MyPy does not consider classes instances of their metaclass

[wittekm]

Playing around with a toy situation as below:

class AsdfMeta(type): pass
class AsdfConcrete1(object):
  __metaclass__ = AsdfMeta

def accepts_an_asdf_class(klass): # type: (AsdfMeta) -> None
  pass

accepts_an_asdf_class(AsdfConcrete1)
accepts_an_asdf_class(AsdfConcrete1())
accepts_an_asdf_class("hello, i am a string and not a class with metaclass AsdfMeta")

I get back

56: error: Argument 1 to "accepts_an_asdf_class" has incompatible type "AsdfConcrete1"; expected "AsdfMeta"
57: error: Argument 1 to "accepts_an_asdf_class" has incompatible type "AsdfConcrete1"; expected "AsdfMeta"
58: error: Argument 1 to "accepts_an_asdf_class" has incompatible type "unicode"; expected "AsdfMeta"

(This is super low-priority in my opinion.)

[Diggsey]

This is problematic with Enums, because enums have the __members__ property on their metaclass. Annoyingly it can't be faked by adding a __members__ field to the enum class itself, because it gets retyped by mypy as though it were a variant of the enum.

[gvanrossum]

Another curious case where this is problematic: ctypes defines classes (e.g. c_char) that can be multiplied (e.g. c_char*16 is an array of 16 bytes).

[ilevkivskyi]

This issue seems to be fixed by #2475 (but only with python 3 syntax, metaclass=AsdfMeta). The first function call passes. two other fail.

[JukkaL]

Updated priority since it looks like this should be easier to fix now that #2475 has landed.

[gvanrossum]

Looks like this is now fixed (in master). I get the last two errors (which seem expected) but this line is now error-free in the example:

accepts_an_asdf_class(AsdfConcrete1)

[wittekm]

Cool! Yes, the latter two are counterexamples.

…

On Tue, Feb 7, 2017, 9:57 AM Guido van Rossum ***@***.***> wrote: Looks like this is now fixed (in master). I get the last two errors (which seem expected) but this line is now error-free in the example: accepts_an_asdf_class(AsdfConcrete1) — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1267 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAjT4s7Z2hZmXQLbMkdIMeuyMtXlf8q-ks5raLCagaJpZM4HoAV1> .