Support for no_type_check decorator

[spkersten]

Implements #557.

I've renamed the no_type_check decorator to no_type_checks.

Currently the typexport-basic tests fails because the no_type_check decorator I added to the typing.py stub shows up in the output of all tests.

[gvanrossum]

Why the rename? AFAIK the e PEP had the he singular.
On Apr 19, 2015 11:14 AM, "Sander Kersten" [email protected] wrote:

Implements #557 #557.

I've renamed the no_type_check decorator to no_type_checks.

Currently the typexport-basic tests fail for a weird reason.

You can view, comment on, or merge this pull request online at:

#645
Commit Summary

• Renamed no_type_check decorator to no_type_checks.
• Parse annotations of function decorated with 'no_type_checks' as
  regular expression instead of type annotations.
• Don't type check functions decorated with 'no_type_checks'

File Changes

• M lib-typing/3.2/typing.py
  https://github.com/JukkaL/mypy/pull/645/files#diff-0 (4)
• M mypy/checker.py
  https://github.com/JukkaL/mypy/pull/645/files#diff-1 (6)
• M mypy/parse.py
  https://github.com/JukkaL/mypy/pull/645/files#diff-2 (49)
• M mypy/test/data/check-functions.test
  https://github.com/JukkaL/mypy/pull/645/files#diff-3 (10)
• M mypy/test/data/lib-stub/typing.py
  https://github.com/JukkaL/mypy/pull/645/files#diff-4 (3)

Patch Links:

• https://github.com/JukkaL/mypy/pull/645.patch
• https://github.com/JukkaL/mypy/pull/645.diff

—
Reply to this email directly or view it on GitHub
#645.

[spkersten]

@gvanrossum Both [https://github.com/python/typing/issues/26](your comment) and #557 have the plural form. But I'll change it back to what the PEP say.

[JukkaL]

Thanks for the PR! I've been busy recently, but I had time for a quick a pass.

The implementation has a small limitation in that it detects the decorator based on the name. Something like this would not be detected:

from typing import no_type_check as my_annotations

@my_annotations
def foo(x: {1: 'x'}) -> 3: ...

I think this would also not work, which is may be a more realistic use case:

import typing

@typing.no_type_check
def foo(...) -> 2: ... 

To implement these properly, we could accept arbitrary expressions in annotations during parsing, and only translate them to types during semantic analysis (but only if there is no no_type_check_ decorator). Not sure if this is important enough to implement now, but it would be more consistent with how names are bound elsewhere.

[gvanrossum]

I think the 'as' rename is uninteresting, but the @typing.no_type_check
form should be supported.

On Tuesday, April 28, 2015, Jukka Lehtosalo [email protected]
wrote:

Thanks for the PR! I've been busy recently, but I had time for a quick a
pass.

The implementation has a small limitation in that it detects the decorator
based on the name. Something like this would not be detected:

from typing import no_type_check as my_annotations
@my_annotationsdef foo(x: {1: 'x'}) -> 3: ...

I think this would also not work, which is may be a more realistic use
case:

import typing
@typing.no_type_checkdef foo(...) -> 2: ...

To implement these properly, we could accept arbitrary expressions in
annotations during parsing, and only translate them to types during
semantic analysis (but only if there is no no_type_check_ decorator). Not
sure if this is important enough to implement now, but it would be more
consistent with how names are bound elsewhere.

—
Reply to this email directly or view it on GitHub
#645 (comment).

--Guido van Rossum (on iPad)

[spkersten]

This is what I got stuck on; I wasn't happy with the string comparison on 'no_type_check'. Parsing types only after semantic analysis seemed the best solution, but is a lot more work.

If it's okay with you, I'll keep the string comparison for now (with both 'no_type_check' and 'typing.no_type_check'), as this feature is quite useful given the existing bugs/limitations in mypy. Separate of this, I'll try to figure out how to move the parsing of type to the semantic analysis phases.

On 29 apr. 2015, at 07:06, Jukka Lehtosalo [email protected] wrote:

Thanks for the PR! I've been busy recently, but I had time for a quick a pass.

The implementation has a small limitation in that it detects the decorator based on the name. Something like this would not be detected:

from typing import no_type_check as my_annotations

@my_annotations
def foo(x: {1: 'x'}) -> 3: ...
I think this would also not work, which is may be a more realistic use case:

import typing

@typing.no_type_check
def foo(...) -> 2: ...
To implement these properly, we could accept arbitrary expressions in annotations during parsing, and only translate them to types during semantic analysis (but only if there is no no_type_check_ decorator). Not sure if this is important enough to implement now, but it would be more consistent with how names are bound elsewhere.

—
Reply to this email directly or view it on GitHub.

[spkersten]

@JukkaL The no_type_check decorator I added to the typing.py stub shows up in the output of many typexport-basic tests. The comments intesttypegen.py suggest that it is not easy to suppress that from the output; and the things I tried were unsuccessful. Do you have any ideas how to fix this?

[JukkaL]

Here's a diff that fixes the test failures:

diff --git a/mypy/test/data/lib-stub/typing.py b/mypy/test/data/lib-stub/typing.py
index c96fa78..35d9977 100644
--- a/mypy/test/data/lib-stub/typing.py
+++ b/mypy/test/data/lib-stub/typing.py
@@ -16,6 +16,7 @@ Callable = object()
 builtinclass = object()
 _promote = object()
 NamedTuple = object()
+no_type_check = object()

 # Type aliases.
 List = object()
@@ -35,6 +36,3 @@ class Iterator(Iterable[T], Generic[T]):
 class Sequence(Generic[T]):
     @abstractmethod
     def __getitem__(self, n: Any) -> T: pass
-
-def no_type_check(func):
-    return func
\ No newline at end of file

The current convention is to initialize all 'special' names in typing.py stubs with object(). The initializer is actually arbitrary and it might be slightly better to use something like ... (literal ellipsis).

[spkersten]

Thanks for the explanation of how the stub works. I've applied your diff and fixed the handling of the @typing.no_type_check case.

[JukkaL]

Looks good now! Thanks again for the PR!