Fixes#4288.
- Default imports to THIRD_PARTY, so in effect we merge the FIRST_PARTY and THIRD_PARTY stubs. This means import order is no longer affected by whether typing_extensions is installed locally.
- Treat typing_extensions, _typeshed and some others as standard library modules.
Note that isort master is very different from the latest release; we'll have to do something
different if and when the next isort release comes out.
* make io classes inherit from typing IO classes
This makes these classes usable if type annotations are given as "IO"
or "TextIO". In the future, we'll then be able to move open() to
return a concrete class instead (#3951).
* open: introduce concrete return types
Fixes#3951.
We use the values of the "mode" and "buffering" arguments to figure out
the concrete type open() will return at runtime. (Compare the CPython
code in https://github.com/python/cpython/blob/master/Modules/_io/_iomodule.c#L231.)
This makes these classes usable if type annotations are given as "IO"
or "TextIO". In the future, we'll then be able to move open() to
return a concrete class instead (#3951).
Fixes#3547
This removes some type safety in exceptional cases, like code that interacts
directly with cached_property objects, but that seems like a price worth
paying.
This is important because mypy doesn't generally think functions are compatible with `FunctionType`, so `inspect.getsource` on arbitrary functions is rejected by the current annotations.
I realized while working on srittau/type-stub-pep#64 that a
few things we do in existing enum definitions in typeshed are
problematic:
- Using "= ..." doesn't allow type checkers to correctly type the
result of Enum.MEMBER.value. In fact, mypy at least infers
.value to be "Ellipsis" if you do this.
- Properties on the enum values themselves, like HTTPStatus.phrase,
should not be specified directly as attributes, because it makes
type checkers think that the properties themselves are enum
members.
I ended up doing a bit more cleanup to the signal module:
- Remove unnecessary ... initializers.
- Remove unnecessary _SIG = Signals alias.
- I don't have Windows to test, but the C code for _signal suggests
that CTRL_C_EVENT and CTRL_BREAK events are not Signals, but just ints:
https://github.com/python/cpython/blob/1dbd084f1f68d7293718b663df675cfbd0c65712/Modules/signalmodule.c#L1575
Previously, constructing a TimeoutExpired directly worked because Exception allowed arbitrary kwargs. We fixed that recently, but now mypy gives an error on creating a TimeoutExpired with legal arguments.
This is needed to make TextIOWrapper and a few classes that inherit
from it concrete in 3.3. I am actually not sure whether this method
exists at runtime; there's nothing in
https://docs.python.org/3/library/io.html#io.TextIOWrapper
that suggests it was added in 3.4. In any case, it hardly matters
since 3.3 usage should be very rare now.
This was already the configuration in Travis and it's been working fine. It's
confusing that Travis was using a different configuration than the default
when you run tests/mypy_test.py yourself.
Started out as progress towards #1476, but I ended up fixing a few more things:
- fixed the signature of _encode_type, which actually returns a pair, not a string
- made some attributes into properties in order to prevent the descriptor protocol from turning them into methods
- found a bug in CPython in the process (python/cpython#6779)
I used the following test file to make sure these classes are now instantiable:
```python
import codecs
import io
from typing import IO
bio = io.BytesIO()
cod = codecs.lookup('utf-8')
codecs.StreamReaderWriter(bio, codecs.StreamReader, codecs.StreamWriter)
codecs.StreamRecoder(bio, cod.encode, cod.decode, codecs.StreamReader, codecs.StreamWriter)
```
Fixes#1850.
The fix was already applied to Python 2, but the typevar-based solution there
leads to "cannot infer value of type variable" in mypy. I used the following
script to check:
```python
from itertools import product
reveal_type(product([1]))
reveal_type(product([1], ['x'], [False], [3.0], [(1,)], [('x',)], [{1}], [{1: 2}], repeat=5))
```
Fixes#1940.
This makes it so that mypy will infer the common base class of these
classes to be Awaitable instead of Iterable. I verified that this
fixes the errors in the script posted by @neilconway.
Fixes#2124.
As mentioned in the issue, the current guidelines lead to extremely long lines of many hundreds of characters, which make code hard to read and review. I think there should be some limit to line length, but it's OK for stubs to have somewhat longer lines than normal code.
For reference, there are currently 115 lines in typeshed over 200 characters; 523 over 120 characters; and 2980 over 79 characters.
We picked 130 because that's the longest line GitHub displays on a 13" laptop in a unified diff without folding. There are currently 382 lines in typeshed that are over 130 characters.
Fixes#1997, #2068.
This is tricky because we need to get the return values right (see #1960 for
prior attempts) and we often run into python/mypy#3644. I found that I
could express most signatures correctly using a series of overloads.
A few other changes in here:
- Added splitunc, which according to https://docs.python.org/3/library/os.path.html
should exist in both Unix and Windows.
- Made the second argument to os.path.curdir Optional to match the implementation.
- Fixed os.path.split, whose previous Path-aware signature triggered python/mypy#3644.
Fixes#1961.
I mostly just replaced all str annotations with Text, including in return types. This is
only broadly correct; diffing a str and a unicode sequence actually results in a mixed
output of str and unicode. We could also keep the return types as str if using Text
causes errors in real code. For callbacks that take str, I introduced a Union alias
because a callable taking a str would not be a compatible with a parameter of type
Callable[[Text], bool].
I also fixed the return type of difflib.restore.
Jelle Zijlstra