Make HTTPError instantiable. #1379

carljm · 2017-06-01T06:22:59Z

Demonstration of fix (../typeshed is a clone of typeshed with this branch checked out):

$ cat httperror.py 
from io import StringIO
from urllib.error import HTTPError

http_error = HTTPError("", 500, "message", dict(), StringIO(''))

$ mypy httperror.py 
httperror.py:4: error: Cannot instantiate abstract class 'HTTPError' with abstract attributes '__enter__', '__exit__', ... and 'writelines' (16 methods suppressed)

$ mypy --custom-typeshed-dir ../typeshed/ httperror.py

…O in typing.pyi.

JelleZijlstra · 2017-06-01T06:39:22Z

stdlib/3/typing.pyi

+    def __enter__(self) -> BinaryIO: ...
+    def __exit__(self, t: Optional[Type[BaseException]], value: Optional[BaseException],
+                 # TODO: traceback should be TracebackType but that's defined in types
+                 traceback: Optional[Any]) -> bool: ...


TracebackType is already imported, so it should be fine to use.

carljm · 2017-06-01T06:45:15Z

Good point! Just copied that TODO from BinaryIO. Updated both locations to just use TracebackType.

gvanrossum · 2017-06-01T23:45:46Z

stdlib/3/typing.pyi

@@ -404,6 +404,41 @@ class BinaryIO(IO[bytes]):
    @abstractmethod
    def __enter__(self) -> BinaryIO: ...

+class _ConcreteBinaryIO(BinaryIO):


I don't think we should do this. Instead we should make some changes in io.pyi and inherit from io.BufferedRandom when we need a concrete subclass of BinaryIO.

Ok, I'll see what I can put together that works.

carljm · 2017-06-02T16:41:51Z

So it turns out that just having addinfourl inherit BufferedRandom instead of BinaryIO fixes the immediate issue, without any other changes to io.pyi. Is that a workable fix? I looked briefly at the other TODOs in io.pyi, but it seemed like some significant refactoring would be needed to address them and avoid lots of "definition incompatible with base class" errors; doesn't seem like that refactoring needs to be conflated with this issue.

gvanrossum · 2017-06-02T16:55:30Z

So if addinfourl() inherits from io.BufferedIO, it's probably not compatible with something that requires a BinaryIO or IO[bytes]. I'm not sure if that's important?

carljm · 2017-06-03T03:19:13Z

Ugh. Yeah, that probably matters.

It also works to subclass BytesIO (which inherits BinaryIO already) instead of BinaryIO. But that gives it some additional methods from BytesIO that it probably shouldn't have.

I'll see what I can figure out with the io.pyi refactor.

gvanrossum · 2017-06-03T03:49:56Z

Thanks, given that you only encountered it once it's not a high priority to get it fixed, so then I'd rather do it right or not at all (you can keep the # type: ignore in the meantime :-).

JelleZijlstra · 2017-06-22T03:07:53Z

@carljm any updates?

gvanrossum · 2017-06-22T03:17:47Z

This should be easier with the changes to io.pyi that just landed.

…

On Jun 21, 2017 8:07 PM, "Jelle Zijlstra" ***@***.***> wrote: @carljm <https://github.com/carljm> any updates? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1379 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ACwrMtCceIsdGvOnjWFVrqLBxaLXU0K2ks5sGdqKgaJpZM4NsijB> .

carljm · 2017-06-22T05:52:55Z

I played with this for a bit but didn't get very far. If some of the io.pyi cleanup has been done for me, I'll take another look (probably next week).

gvanrossum · 2017-06-23T01:13:27Z

Great. Check out the changes that were made in #1395

JelleZijlstra · 2017-08-09T15:38:09Z

This was just reported again, so perhaps we should try to apply a quick fix to make HTTPError instantiable. Perhaps we can just drop the IO base class?

carljm · 2017-08-09T17:12:53Z

Yeah, sorry... I took a look at this, realized that the io.pyi cleanup had only been done for Py2 and I care about Py3, so would have to figure out how to correctly reapply that cleanup for Py3, and then ran out of time.

gvanrossum · 2018-03-14T04:43:12Z

Ping? We should probably just close this if nobody's going to get to it.

carljm · 2018-03-14T16:59:07Z

I'm not likely to get back to this soon, sorry. It's an unfortunate wart, but it doesn't crop up often (only one instance in our entire codebase AFAIK) and it's easy enough to # type: ignore.

gvanrossum · 2018-03-14T19:11:46Z

OK, given the amount of back and forth this has already led to, I'm closing it.

Make HTTPError instantiable by introducing fictional _ConcreteBinaryI…

94592d4

…O in typing.pyi.

JelleZijlstra reviewed Jun 1, 2017

View reviewed changes

Fix obsolete TODO regarding TracebackType.

14a0c0e

gvanrossum reviewed Jun 1, 2017

View reviewed changes

Use BufferedRandom instead of _ConcreteBinaryIO.

9848b6b

carljm changed the title ~~Make HTTPError instantiable by introducing fictional _ConcreteBinaryIO in typing.pyi.~~ Make HTTPError instantiable by subclassing BufferedRandom instead of BinaryIO. Jun 2, 2017

carljm changed the title ~~Make HTTPError instantiable by subclassing BufferedRandom instead of BinaryIO.~~ Make HTTPError instantiable. Jun 3, 2017

JelleZijlstra mentioned this pull request Jul 11, 2017

Accidentally abstract classes #1476

Closed

25 tasks

gvanrossum closed this Mar 14, 2018

Uh oh!

Make HTTPError instantiable. #1379

Make HTTPError instantiable. #1379

Uh oh!

Conversation

carljm commented Jun 1, 2017

Uh oh!

JelleZijlstra Jun 1, 2017

Choose a reason for hiding this comment

Uh oh!

carljm commented Jun 1, 2017

Uh oh!

gvanrossum Jun 1, 2017

Choose a reason for hiding this comment

Uh oh!

carljm Jun 2, 2017

Choose a reason for hiding this comment

Uh oh!

carljm commented Jun 2, 2017

Uh oh!

gvanrossum commented Jun 2, 2017

Uh oh!

carljm commented Jun 3, 2017

Uh oh!

gvanrossum commented Jun 3, 2017

Uh oh!

JelleZijlstra commented Jun 22, 2017

Uh oh!

gvanrossum commented Jun 22, 2017 via email

Uh oh!

carljm commented Jun 22, 2017

Uh oh!

gvanrossum commented Jun 23, 2017 via email

Uh oh!

JelleZijlstra commented Aug 9, 2017

Uh oh!

carljm commented Aug 9, 2017

Uh oh!

gvanrossum commented Mar 14, 2018

Uh oh!

carljm commented Mar 14, 2018

Uh oh!

gvanrossum commented Mar 14, 2018

Uh oh!

Uh oh!