Why not use exceptions as regular flow of control?

To avoid all standard-answers I could have Googled on, I will provide an example you all can attack at will.

C# and Java (and too many others) have with plenty of types some of ‘overflow’ behaviour I don’t like at all (e.g type.MaxValue + type.SmallestValue == type.MinValue for example : int.MaxValue + 1 == int.MinValue).

But, seen my vicious nature, I’ll add some insult to this injury by expanding this behaviour to, let’s say an Overridden DateTime type. (I know DateTime is sealed in .NET, but for the sake of this example, I’m using a pseudo language that is exactly like C#, except for the fact that DateTime isn’t sealed).

The overridden Add method:

/// <summary>
/// Increments this date with a timespan, but loops when
/// the maximum value for datetime is exceeded.
/// </summary>
/// <param name="ts">The timespan to (try to) add</param>
/// <returns>The Date, incremented with the given timespan. 
/// If DateTime.MaxValue is exceeded, the sum wil 'overflow' and 
/// continue from DateTime.MinValue. 
/// </returns>
public DateTime override Add(TimeSpan ts) 
{
    try
    {                
        return base.Add(ts);
    }
    catch (ArgumentOutOfRangeException nb)
    {
        // calculate how much the MaxValue is exceeded
        // regular program flow
        TimeSpan saldo = ts - (base.MaxValue - this);
        return DateTime.MinValue.Add(saldo)                			
    }
    catch(Exception anyOther) 
    {
        // 'real' exception handling.
    }
}

Of course an if could solve this just as easy, but the fact remains that I just fail to see why you couldn’t use exceptions (logically that is, I can see that when performance is an issue that in certain cases exceptions should be avoided).

I think in many cases they are more clear than if-structures and don’t break any contract the method is making.

IMHO the “Never use them for regular program flow” reaction everybody seems to have is not that well underbuild as the strength of that reaction can justify.

Or am I mistaken?

I've read other posts, dealing with all kind of special cases, but my point is there's nothing wrong with it if you are both:

1. Clear
2. Honour the contract of your method

Shoot me.

Have you ever tried to debug a program raising five exceptions per second in the normal course of operation ?

I have.

The program was quite complex (it was a distributed calculation server), and a slight modification at one side of the program could easily break something in a totally different place.

I wish I could just have launched the program and wait for exceptions to occur, but there were around 200 exceptions during the start-up in the normal course of operations

My point : if you use exceptions for normal situations, how do you locate unusual (ie exceptional) situations ?

Of course, there are other strong reasons not to use exceptions too much, especially performance-wise

Exceptions are basically non-local goto statements with all the consequences of the latter. Using exceptions for flow control violates a principle of least astonishment, make programs hard to read (remember that programs are written for programmers first).

Moreover, this is not what compiler vendors expect. They expect exceptions to be thrown rarely, and they usually let the throw code be quite inefficient. Throwing exceptions is one of the most expensive operations in .NET.

However, some languages (notably Python) use exceptions as flow-control constructs. For example, iterators raise a StopIteration exception if there are no further items. Even standard language constructs (such as for) rely on this.

My rule of thumb is:

• If you can do anything to recover from an error, catch exceptions
• If the error is a very common one (eg. user tried to log in with the wrong password), use returnvalues
• If you can't do anything to recover from an error, leave it uncaught (Or catch it in your main-catcher to do some semi-graceful shutdown of the application)

The problem I see with exceptions is from a purely syntax point of view (I'm pretty sure the perfomance overhead is minimal). I don't like try-blocks all over the place.

Take this example:

try
{
   DoSomeMethod();  //Can throw Exception1
   DoSomeOtherMethod();  //Can throw Exception1 and Exception2
}
catch(Exception1)
{
   //Okay something messed up, but is it SomeMethod or SomeOtherMethod?
}

.. Another example could be when you need to assign something to a handle using a factory, and that factory could throw an exception:

Class1 myInstance;
try
{
   myInstance = Class1Factory.Build();
}
catch(SomeException)
{
   // Couldn't instantiate class, do something else..
}
myInstance.BestMethodEver();   // Will throw a compile-time error, saying that myInstance is uninitalized, which it potentially is.. :(

Soo, personally, I think you should keep exceptions for rare error-conditions (out of memory etc.) and use returnvalues (valueclasses, structs or enums) to do your error checking instead.

Hope I understood your question correct :)

A first reaction to a lot of answers :

> you're writing for the programmers and the principle of least astonishment

Of course! But an if just isnot more clear all the time.

It shouldn't be astonishing eg : divide (1/x) catch (divisionByZero) is more clear than any if to me (at Conrad and others) . The fact this kind of programming isn't expected is purely conventional, and indeed, still relevant. Maybe in my example an if would be clearer.

