Exceptions for controlling program flow: take it easy on your poor CPU

A lot of developers don’t see what’s so bad about using exceptions to control program flow.

I have weighed in on this question, and my general opinion is that the main problem—from a practical standpoint—is one of performance. Yes, it’s an implementation detail and varies by programming language/framework; but I would wager that 9 times out of 10, writing code that is not unlikely to throw an exception (what Eric Lippert calls a “boneheaded exception”) and catching that exception is less efficient than dealing with the problem beforehand—nipping it in the bud, if you will—sans exception.

But this blog is titled “The Philosopher Developer,” after all; so now I’m going to offer my philosophical explanation. And, as has become customary for my blog posts, I will do so in the form of an analogy.

The characters:

  • You: the developer
  • Servant: the computer, running your software
  • Your Instructions: your software (obviously)
  • The Store: the operating environment

You have a servant whom you use to accomplish various small tasks for you—running errands, performing simple work around the house, etc. (Sounds nice, doesn’t it?)

One day you decide you want something cold to drink, but there’s nothing in your fridge. So you give your servant some cash and send him to the grocery store with these instructions: “Buy me a Coke.”

Now, let’s say your servant is extremely literal and incapable of making his own decisions. He can only follow instructions. You realize this, and so in your mind you think: If the store doesn’t have any Coke, maybe I’d be fine with a Pepsi instead. As a last resort, I could even go for some orange juice. (This is of course supposing you know for a fact that the store will always have orange juice.) Your plan is, if your servant can’t find a Coke, he will have a panic attack and come back to you very distressed. Then you will let him know your alternate choice.

This is what it’s like to throw and catch exceptions: you’re accounting for disastrous scenarios without letting your servant “in” on the details of your backup plan. Consider the energy your servant has wasted going to the store, trying to execute your instructions, failing, and then coming back to you to report on his failure, when all along you knew exactly what you wanted to do if he were unsuccessful. You have anticipated the disaster and deliberately allowed it to happen as a means of controlling your servant’s workflow.

The alternative, obviously, would have been to let your servant know of your 2nd and 3rd choices in advance. This would have allowed him to make only one trip to the grocery store, where he could have looked for Coke, Pepsi, and orange juice—in that order—without any panic attacks or distress, and most importantly without needing to return to you empty-handed. The disaster (the exception) thus could have been prevented altogether.

When you think of it this way, it really is just the sensible thing to do all-around to try and avoid exceptions when it’s possible to do so.

Advertisements

3 thoughts on “Exceptions for controlling program flow: take it easy on your poor CPU

  1. Bragaadeesh says:

    Till now, i have thought exceptions were fool proof and I still tend to think that way. But one thing that i could relate real life to programming is the stack that the program has to painfully come out of during an exception (it has to go all through the stack to get the proper exception handler) which ofcourse is costly.
    By the way, I have started to check your blog daily and am really excited if there is a new post!

    • Daniel says:

      Yeah, unwinding the stack to assemble a stack trace is definitely one cost. Another is undergoing whatever trauma had to occur to throw the exception in the first place! For example, attempting to read from a file that doesn’t exist can pose a lot more overhead than would simply constructing a new Exception object with a stack trace. Of course, sometimes the occurrence of an exception might be outside one’s control (e.g., when dealing with I/O, unmanaged code, etc.); mainly I’m talking about situations where exceptions are used instead of more lightweight precautions such as argument validation, index range checks—stuff like that. Also, thanks for reading!

  2. Mark T says:

    There can also be the mental cost of trying to understand the code as you read it – exceptions are really just a special form of goto (as mentioned here: http://www.joelonsoftware.com/articles/Wrong.html)!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: