Add --allow-untyped-globals flag that avoids partial type errors
#5670
Conversation
A major cause of false positives when checking new codebases is unannotated top level containers and Nones. Add a flag that suppresses both the "needs type annotation" errors and the assigning of NoneTyp to variables at the "toplevel" (not when nested in functions). To avoid throwing away too much type information, we give more precise types in these situations than previously, filling in the arguments to the type with `Any`. Closes #5639.
msullivan
force-pushed
the
lenient-toplevel
branch
from
7c23ac4 to
c352f83
Compare
| reveal_type(x) # E: Revealed type is 'builtins.list[Any]' | ||
| reveal_type(y) # E: Revealed type is 'builtins.dict[Any, Any]' | ||
|
|
||
| [builtins fixtures/dict.pyi] |
There was a problem hiding this comment.
I'm curious why this doesn't pass:
class C:
a = [] # Error here
def f(self) -> None:
reveal_type(self.a)This still gives an error "Need type annotation for 'a'" on line 2.
There was a problem hiding this comment.
Because member checking has a totally different partial type path apparently, which I'm now going to merge.
|
Just a few thoughts about how we're categorizing this flag:
|
|
Re categorization: |
|
I still think it belongs better in strictness_group. This looks as follows currently: |
gvanrossum
changed the title
--permissive-toplevel flag that avoids partial type errors --allow-untyped-globals flag that avoids partial type errors
|
(Note that the flag ended up being renamed to |
A major cause of false positives when checking new codebases is
unannotated top level containers and Nones.
Add a flag that suppresses both the "needs type annotation" errors and
the assigning of NoneTyp to variables at the "toplevel" (not when
nested in functions). To avoid throwing away too much type
information, we give more precise types in these situations than
previously, filling in the arguments to the type with
Any.Closes #5639.
I don't love the flag name and would be happy to change it. This just adds an individual flag. Grouping into a
--strictness=0configuration or something we should spin off as another bug, I think.