Last update: 2022-05-18 10:48 Irish Time
Recently a colleague was asking me for advice on their design of error handling in a Python application.
They were catching an error and raising an Exception, inside the except part of a method, to be catch outside the method.
And at some point a simple logic got really messy and unnecessarily complicated. Also troubleshooting and debugging an error was painful because they were only getting a Custom Exception and not context.
I explained to my colleague that I believed that the person that created that Exception chain of catch came from Java background and why I think they choose that path, and why I think in Python it’s a bad idea.
In Java, functions and methods can only return one object.
I programmed a lot in Java in my career, and it was a pain having to create value objects, and having to create all kind of objects for the return. Is it a good thing that types are strongly verified by the language? Yes. It worked? Yes. It made me invest much more time than necessary? Also yes.
Having the possibility to return only one object makes it mandatory having a way to return when there was an error. Otherwise you would need to encapsulate an error code and error description fields in each object, which is contrary to the nature of the object.
For example, a class Persona. Doesn’t make any sense having an attribute inside the class Persona to register if an operation related to this object went wrong.
For example, if we are in a class Spaceship that has a method GetPersonaInCommand() and there is a problem in that method, doesn’t make any sense to return an empty Persona object with attributes idError, errorDescription. Probably the Constructor or Persona will require at least a name or Id to build the object…. so in this case, makes sense that the method raises an Exception so the calling code catches it and knows that something went wrong or when there is no data to return.
This will force to write Custom Exceptions, but it’s a solution.
Another solution is creating a generic response object which could be an Object with these attributes:
- an Object which is the response, in our example Persona or null
I created this kind of approach for my Cassandra libraries to easily work with Cassandra from Java and from PHP, and for Cassandra Universal Driver (a http/s gateway created in year 2014).
Why this in not necessary in Python
Python allows you to return multiple values, so I encourage you tor return a boolean for indicating the success of the operation, and the object/value you’re interested.
You can see it easily if you take a look to FileUtils class from my OpenSource libraries carleslibs.
The method get_file_size_in_bytes(self, s_file) for example:
def get_file_size_in_bytes(self, s_file):
b_success = False
i_file_size = 0
# This will help with Unit Testing by raisin IOError Exception
i_file_size = os.path.getsize(s_file)
b_success = True
b_success = False
return b_success, i_file_size
It will always return a boolean value to indicate success or failure of the operation and an integer for the size of the file.
The calling code will do something like this:
o_fileutils = FileUtils()
b_success, i_bytes = o_fileutils.get_file_size_in_bytes("profile.png")
if b_succes is False:
print("Error! The file does not exist or cannot be accessed!")
if i_bytes < 1024:
print("The profile picture should be at least 1KB")
print("Profile picture exists and is", i_bytes, " bytes in length!")
The fact that Python can return multiple variables makes super easy dealing with error handling without having to take the road of Custom Exceptions.
And it is Ok if you want to follow this path, but in my opinion, for most of the developers up to Senior levels, it only over complicates the logic of your code and the amount of try/excepts you have to have everywhere.
If you use PHP you can mix different types in an Array, so you can always return an Array with a boolean, or an i_id_error, and your object or data of whatever type it’s.
Getting back to my carleslibs Open Source package, it is super easy to Unit Test these methods.
In my opinion, this level of simplicity, brings only advantages. Including Software Development speed, which is good for the business.
I’m not advocating for not using Custom Exceptions or to not develop a Exceptions Raising strategy if you need it and you know what you’re doing. I’m just suggesting why I think most of the developments in Python do not really need this and only over complicates the development. There are situations where raising exceptions will be a perfectly useful or even the best approach, there are many scenarios, but I think that in most of cases, using raise inside except will only multiply the time of the development and slow down the speed of bringing new features to the business, over complicating Unit Test as well, and be a real pain for the Junior and Intermediate developers.
Obviously, as the Constructor doesn’t return any value, it is perfectly fine to raise an exception in there, or just to use try/except in the code that is instancing the objects.