Repository navigation
This library should have type annotations #795
Description
Activity
There are stubs available here: https://pypi.org/project/types-babel/
There are stubs available here: https://pypi.org/project/types-babel/
I'm not sure whether that's what you meant to imply, but these stubs pretty much insufficient, as most method signatures in there are untyped.
Question for the maintainers: The type hints would need to be python3.6 compatible, as that's the lowest advertised python version in the
setup.py, correct?@lukasjuhrich Yes, I'd say at least one more version should be compatible with Python 3.6, if possible (even if it is EOL already).
I wonder if there's a tool that could automagically merge those type stubs to our source here...
@akx that wouldn't be really helpful as the aforementioned stubs are just the autogenerated ones in typeshed, see for instance the content of
babel/support.pyi. They're all either untyped or annotated withAny.Well - whatever isn't any-typed, of course.
Monkeytype has proven an useful tool for generating initial typing from test runs, too.
Reacted by Lukas JuhrichWell - whatever isn't any-typed, of course.
Oh, I didn't look far enough. You're right, other modules contain nontrivial hints.
I wonder if there's a tool that could automagically merge those type stubs to our source here...
There is (was?) retype, but I've never used it and it claims to be unsupported as of August 14th.
Okay, so I tried the following on my fork:
pip monkeytype pytest-monkeytypep- Run the tests with
pytest --monkeytype-output=./monkeytype.sqlite3 tests, which took ~1500s and produced a 1.2G sample file monkeytype apply babel.support(see diff) to get an example module I started type hinting merged- Run
darkerto automatically format lines which are overly long due to the diff nicely.
I've noticed the following things which required manual intervention:
- Some imports are janky. Especially the existence of methods named
datetimeordateshadow imports which breaks the type annotations; one has to rewrite them tofrom datetime import datetime as std_datetimeor similar. - Many types are too strict: for instance,
- parameters which are only instantiated as
Noneare typed asNone, which is clearly wrong - parameters like
timeare only instantiated withdatetimes in tests, while they clearly also accepttimes as well
- parameters which are only instantiated as
- The imports are imported in a
from typing import […]-style, and I didn't find a way to configure it differently.
Personally I would recommend aimport typing as tas it strikes a nice balance between- having to touch imports when changing signatures and thus unnecessary diff headache
- having too much visual noise in the signature due to
arg: typing.Optional[typing.Dict[…]]-like constructs where thetypingqualifier just becomes more annoying than anything else
So at the end of the day one has to carefully go through the modules manually (and, of course, at least validate the annotations' relative consistency by running
mypy --check-untyped-defs --follow-imports=silent).
I think especially point 2 is tricky because ideally the type hints of arguments would be as weak as possible (e.g. not demandinglist[T]when you're only iterating over the argument).All of this seems to me that an incremental module-by-module approach with mypy checking every step of the way would be more viable than applying all monkeytype hints at once, then reviewing all of that while probably missing a lot of things, and having a giant PR which takes ages to review and merge.
I think it is important to check type signatures with mypy during CI if a library announces type annotation support via
py.typed.@akx I'd still add it to CI even if you need to have a lot of exceptions defined for it.
(master) $ mypy babel Found 268 errors in 17 files (checked 25 source files)If by "a lot of exceptions" you mean adding about 268
# type: ignorecomments, I don't think that's a good way to go about this.If by "a lot of exceptions" you mean adding about 268 # type: ignore comments, I don't think that's a good way to go about this.
No I meant to ignore some files and rules:
[tool.mypy] exclude = [ 'babel/localtime/_fallback.py', ... ]
https://github.com/pallets-eco/flask-caching/blob/master/setup.cfg#LL83-L96C19
But I guess you're right: With that huge amount of type errors it may be to tedious to add all the exceptions.