How should I correctly handle exceptions in Python3

Question

I can\'t understand what sort of exceptions I should handle \'here and now\', and what sort of exceptions I should re-raise or just don\'t handle here, and what to do with t

Answer 1

In general, you should "catch" the exceptions that you expect to happen (because they may be caused by user error, or other environmental problems outside of your program's control), especially if you know what your code might be able to do about them. Just giving more details in an error report is a marginal issue, though some programs' specs may require doing that (e.g. a long-running server that's not supposed to crash due to such problems, but rather log a lot of state information, give the user a summary explanation, and just keep working for future queries).

NameError, TypeError, KeyError, ValueError, SyntaxError, AttributeError, and so on, can be thought of as due to errors in the program -- bugs, not problems outside of the programmer's control. If you're releasing a library or framework, so that your code is going to be called by other code outside of your control, then such bugs may quite likely be in that other code; you should normally let the exception propagate to help the other programmer debug their own bugs. If you're releasing an application, you own the bugs, and you must pick the strategy that helps you find them.

If your bugs show up while an end-user is running the program, you should log a lot of state information, and give the user a summary explanation and apologies (perhaps with a request to send you the log info, if you can't automate that -- or, at least, ask permission before you send anything from the user's machine to yours). You may be able to save some of the user's work so far, but often (in a program that's known to be buggy) that may not work anyway.

Most bugs should show up during your own testing of course; in that case, propagating the exception is useful as you can hook it up to a debugger and explore the bug's details.

Sometimes some exceptions like these show up just because "it's easier to ask forgiveness than permission" (EAFP) -- a perfectly acceptable programming technique in Python. In that case of course you should handle them at once. For example:

try:
    return mylist[theindex]
except IndexError:
    return None

here you might expect that theindex is generally a valid index into mylist, but occasionally outside of mylist's bounds -- and the latter case, by the semantics of the hypothetic app in which this snippet belongs, is not an error, just a little anomaly to be fixed by considering the list to be conceptually extended on both sides with infinite numbers of Nones. It's easier to just try/except than to properly check for positive and negative values of the index (and faster, if being out of bounds is a truly rare occurrence).

Similarly appropriate cases for KeyError and AttributeError happen less frequently, thanks to the getattr builtin and get method of dicts (which let you provide a default value), collections.defaultdict, etc; but lists have no direct equivalent of those, so the try/except is seen more frequently for IndexError.

Trying to catch syntax errors, type errors, value errors, name errors, etc, is a bit rarer and more controversial -- though it would surely be appropriate if the error was diagnosed in a "plug-in", third-party code outside your control which your framework/application is trying to load and execute dynamically (indeed that's the case where you're supplying a library or the like and need to coexist peacefully with code out of your control which might well be buggy). Type and value errors may sometimes occur within an EAFP pattern -- e.g. when you try to overload a function to accept either a string or a number and behave slightly differently in each case, catching such errors may be better than trying to check types -- but the very concept of functions thus overloaded is more often than not quite dubious.

Back to "user and environmental errors", users will inevitably make mistakes when they give you input, indicate a filename that's not actually around (or that you don't have permission to read, or to write if that's what you're supposed to be doing), and so on: all such errors should of course be caught and result in a clear explanation to the user about what's gone wrong, and another chance to get the input right. Networks sometime go down, databases or other external servers may not respond as expected, and so forth -- sometimes it's worth catching such problems and retrying (maybe after a little wait -- maybe with an indication to the user about what's wrong, e.g. they may have accidentally unplugged a cable and you want to give them a chance to fix things and tell you when to try again), sometimes (especially in unattended long-running programs) there's nothing much you can do except an ordered shutdown (and detailed logging of every possibly-relevant aspect of the environment).

So, in brief, the answer to your Q's title is, "it depends";-). I hope I have been of use in listing many of the situations and aspects on which it can depend, and recommending what's generally the most useful attitude to take towards such issues.

Answer 2

To start with, you don't need any _mainLogger. If you want to catch any exceptions, maybe to log or send them by email or whatever, do that at the highest possible level -- certainly not inside this class.

Also, you definitely don't want to convert every Exception to a RuntimeError. Let it emerge. The stopClient() method has no purpose right now. When it has, we'll look at it..

You could basically wrap the ConnectionError, IOError and OSError together (like, re-raise as something else), but not much more than that...