When we import typeshed internally at Dropbox, somehow the fact that
these files are all stubs gets lost (it's a long story...). This
causes errors like this:
.../stdlib/2/ast.pyi:6: error: Name 'typing' already defined (by an import)
The quickest way around this is to rename the import to _typing.
I found some code that called DEFINE_integer() and DEFINE_float() with default set to None. Perusing the code that seems to be a totally valid usage. I am not super familiar with gflags but I believe this PR consistently changes all those default values to Optional[].
We (I) introduced the BLOCKED label ~2 years ago, but it never got much usage, and I remember it startled some developers (Ivan may recall...). A better convention seems to put [WIP] in the subject, so let's standardize on that and get rid of the BLOCKED label.
I already removed the BLOCKED label and edited the subject of the one issue under it.
We also don't make good use of the other labels, but they seem more useful and less controversial.
There were two problems AFAICT:
- There are multiple Loader classes without a common base class,
so this should really be a union or a protocol.
- The argument is supposed to be a class, not an instance.
Rather than figuring out a fix (which would probably require some
nasty refactoring) I'm just going back to the old situation where
these arguments are implicitly annotated with `Any`.
See also https://github.com/python/typeshed/pull/1657#pullrequestreview-72049880
* Add flags to pass on --warn-unused-ignores and --no-implicit-optional to mypy
* Make implicit Optional explicit in arg types (2and3 part)
* Convert {stdlib,third_party}/2 to explicit Optional
* DictReader should not be abstract. Reformat long lines.
* Make restval optional for DictWriter.__init__.
* The arg to reader() is *Iterable*, not *Iterator*.
* Improve signature of float() (use Union instead of overload).
The previous change was still right. Unique among `poplib` functions, `uidl()` returns a long response when called without arguments, but just a bytes string when called with a `which` argument.
* Change Uses of List[] to Sequence[] in argument positions.
One of these (parse_args()) broke in real-world code, because the
expected type was List[Union[str, bytes]] but the actual type was
List[str] (because List is invariant). That is ridiculous, so I
changed the accepted type to Sequence[_Text] which is covariant. Then
I found a few other methods/functions that probably should also be
changed to Sequence, but I'm less certain of those. I'm not at all
sure about the instance/class attributes, so I left those alone
(though I suspect those might also have to switch).
* Fixes suggested by code review
- Changed Sequence[Option] to Iterable[Option] everywhere
- Changed _process_args to plain List
- Switch requirements-tests-py3.txt to use mypy GitHub master.
- Exclude .venv{2,3} (suggested in README.md) in .flake8 and .gitignore.
- Update instructions in README.md to be working.
- Remove runtests.sh script and mentions of it -- I can't get it to work
and the three steps are easily run by hand.
The latest mypy complained that the signature didn't match that in the
superclass: Serializer.load_payload() has a serializer argument. I
don't know this project but I think it's best to just add that
argument (rather than adding `# type: ignore`).
* Make Mapping covariant.
Fixes#510.
This requires a `# type: ignore` on `__getitem__` and `get` because
mypy complains about covariant parameters.
FWIW typing.py needs to be changed too. (It was covariant in the
value but invariant in the key -- typeshed was invariant in both.)
* Delete outdated comment.
* Backpeddle a bit -- Mapping key type should not be covariant.
* Add stub for cgi.parse_header().
* Improve asyncio stubs (far from complete)
* More asyncio changes.
* Use @overload to solve strange test failures.
* Add some TODOs. Make ProactorEventLoop conditional.
* Future should not inherit from Awaitable or implement __await__.
At least not yet.
* Fix AbstractServer.wait_closed() return type.
It's a generator, not a future.
* Fix stubs for 2.7/dateutil. They were pretty broken.
(The 3/dateutil share some of the brokenness but that's still a TODO.)
* Fix argparse stubs.
- Container is not strong enough for choices.
- add_subparsers() returns something with an add_parser() method.
- Move Loader and ModuleType into _importlib_modulespec.pyi.
- Add "import X as X" for these to types.pyi and importlib/abc.pyi.
The goal is to ensure mypy -i still works, to fix https://github.com/python/mypy/issues/1797.
* Almost all re functions take a compiled pattern. (Even re.compile()!)
Fixes#188
Note: I'm using AnyStr so that the type of string used for pattern and
for the rest of the arguments must match. This is not 100% correct,
since Python 2 sometimes allows mixed types. But sometimes it
doesn't, depending on the values (e.g. non-ASCII bytes), and Python 3
always insists on matching, so I think this is actually a good idea.
* Same treatment for stdlib/3/re.pyi.
Guido van Rossum