Metaclass related internal error

[arnimarj]

Hi,

We just ran into this stacktrace:

https://mypy.rtfd.io/en/latest/common_issues.html#using-a-development-mypy-build
Please report a bug at https://github.com/python/mypy/issues
version: 0.800
Traceback (most recent call last):
  File "mypy/semanal.py", line 4835, in accept
  File "mypy/nodes.py", line 950, in accept
  File "mypy/semanal.py", line 1048, in visit_class_def
  File "mypy/semanal.py", line 1125, in analyze_class
  File "mypy/semanal.py", line 1134, in analyze_class_body_common
  File "mypy/semanal.py", line 1187, in apply_class_plugin_hooks
  File "/home/travis/virtualenv/python3.8.1/lib/python3.8/site-packages/mypy_zope/plugin.py", line 266, in analyze_metaclass
    if info and any(node.fullname == expected for node in info.mro):
AttributeError: 'Var' object has no attribute 'mro'

I haven't had time find exactly which piece of code is causing this, but will do so as soon as I can.

EDIT: this was tested on mypy 0.800. The line which hit this was something like this:

@attr.s(slots=True, frozen=True)
class Leg(metaclass=CustomMetaclass):
	a: str = attr.ib()
	b: str = attr.ib()
	c: str = attr.ib()

[arnimarj]

I modified get_metaclass_hook to be:

    def get_metaclass_hook(
        self, fullname: str
    ) -> Optional[Callable[[ClassDefContext], None]]:
        def analyze_metaclass(ctx: ClassDefContext) -> None:
            metaclass = cast(NameExpr, ctx.cls.metaclass)

            if isinstance(metaclass.node, Var):
                return

            info = cast(TypeInfo, metaclass.node)
            expected = "zope.interface.interface.InterfaceClass"
            if info and any(node.fullname == expected for node in info.mro):
                self.log(f"Found zope interface: {ctx.cls.fullname}")
                md = self._get_metadata(ctx.cls.info)
                md["is_interface"] = True

        return analyze_metaclass

...after looking at analyze_metaclass in MyPy. There is a short-circuit along the lines of:

            if isinstance(sym.node, Var) and isinstance(get_proper_type(sym.node.type), AnyType):
                # 'Any' metaclass -- just ignore it.
                #
                # TODO: A better approach would be to record this information
                #       and assume that the type object supports arbitrary
                #       attributes, similar to an 'Any' base class.
                return

That made the error go away

[kedder]

Yeah, these casts in front of analyze_metaclass are not right. I'll try a more correct approach there, let me make a branch to test it.

[arnimarj]

I've got a small test-case:

from i.dont.exist import MetaKlass


class Leg(metaclass=MetaKlass):
	pass

I believe mypy gives you a Var node when the metaclass has not been resolved

[kedder]

Oh, thanks, I can add a test too then!

[arnimarj]

my mypy.ini:

[mypy]
scripts_are_modules = True
show_traceback = True
disallow_any_generics = True
warn_no_return = True
no_implicit_optional = True
ignore_missing_imports = True
namespace_packages = True
plugins = mypy_zope:plugin

And the run:

$ PYTHONPATH=$(pwd) mypy --show-traceback --show-error-codes --config mypy.ini z.py 
z.py:4: error: INTERNAL ERROR -- Please try using mypy master on Github:
https://mypy.rtfd.io/en/latest/common_issues.html#using-a-development-mypy-build
Please report a bug at https://github.com/python/mypy/issues
version: 0.800
Traceback (most recent call last):
  File "mypy/semanal.py", line 4835, in accept
  File "mypy/nodes.py", line 950, in accept
  File "mypy/semanal.py", line 1048, in visit_class_def
  File "mypy/semanal.py", line 1125, in analyze_class
  File "mypy/semanal.py", line 1134, in analyze_class_body_common
  File "mypy/semanal.py", line 1187, in apply_class_plugin_hooks
  File "/home/arni/.config/dohop/tools_venv3/lib/python3.8/site-packages/mypy_zope/plugin.py", line 270, in analyze_metaclass
    if info and any(node.fullname == expected for node in info.mro):
AttributeError: 'Var' object has no attribute 'mro'
z.py:4: : note: use --pdb to drop into pdb

[kedder]

@arnimarj, the fix is in #29. I think it should solve this issue.

[arnimarj]

@arnimarj, the fix is in #29. I think it should solve this issue.

The issue seems to be gone. Thanks for the great turnaround speed.

[kedder]

Do you want a new release with this fix?

[arnimarj]

Do you want a new release with this fix?

Yes please :)

[kedder]

0.2.10 is released

[arnimarj]

0.2.10 is released

Thanks. We hit this bug in our regression tests, so a new release helps.