes, one can always just use another type variable, so this works:

T = TypeVar('T')
S = TypeVar('S')
class C(Generic[T]):
    A = List[S]
    x: A[T]

…euse

…ases

…i/mypy into fix-qualified-aliases

JukkaL

Thanks for the PR! Looks pretty good, but I was confused in a few places.

JukkaL · 2017-10-09T16:34:50Z

mypy/semanal.py

        """Check if 'rvalue' represents a valid type allowed for aliasing
        (e.g. not a type variable). If yes, return the corresponding type and a list of
        qualified type variable names for generic aliases.
-        If 'allow_unnormalized' is True, allow types like builtins.list[T].


Does this PR also affect whether list[x] is accepted in stubs?

No, I just noticed that this parameter is unused in the function body when I added the new one.

JukkaL · 2017-10-09T16:35:05Z

mypy/checkexpr.py

@@ -650,7 +650,7 @@ def analyze_type_type_callee(self, item: Type, context: Context) -> Type:
            res = type_object_type(item.type, self.named_type)
            if isinstance(res, CallableType):
                res = res.copy_modified(from_type_type=True)
-            return res
+            return expand_type_by_instance(res, item)


What's this change?

This is not directly related to this PR, but this was exposed by a test I updated. For example:

x: Type[C[int]] x() # this should be 'C[int]', not 'C[<nothing>]' y: Type[List] y() # this should be 'List[Any]', not 'List[<nothing>]'

By the way, this use case is not mentioned in PEP 484, but there are tests for this, for example testClassValuedAttributesGeneric. There are of course subtleties related to Type[...], but I think this change doesn't add unsafety, but removes some false positives.

JukkaL · 2017-10-09T16:42:53Z

mypy/semanal.py

-        if (len(s.lvalues) == 1 and not self.is_func_scope() and
-                not (self.type and isinstance(s.rvalue, NameExpr) and lvalue.is_def)
+        if (len(s.lvalues) == 1 and
+                not ((self.type or self.is_func_scope()) and


This condition is hard to understand. Add a comment describing what it does at a high level and maybe also give some concrete examples. It's also worth trying to simplify the nested boolean expression. Ideas:

Make the top-level expression of form e1 and (...) and not e2 instead of e1 and not (...) and not e2 -- the first not operator makes this a bit harder to understand, I think.

Use a temp variable to hold the result of self.type or self.is_func_scope() and give it a semantically meaningful name.

OK, fixed. I replaced this with three ifs and thus saved one nesting level for the rest of the body.

ilevkivskyi

@JukkaL Thanks for review!

ilevkivskyi · 2017-10-09T16:57:26Z

mypy/checkexpr.py

@@ -650,7 +650,7 @@ def analyze_type_type_callee(self, item: Type, context: Context) -> Type:
            res = type_object_type(item.type, self.named_type)
            if isinstance(res, CallableType):
                res = res.copy_modified(from_type_type=True)
-            return res
+            return expand_type_by_instance(res, item)


This is not directly related to this PR, but this was exposed by a test I updated. For example:

x: Type[C[int]] x() # this should be 'C[int]', not 'C[<nothing>]' y: Type[List] y() # this should be 'List[Any]', not 'List[<nothing>]'

By the way, this use case is not mentioned in PEP 484, but there are tests for this, for example testClassValuedAttributesGeneric. There are of course subtleties related to Type[...], but I think this change doesn't add unsafety, but removes some false positives.

ilevkivskyi · 2017-10-09T17:12:54Z

mypy/semanal.py

        """Check if 'rvalue' represents a valid type allowed for aliasing
        (e.g. not a type variable). If yes, return the corresponding type and a list of
        qualified type variable names for generic aliases.
-        If 'allow_unnormalized' is True, allow types like builtins.list[T].


No, I just noticed that this parameter is unused in the function body when I added the new one.

ilevkivskyi · 2017-10-09T17:24:41Z

mypy/semanal.py

-        if (len(s.lvalues) == 1 and not self.is_func_scope() and
-                not (self.type and isinstance(s.rvalue, NameExpr) and lvalue.is_def)
+        if (len(s.lvalues) == 1 and
+                not ((self.type or self.is_func_scope()) and


OK, fixed. I replaced this with three ifs and thus saved one nesting level for the rest of the body.

JukkaL

Thanks for the updates! The code is now much easier to read.

ilevkivskyi and others added 6 commits September 25, 2017 01:37

Allow subscripted aliases at function scope; prohibit bound typevar r…

4c9b023

…euse

Improve one test

a464a53

Merge remote-tracking branch 'upstream/master' into fix-qualified-ali…

e4c1ef4

…ases

Merge branch 'master' into fix-qualified-aliases

38ecc68

Merge branch 'fix-qualified-aliases' of https://github.com/ilevkivsky…

178826c

…i/mypy into fix-qualified-aliases

Update tests after merge

358ca96

ilevkivskyi mentioned this pull request Oct 9, 2017

Release 0.540 planning #4084

Closed

JukkaL self-assigned this Oct 9, 2017

JukkaL reviewed Oct 9, 2017

View reviewed changes

Address CR

5cbe056

ilevkivskyi commented Oct 9, 2017

View reviewed changes

JukkaL approved these changes Oct 10, 2017

View reviewed changes

JukkaL merged commit 3e49ef9 into python:master Oct 10, 2017

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Allow subscripted aliases at function scope #4000

Allow subscripted aliases at function scope #4000

ilevkivskyi commented Sep 24, 2017

JukkaL left a comment

JukkaL Oct 9, 2017

ilevkivskyi Oct 9, 2017

JukkaL Oct 9, 2017

ilevkivskyi Oct 9, 2017

JukkaL Oct 9, 2017

ilevkivskyi Oct 9, 2017

ilevkivskyi left a comment

ilevkivskyi Oct 9, 2017

ilevkivskyi Oct 9, 2017

ilevkivskyi Oct 9, 2017

JukkaL left a comment

Allow subscripted aliases at function scope #4000

Allow subscripted aliases at function scope #4000

Conversation

ilevkivskyi commented Sep 24, 2017

JukkaL 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

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

JukkaL left a comment

Choose a reason for hiding this comment