如何实现父类方法分派？解决shapely.Point多构造方式问题

DispatchError Traceback (most recent call last)
Cell In [461], line 1
----> 1 Point(6, 4)

File ~\anaconda3\envs\bioinfo\lib\site-packages\multimethod_init_.py:313, in multimethod.call(self, *args, **kwargs)
311 if self.pending: # check first to avoid function call
312 self.evaluate()
--> 313 func = self[tuple(func(arg) for func, arg in zip(self.type_checkers, args))]
314 try:
315 return func(*args, **kwargs)

File ~\anaconda3\envs\bioinfo\lib\site-packages\multimethod_init_.py:307, in multimethod.missing(self, types)
305 return self.setdefault(types, *funcs) # type: ignore
306 msg = f"{self.name}: {len(keys)} methods found" # type: ignore
--> 307 raise DispatchError(msg, types, keys)

DispatchError: ('init: 0 methods found', (<class 'main.Point'>, <class 'int'>), [])

请问除了`multimethod`之外，还有哪些方法可以实现父类方法的分派？

# 可行替代方案

## 1. 手动在`__init__`中做类型判断
直接在自定义`Point`类的`__init__`里区分参数类型，分别处理`AnnotationSet`和父类支持的参数格式：
```python
from shapely.geometry import Point as ShapelyPoint
from dendropy.datamodel.basemodel import AnnotationSet

LONGITUDE_ALIAS = "lon"
LATITUDE_ALIAS = "lat"

class Point(ShapelyPoint):
    def __init__(self, *args, **kwargs):
        if len(args) == 1 and isinstance(args[0], AnnotationSet):
            annotations = args[0]
            try:
                lon = float(annotations.get_value(LONGITUDE_ALIAS))
                lat = float(annotations.get_value(LATITUDE_ALIAS))
                super().__init__(lon, lat)
            except (TypeError, ValueError):
                raise ValueError(f"There is no coordinates in the annotations:\n{annotations}")
        else:
            super().__init__(*args, **kwargs)

该方案简单直接，既支持Point(annotations)，也能正常使用Point(6,4)或父类的其他初始化方式。

2. 使用类工厂方法 #

不修改__init__，新增类方法专门处理AnnotationSet的实例化，完全保留父类的初始化逻辑：

from shapely.geometry import Point as ShapelyPoint
from dendropy.datamodel.basemodel import AnnotationSet

LONGITUDE_ALIAS = "lon"
LATITUDE_ALIAS = "lat"

class Point(ShapelyPoint):
    @classmethod
    def from_annotations(cls, annotations: AnnotationSet):
        try:
            lon = float(annotations.get_value(LONGITUDE_ALIAS))
            lat = float(annotations.get_value(LATITUDE_ALIAS))
            return cls(lon, lat)
        except (TypeError, ValueError):
            raise ValueError(f"There is no coordinates in the annotations:\n{annotations}")

使用时，从AnnotationSet创建实例用Point.from_annotations(annotations)，常规创建用Point(6,4)，符合Python惯用的工厂模式设计，避免重载__init__带来的潜在问题。

3. 利用functools.singledispatchmethod #

Python 3.8+自带的singledispatchmethod支持基于第一个参数的类型分派，针对__init__（第一个参数是self）的特性，可将初始化逻辑转移到辅助方法实现分派：

from functools import singledispatchmethod
from shapely.geometry import Point as ShapelyPoint
from dendropy.datamodel.basemodel import AnnotationSet

LONGITUDE_ALIAS = "lon"
LATITUDE_ALIAS = "lat"

class Point(ShapelyPoint):
    def __init__(self, *args, **kwargs):
        self._init(*args, **kwargs)
    
    @singledispatchmethod
    def _init(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
    
    @_init.register(AnnotationSet)
    def _(self, annotations: AnnotationSet):
        try:
            lon = float(annotations.get_value(LONGITUDE_ALIAS))
            lat = float(annotations.get_value(LATITUDE_ALIAS))
            super().__init__(lon, lat)
        except (TypeError, ValueError):
            raise ValueError(f"There is no coordinates in the annotations:\n{annotations}")

该方案实现了类似多分派的效果，同时兼容父类所有初始化方式。

4. 自定义元类实现分派 #

如果必须用元类方案，可以手动编写元类收集并匹配__init__方法：

import inspect
from shapely.geometry import Point as ShapelyPoint
from dendropy.datamodel.basemodel import AnnotationSet

LONGITUDE_ALIAS = "lon"
LATITUDE_ALIAS = "lat"

class InitDispatchMeta(type):
    def __new__(cls, name, bases, namespace):
        init_methods = []
        # 收集父类的__init__作为兜底
        for base in bases:
            if hasattr(base, '__init__'):
                init_methods.append((None, base.__init__))
        # 收集当前类带类型注解的__init__
        if '__init__' in namespace:
            func = namespace['__init__']
            sig = inspect.signature(func)
            params = list(sig.parameters.values())
            if len(params) > 1:
                param = params[1]
                if param.annotation is not param.empty:
                    init_methods.append((param.annotation, func))
        # 实现分派逻辑
        def dispatch_init(self, *args, **kwargs):
            for annot, method in init_methods:
                if annot is None:
                    try:
                        return method(self, *args, **kwargs)
                    except TypeError:
                        continue
                if len(args) >=1 and isinstance(args[0], annot):
                    return method(self, *args, **kwargs)
            raise TypeError(f"No valid __init__ for arguments: {args}")
        namespace['__init__'] = dispatch_init
        return super().__new__(cls, name, bases, namespace)

class Point(ShapelyPoint, metaclass=InitDispatchMeta):
    def __init__(self, annotations: AnnotationSet):
        try:
            lon = float(annotations.get_value(LONGITUDE_ALIAS))
            lat = float(annotations.get_value(LATITUDE_ALIAS))
            super().__init__(lon, lat)
        except (TypeError, ValueError):
            raise ValueError(f"There is no coordinates in the annotations:\n{annotations}")

该元类优先匹配带类型注解的__init__，失败则调用父类方法，但实现复杂度较高，仅推荐有特殊元类需求时使用。

内容的提问来源于stack exchange，提问作者Vovin