sizmailov / pybind11-stubgen

Generate stubs for python modules
Other
218 stars 44 forks source link

When parsing enums derived from `enum.Enum`, non-canonical values are ignored #223

Open bindreams opened 2 months ago

bindreams commented 2 months ago

My package does not use pybind11 hacky enums, but instead derives enums from enum.Enum using a popular helper macro.

However, in that case the generated stub does not contain some values, the ones that are considered non-canonical (as defined in this comment). These are zero values (maybe just in enum.Flag) and repeated/alias values.

I haven't looked at the implementation, but the issue is probably related to this: https://github.com/python/cpython/issues/109633. If I'm correct, the issue can probable be fixed by iterating over the __members__ of the enum class.

Example of what I mean:

>>> list(iter(InputMode))
[<InputMode.UNICODE: 1>,
 <InputMode.GS1: 2>,
 <InputMode.ESCAPE: 8>,
 <InputMode.GS1PARENS: 16>,
 <InputMode.GS1NOCHECK: 32>,
 <InputMode.HEIGHTPERROW: 64>,
 <InputMode.FAST: 128>,
 <InputMode.EXTRA_ESCAPE: 256>]

>>> InputMode.__members__
mappingproxy({'DATA': <InputMode.DATA: 0>,
              'UNICODE': <InputMode.UNICODE: 1>,
              'GS1': <InputMode.GS1: 2>,
              'ESCAPE': <InputMode.ESCAPE: 8>,
              'GS1PARENS': <InputMode.GS1PARENS: 16>,
              'GS1NOCHECK': <InputMode.GS1NOCHECK: 32>,
              'HEIGHTPERROW': <InputMode.HEIGHTPERROW: 64>,
              'FAST': <InputMode.FAST: 128>,
              'EXTRA_ESCAPE': <InputMode.EXTRA_ESCAPE: 256>})

As you can see, with iter(), the DATA value is missing.

sizmailov commented 2 months ago

I don't see how it's relevant to stub generation and this repo.

bindreams commented 2 months ago

I think it's relevant because I'm generating stubs for a pybind11 project and they are incorrect :) I would have used mypy stubgen (which parses enums correctly) but it has multiple other problems.

Do you consider issues related to user-defined type casters out of scope for this project?

sizmailov commented 2 months ago

What is the generated stub, and what is expected?

bindreams commented 2 months ago

Sorry for the back and forth, I should have started with an example.

Here is an example enum in C++

enum class ExampleEnum : int {
    A = 0,
    B = 1,
    C = 1,
    D = 2,
};
Code to generate a pybind11 binding using the helper macro I mentioned above ```c++ // Global scope P11X_DECLARE_ENUM( "ExampleEnum", "enum.Enum", // or "enum.Flag", etc. {"A", ExampleEnum::A}, {"B", ExampleEnum::B}, {"C", ExampleEnum::C}, {"D", ExampleEnum::D} ) PYBIND11_MODULE(module_name, m) { p11x::bind_enums(m); } ```

Here is what pybind11-stubgen generates, when this enum is bound to python's enum.Enum or enum.Flag (or derived classes) respectively:

class ExampleEnum(enum.Enum):
    A: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.A: 0>
    B: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.B: 1>
    D: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.D: 2>
class ExampleEnum(enum.Flag):
    B: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.B: 1>
    D: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.D: 2>

You can see that C is missing in both cases because it has a repeated value, and A is missing in the enum.Flag case becase it's value is 0. However, the real enum you get in the module obviously contains all of the elements, and they are accessible via the __members__ dict:

>>> dir(ExampleEnum)
['B', 'D', '__class__', ...]

>>> list(iter(ExampleEnum))
[<ExampleEnum.B: 1>, <ExampleEnum.D: 2>]

>>> ExampleEnum.__members__
mappingproxy({'A': <ExampleEnum.A: 0>,
              'B': <ExampleEnum.B: 1>,
              'C': <ExampleEnum.B: 1>,
              'D': <ExampleEnum.D: 2>})

For completeness, the desired stub would look like this:

class ExampleEnum(enum.Enum):
    A: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.A: 0>
    B: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.B: 1>
    C: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.C: 1>
    D: typing.ClassVar[ExampleEnum]  # value = <ExampleEnum.D: 2>