The phrase "exceptions must be truly exceptional" is thrown around a lot, but I find that many developers can’t fully understand what that actually means or how to apply that information. They only think they do.
For example, I had a developer argue that an invalid date format is not truly an exception, that it is "simply a MINOR user data entry issue, which could happen often", therefore there was no acceptable case for bubbling, or even throwing an exception for that. He said that having a service go down is a real example of an exception, because that would be a BIG DEAL, and that it rarely occurred.
His distinction was that a web service going down is "rare and exceptional" (true in our case, but it has happened more than once) and that it was a "BIG DEAL", hence the cost of bubbling an exception was acceptable because one would rarely have to pay that price, but at least you're paying the price for something that's EXCEPTIONALLY important and that a user data entry problem is not an exceptionally important problem in the system.
That kind of answer tells me that the developer defines "exceptional" only as something that's SUPERIOR and hence in his mind the issue of a web service going down is superior in importance to the a data entry issue. The developer was ignoring the fact that "exceptional" primarily means "rare".
I had him investigate our logs to see the frequency at which an invalid date format issue, made it through all the layers of our app and he found that it had only happened once, and it never made it past the first service layer.
Now consider the real "expense" between a service going down and a user data entry issue:
The aforementioned service which had an issue that had caused an exception to bubble was 4 layers down in the app stack, and it occurred for hundreds of transactions in the app because of all the users making requests to that area of the app all at the same time. That means that we had hundreds of exceptions bubbling through to the UI, through 4 layers of the app. Also, this happened at least twice in a twelve month period.
So exceptions bubbled by a service going down occurred with much greater frequency, but it’s also important to note that it went through more layers of the app compared with the data entry error. Hence in terms of actual resource EXPENSE, the magnitude of resource cost on our servers for throwing exceptions for data entry errors, is entirely NEGLIGIBLE in this case when compared with the resource expense incurred for a web service going down.
His rebuttal was that an isolated data entry error is negligible compared to the IMPACT of critical service not being available at all. His statement is true and irrefutable; however that is not the point. In fact, that’s the wrong way to look at the issue entirely.
Good quality standards must also dictate that at no point should a developer minimize the importance of the elementary components of an application (such as data entry), simply because their related issues are “low impact”. When one minimizes importance of the elementary components such as data entry validation, simply because they are “low impact”, you end up with COMMON issues in your application for which you would never throw exceptions, because now, you treat them as “common” and “low-impact”. Essentially you’ll end up with spaghetti code where you’re coding to mitigating un-exceptional issues wherever you might encounter them. This approach is unintuitive and highly counter productive.
To me, anything which fails a basic quality standard must be treated with priority, because if you can’t even maintain good standards for the basics, then your whole app is rubbish. Therefore, I would argue that everything should be treated as an exception, simply because problems should be EQUALLY RARE at every level, and when any exception is rare, then expense is not an issue.
If someone you know is creating apps that have so many issues that they have to start worrying about the expense of all the exceptions all over the place in all the applications various layers, then it’s probably time to suggest they scrap their app and start over.
My point is this: if you treat even minor, common issues in your applications as exceptionally important then they will become exceptionally rare, in which case it would be valid and inexpensive to treat them as exceptions.