But DivisionByZero and FileNotFound for that matter are clearer than ifs.

Of course if it's less performant and needed a zillion time per sec, you should of course avoid it, but still i haven't read any good reason to avoid the overal design.

As far as the principle of least astonishment goes : there's a danger of circular reasoning here : suppose a whole community uses a bad design, this design will become expected! Therefore the principle cannot be a grail and should be concidered carefully.

> exceptions for normal situations, how do you locate unusual (ie exceptional) situations ?

In many reactions sth. like this shines trough. Just catch them, no? Your method should be clear, well documented, and hounouring it's contract. I don't get that question I must admit.

Debugging on all exceptions : the same, that's just done sometimes because the design not to use exceptions is common. My question was : why is it common in the first place?

Before exceptions, in C, there were setjmp and longjmp that could be used to accomplish a similar unrolling of the stack frame.

Then the same construct was given a name: "Exception". And most of the answers rely on the meaning of this name to argue about its usage, claiming that exceptions are intended to be used in exceptional conditions. That was never the intent in the original longjmp. There were just situations where you needed to break control flow across many stack frames.

Exceptions are slightly more general in that you can use them within the same stack frame too. This raises analogies with goto that I believe are wrong. Gotos are a tightly coupled pair (and so are setjmp and longjmp). Exceptions follow a loosely coupled publish/subscribe that is much cleaner! Therefore using them within the same stack frame is hardly the same thing as using gotos.

The third source of confusion relates to whether they are checked or unchecked exceptions. Of course, unchecked exceptions seem particularly awful to use for control flow and perhaps a lot of other things.

Checked exceptions however are great for control flow, once you get over all the Victorian hangups and live a little.

My favorite usage is a sequence of throw new Success() in a long fragment of code that tries one thing after the other until it finds what it is looking for. Each thing -- each piece of logic -- may have arbritrary nesting so break's are out as also any kind of condition tests. The if-else pattern is brittle. If I edit out an else or mess up the syntax in some other way, then there is a hairy bug.

Using throw new Success() linearizes the code flow. I use locally defined Success classes -- checked of course -- so that if I forget to catch it the code won't compile. And I don't catch another method's Successes.

Sometimes my code checks for one thing after the other and only succeeds if everything is OK. In this case I have a similar linearization using throw new Failure().

Using a separate function messes with the natural level of compartmentalization. So the return solution is not optimal. I prefer to have a page or two of code in one place for cognitive reasons. I don't believe in ultra-finely divided code.

What JVMs or compilers do is less relevant to me unless there is a hotspot. I cannot believe there is any fundamental reason for compilers to not detect locally thrown and caught Exceptions and simply treat them as very efficient gotos at the machine code level.

As far as using them across functions for control flow -- i. e. for common cases rather than exceptional ones -- I cannot see how they would be less efficient than multiple break, condition tests, returns to wade through three stack frames as opposed to just restore the stack pointer.

I personally do not use the pattern across stack frames and I can see how it would require design sophistication to do so elegantly. But used sparingly it should be fine.

Lastly, regarding surprising virgin programmers, it is not a compelling reason. If you gently introduce them to the practice, they will learn to love it. I remember C++ used to surprise and scare the heck out of C programmers.

The standard anwser is that exceptions are not regular and should be used in exceptional cases.

One reason, which is important to me, is that when I read a try-catch control structure in a software I maintain or debug, I try to find out why the original coder used an exception handling instead of an if-else structure. And I expect to find a good answer.

Remember that you write code not only for the computer but also for other coders. There is a semantic associated to an exception handler that you cannot throw away just because the machine doesn't mind.

How about performance? While load testing a .NET web app we topped out at 100 simulated users per web server until we fixed a commonly-occuring exception and that number increased to 500 users.

Josh Bloch deals with this topic extensively in Effective Java. His suggestions are illuminating and should apply to .NET as well (except for the details).

In particular, exceptions should be used for exceptional circumstances. The reasons for this are usability-related, mainly. For a given method to be maximally usable, its input and output conditions should be maximally constrained.

