引发异常与返回函数中无异常？

在 Python 中用户定义函数的更好做法是: raise是异常还是 return None？例如，我有一个函数可以查找文件夹中最新的文件。

def latestpdf(folder):
# list the files and sort them
try:
latest = files[-1]
except IndexError:
# Folder is empty.
return None  # One possibility
raise FileNotFoundError()  # Alternative
else:
return somefunc(latest)  # In my case, somefunc parses the filename

另一种选择是保留异常并在调用方代码中处理它，但我认为处理 FileNotFoundError比处理 IndexError更清楚。或者用不同的名称重新引发异常是不好的形式？

54406 次浏览
小开
最佳答案

It's really a matter of semantics. What does foo = latestpdf(d) mean?

Is it perfectly reasonable that there's no latest file? Then sure, just return None.

Are you expecting to always find a latest file? Raise an exception. And yes, re-raising a more appropriate exception is fine.

If this is just a general function that's supposed to apply to any directory, I'd do the former and return None. If the directory is, e.g., meant to be a specific data directory that contains an application's known set of files, I'd raise an exception.

小开

In general, I'd say an exception should be thrown if something catastrophic has occured that cannot be recovered from (i.e. your function deals with some internet resource that cannot be connected to), and you should return None if your function should really return something but nothing would be appropriate to return (i.e. "None" if your function tries to match a substring in a string for example).

小开

I usually prefer to handle exceptions internally (i.e. try/except inside the called function, possibly returning a None) because python is dynamically typed. In general, I consider it a judgment call one way or the other, but in a dynamically typed language, there are small factors that tip the scales in favor of not passing the exception to the caller:

  1. Anyone calling your function is not notified of the exceptions that can be thrown. It becomes a bit of an art form to know what kind of exception you are hunting for (and generic except blocks ought to be avoided).
  2. if val is None is a little easier than except ComplicatedCustomExceptionThatHadToBeImportedFromSomeNameSpace. Seriously, I hate having to remember to type from django.core.exceptions import ObjectDoesNotExist at the top of all my django files just to handle a really common use case. In a statically typed world, let the editor do it for you.

Honestly, though, it's always a judgment call, and the situation you're describing, where the called function receives an error it can't help, is an excellent reason to re-raise an exception that is meaningful. You have the exact right idea, but unless you're exception is going to provide more meaningful information in a stack trace than

AttributeError: 'NoneType' object has no attribute 'foo'

which, nine times out of ten, is what the caller will see if you return an unhandled None, don't bother.

(All this kind of makes me wish that python exceptions had the cause attributes by default, as in java, which lets you pass exceptions into new exceptions so that you can rethrow all you want and never lose the original source of the problem.)

小开

I would make a couple suggestions before answering your question as it may answer the question for you.

  • Always name your functions descriptive. latestpdf means very little to anyone but looking over your function latestpdf() gets the latest pdf. I would suggest that you name it getLatestPdfFromFolder(folder).

As soon as I did this it became clear what it should return.. If there isn't a pdf raise an exception. But wait there more..

  • Keep the functions clearly defined. Since it's not apparent what somefuc is supposed to do and it's not (apparently) obvious how it relates to getting the latest pdf I would suggest you move it out. This makes the code much more readable.
for folder in folders:
try:
latest = getLatestPdfFromFolder(folder)
results = somefuc(latest)
except IOError: pass

Hope this helps!

小开

with python 3.5's typing:

example function when returning None will be:

def latestpdf(folder: str) -> Union[str, None]

and when raising an exception will be:

def latestpdf(folder: str) -> str

option 2 seem more readable and pythonic

(+option to add comment to exception as stated earlier.)


提交答案