首先我们定义一个可以打印日志的装饰器：

def log(func):
    def wrapper(*args, **kwargs):
        print('call %s():' % func.__name__)
        return func(*args, **kw)
    return wrapper

它接受一个函数作为输入，再返回一个函数。我们使用一下这个装饰器

@log
def now():
    print('2023-8-31')

我们调用一下now函数，它不仅打印当前时间，还会在前面打印一行日志：

把@log放到now()函数定义处，相当于执行下列语句：

now = log(now)

假如我们的装饰器需要传递参数，那么我们需要再把装饰器包起来，俗称套娃

def log(text):
    def decorator(func):
        def wrapper(*args, **kwargs):
            print('%s %s():' % (text, func.__name__))
            return func(*args, **kwargs)
        return wrapper
    return decorator

我们用一下这个装饰器：

@log('装饰器参数')
def now():
    print('2023-8-31')

执行

上面装饰器等同于

now = log('装饰器参数')(now)

log('装饰器参数')返回函数decorator，decorator(now)返回函数wrapper

附录

我在看 lite transformer 的代码的时候发现了一个装饰器：

def register_model(name):
    """
    New model types can be added to fairseq with the :func:`register_model`
    function decorator.

    For example::

        @register_model('lstm')
        class LSTM(FairseqEncoderDecoderModel):
            (...)

    .. note:: All models must implement the :class:`BaseFairseqModel` interface.
        Typically you will extend :class:`FairseqEncoderDecoderModel` for
        sequence-to-sequence tasks or :class:`FairseqLanguageModel` for
        language modeling tasks.

    Args:
        name (str): the name of the model
    """

    def register_model_cls(cls):
    	#如果函数名字登记过了，报错
        if name in MODEL_REGISTRY:
            raise ValueError('Cannot register duplicate model ({})'.format(name))
        #如果函数不是BaseFairseqModel的子类，报错
        if not issubclass(cls, BaseFairseqModel):
            raise ValueError('Model ({}: {}) must extend BaseFairseqModel'.format(name, cls.__name__))
        #登记一下新函数的名字
        MODEL_REGISTRY[name] = cls
        return cls

    return register_model_cls

可以看到它只有两层，register_model(name)对标上面的log(text)，register_model_cls对标上面的decorator，它把一个函数cls传进来，登记一下，再把函数cls传出去，没有wrapper。看一下它是怎么调用的：

@register_model('transformer_multibranch_v2')
class TransformerMultibranchModel(FairseqEncoderDecoderModel):
    """
    Transformer model from `"Attention Is All You Need" (Vaswani, et al, 2017)
    <https://arxiv.org/abs/1706.03762>`_.

    Args:
        encoder (TransformerEncoder): the encoder
        decoder (TransformerDecoder): the decoder

    The Transformer model provides the following named architectures and
    command-line arguments:

    .. argparse::
        :ref: fairseq.models.transformer_parser
        :prog:
    """

在这里，他把新定义的函数TransformerMultibranchModel作为参数cls传进去登记，给它取名name为transformer_multibranch_v2