For example, the second method is easier to use than the first:

/**
 * Adds two positive numbers.
 *
 * @param addend1 greater than zero
 * @param addend2 greater than zero
 * @throws AdditionException if addend1 or addend2 is less than or equal to zero
 */
int addPositiveNumbers(int addend1, int addend2) throws AdditionException{
  if( addend1 <= 0 ){
     throw new AdditionException("addend1 is <= 0");
  }
  else if( addend2 <= 0 ){
     throw new AdditionException("addend2 is <= 0");
  }
  return addend1 + addend2;
}

/**
 * Adds two positive numbers.
 *
 * @param addend1 greater than zero
 * @param addend2 greater than zero
 */
public int addPositiveNumbers(int addend1, int addend2) {
  if( addend1 <= 0 ){
     throw new IllegalArgumentException("addend1 is <= 0");
  }
  else if( addend2 <= 0 ){
     throw new IllegalArgumentException("addend2 is <= 0");
  }
  return addend1 + addend2;
}

In either case, you need to check to make sure that the caller is using your API appropriately. But in the second case, you require it (implicitly). The soft Exceptions will still be thrown if the user didn't read the javadoc, but:

> 1. You don't need to document it. > 2. You don't need to test for it (depending upon how aggresive your > unit testing strategy is). > 3. You don't require the caller to handle three use cases.

The ground-level point is that Exceptions should not be used as return codes, largely because you've complicated not only YOUR API, but the caller's API as well.

Doing the right thing comes at a cost, of course. The cost is that everyone needs to understand that they need to read and follow the documentation. Hopefully that is the case anyway.

I think that you can use Exceptions for flow control. There is, however, a flipside of this technique. Creating Exceptions is a costly thing, because they have to create a stack trace. So if you want to use Exceptions more often than for just signalling an exceptional situation you have to make sure that building the stack traces doesn't negatively influence your performance.

The best way to cut down the cost of creating exceptions is to override the fillInStackTrace() method like this:

public Throwable fillInStackTrace() { return this; }

Such an exception will have no stacktraces filled in.

Here are best practices I described in my blog post:

• Throw an exception to state an unexpected situation in your software.

• Use return values for input validation.

• If you know how to deal with exceptions a library throws, catch them at the lowest level possible.

• If you have an unexpected exception, discard current operation completely. Don’t pretend you know how to deal with them.

I don't really see how you're controlling program flow in the code you cited. You'll never see another exception besides the ArgumentOutOfRange exception. (So your second catch clause will never be hit). All you're doing is using an extremely costly throw to mimic an if statement.

Also you aren't performing the more sinister of operations where you just throw an exception purely for it to be caught somewhere else to perform flow control. You're actually handling an exceptional case.

Apart from the reasons stated, one reason not to use exceptions for flow control is that it can greatly complicate the debugging process.

For example, when I'm trying to track down a bug in VS I'll typically turn on "break on all exceptions". If you're using exceptions for flow control then I'm going to be breaking in the debugger on a regular basis and will have to keep ignoring these non-exceptional exceptions until I get to the real problem. This is likely to drive someone mad!!

Because the code is hard to read, you may have troubles debugging it, you will introduce new bugs when fixing bugs after a long time, it is more expensive in terms of resources and time, and it annoys you if you are debugging your code and the debugger halts on the occurence of every exception ;)

Lets assume you have a method that does some calculations. There are many input parameters it has to validate, then to return a number greater then 0.

Using return values to signal validation error, it's simple: if method returned a number lesser then 0, an error occured. How to tell then which parameter didn't validate?

I remember from my C days a lot of functions returned error codes like this:

-1 - x lesser then MinX
-2 - x greater then MaxX
-3 - y lesser then MinY

etc.

Is it really less readable then throwing and catching an exception?

You can use a hammer's claw to turn a screw, just like you can use exceptions for control flow. That doesn't mean it is the intended usage of the feature. The if statement expresses conditions, whose intended usage is controlling flow.

If you are using a feature in an unintended way while choosing to not use the feature designed for that purpose, there will be an associated cost. In this case, clarity and performance suffer for no real added value. What does using exceptions buy you over the widely-accepted if statement?

Said another way: just because you can doesn't mean you should.

As others have mentioned numerously, the principle of least astonishment will forbid that you use exceptions excessively for control flow only purposes. On the other hand, no rule is 100% correct, and there are always those cases where an exception is "just the right tool" - much like goto itself, by the way, which ships in the form of break and continue in languages like Java, which are often the perfect way to jump out of heavily nested loops, which aren't always avoidable.

The following blog post explains a rather complex but also rather interesting use-case for a non-local ControlFlowException:

• http://blog.jooq.org/2013/04/28/rare-uses-of-a-controlflowexception

It explains how inside of jOOQ (a SQL abstraction library for Java), such exceptions are occasionally used to abort the SQL rendering process early when some "rare" condition is met.

