Python抽象基类子类方法执行时自动实现耗时统计的方法

直接替换原有Encoder基类代码即可，全程不需要修改任何子类实现，所有继承该基类的编码器会自动获得编解码耗时统计能力。

推荐使用Python 3.6+原生支持的__init_subclass__钩子实现，侵入性最低：

import abc
import time
from functools import wraps
from typing import Tuple

# 项目内已有的DataMapper、MetadataCollector类型保持原有导入逻辑即可
class Encoder(abc.ABC):
    def __init__(self):
        self._encode_time = []
        self._decode_time = []

    def __init_subclass__(cls, **kwargs):
        super().__init_subclass__(**kwargs)
        
        # 自动包装子类实现的encode方法
        original_encode = cls.encode
        @wraps(original_encode)
        def timed_encode(self, decoded_data, **kwargs):
            start = time.perf_counter()
            try:
                return original_encode(self, decoded_data, **kwargs)
            finally:
                self._encode_time.append(time.perf_counter() - start)
        cls.encode = timed_encode

        # 自动包装子类实现的decode方法
        original_decode = cls.decode
        @wraps(original_decode)
        def timed_decode(self, encoded_data, metadata, **kwargs):
            start = time.perf_counter()
            try:
                return original_decode(self, encoded_data, metadata, **kwargs)
            finally:
                self._decode_time.append(time.perf_counter() - start)
        cls.decode = timed_decode

    @abc.abstractmethod
    def encode(self, decoded_data: DataMapper, **kwargs) -> Tuple[DataMapper, MetadataCollector]:
        pass

    @abc.abstractmethod
    def decode(self, encoded_data: DataMapper, metadata: MetadataCollector, **kwargs) -> DataMapper:
        pass

实现说明 #

• 逻辑对业务代码完全透明：子类原有入参、返回值、异常抛出逻辑完全不受影响，不需要做任何适配
• 用try/finally块保证即使编解码过程抛出异常，本次操作的耗时也会被正常记录，不会出现统计漏项
• 用functools.wraps保留原有方法的签名、文档字符串等元信息，不会破坏依赖方法元信息的框架校验逻辑
• 替换了示例中已在Python3.8版本移除的time.clock()，改用官方推荐的time.perf_counter()做耗时统计，精度更高、不受系统时钟调整影响
• 统计结果依然存储在实例的_encode_time、_decode_time列表中，原有读取统计数据的逻辑不需要做任何调整

如果项目运行在Python3.6以下版本（不支持__init_subclass__），可以换用元类实现完全相同的效果：

import abc
import time
from functools import wraps
from typing import Tuple

class EncoderMeta(abc.ABCMeta):
    def __init__(cls, name, bases, namespace):
        super().__init__(name, bases, namespace)
        # 跳过基类本身，只包装具体子类的实现
        if name != 'Encoder' and not cls.__abstractmethods__:
            original_encode = cls.encode
            @wraps(original_encode)
            def timed_encode(self, decoded_data, **kwargs):
                start = time.perf_counter()
                try:
                    return original_encode(self, decoded_data, **kwargs)
                finally:
                    self._encode_time.append(time.perf_counter() - start)
            cls.encode = timed_encode

            original_decode = cls.decode
            @wraps(original_decode)
            def timed_decode(self, encoded_data, metadata, **kwargs):
                start = time.perf_counter()
                try:
                    return original_decode(self, encoded_data, metadata, **kwargs)
                finally:
                    self._decode_time.append(time.perf_counter() - start)
            cls.decode = timed_decode

class Encoder(metaclass=EncoderMeta):
    def __init__(self):
        self._encode_time = []
        self._decode_time = []

    @abc.abstractmethod
    def encode(self, decoded_data: DataMapper, **kwargs) -> Tuple[DataMapper, MetadataCollector]:
        pass

    @abc.abstractmethod
    def decode(self, encoded_data: DataMapper, metadata: MetadataCollector, **kwargs) -> DataMapper:
        pass

内容的提问来源于stack exchange，提问作者graille