Examples of such conditions are:

• Too many bind values are encountered. Some databases do not support arbitrary numbers of bind values in their SQL statements (SQLite: 999, Ingres 10.1.0: 1024, Sybase ASE 15.5: 2000, SQL Server 2008: 2100). In those cases, jOOQ aborts the SQL rendering phase and re-renders the SQL statement with inlined bind values. Example:

    // Pseudo-code attaching a "handler" that will
    // abort query rendering once the maximum number
    // of bind values was exceeded:
    context.attachBindValueCounter();
    String sql;
    try {
  
      // In most cases, this will succeed:
      sql = query.render();
    }
    catch (ReRenderWithInlinedVariables e) {
      sql = query.renderWithInlinedBindValues();
    }
  

  If we explicitly extracted the bind values from the query AST to count them every time, we'd waste valuable CPU cycles for those 99.9% of the queries that don't suffer from this problem.

• Some logic is available only indirectly via an API that we want to execute only "partially". The UpdatableRecord.store() method generates an INSERT or UPDATE statement, depending on the Record's internal flags. From the "outside", we don't know what kind of logic is contained in store() (e.g. optimistic locking, event listener handling, etc.) so we don't want to repeat that logic when we store several records in a batch statement, where we'd like to have store() only generate the SQL statement, not actually execute it. Example:

    // Pseudo-code attaching a "handler" that will
    // prevent query execution and throw exceptions
    // instead:
    context.attachQueryCollector();
    
    // Collect the SQL for every store operation
    for (int i = 0; i < records.length; i++) {
      try {
        records[i].store();
      }
    
      // The attached handler will result in this
      // exception being thrown rather than actually
      // storing records to the database
      catch (QueryCollectorException e) {
    
        // The exception is thrown after the rendered
        // SQL statement is available
        queries.add(e.query());                
      }
    }
  

  If we had externalised the store() logic into "re-usable" API that can be customised to optionally not execute the SQL, we'd be looking into creating a rather hard to maintain, hardly re-usable API.

Conclusion

In essence, our usage of these non-local gotos is just along the lines of what [Mason Wheeler][5] said in his answer:

> "I just encountered a situation that I cannot deal with properly at this point, because I don't have enough context to handle it, but the routine that called me (or something further up the call stack) ought to know how to handle it."

Both usages of ControlFlowExceptions were rather easy to implement compared to their alternatives, allowing us to reuse a wide range of logic without refactoring it out of the relevant internals.

But the feeling of this being a bit of a surprise to future maintainers remains. The code feels rather delicate and while it was the right choice in this case, we'd always prefer not to use exceptions for local control flow, where it is easy to avoid using ordinary branching through if - else.

Typically there is nothing wrong, per se, with handling an exception at a low level. An exception IS a valid message that provides a lot of detail for why an operation cannot be performed. And if you can handle it, you ought to.

In general if you know there is a high probability of failure that you can check for... you should do the check... i.e. if(obj != null) obj.method()

In your case, i'm not familiar enough with the C# library to know if date time has an easy way to check whether a timestamp is out of bounds. If it does, just call if(.isvalid(ts)) otherwise your code is basically fine.

So, basically it comes down to whichever way creates cleaner code... if the operation to guard against an expected exception is more complex than just handling the exception; than you have my permission to handle the exception instead of creating complex guards everywhere.

If you are using exception handlers for control flow, you are being too general and lazy. As someone else mentioned, you know something happened if you are handling processing in the handler, but what exactly? Essentially you are using the exception for an else statement, if you are using it for control flow.

If you don't know what possible state could occur, then you can use an exception handler for unexpected states, for example when you have to use a third-party library, or you have to catch everything in the UI to show a nice error message and log the exception.

However, if you do know what might go wrong, and you don't put an if statement or something to check for it, then you are just being lazy. Allowing the exception handler to be the catch-all for stuff you know could happen is lazy, and it will come back to haunt you later, because you will be trying to fix a situation in your exception handler based on a possibly false assumption.

If you put logic in your exception handler to determine what exactly happened, then you would be quite stupid for not putting that logic inside the try block.

Exception handlers are the last resort, for when you run out of ideas/ways to stop something from going wrong, or things are beyond your ability to control. Like, the server is down and times out and you can't prevent that exception from being thrown.

Finally, having all the checks done up front shows what you know or expect will occur and makes it explicit. Code should be clear in intent. What would you rather read?

You might be interested in having a look at Common Lisp's condition system which is a sort of generalization of exceptions done right. Because you can unwind the stack or not in a controlled way, you get "restarts" as well, which are extremely handy.

This doesn't have anything much to do with best practices in other languages, but it shows you what can be done with some design thought in (roughly) the direction you are thinking of.

Of course there are still performance considerations if you're bouncing up and down the stack like a yo-yo, but it's a much more general idea than "oh crap, lets bail" kind of approach that most catch/throw exception systems embody.

I don't think there is anything wrong with using Exceptions for flow-control. Exceptions are somewhat similar to continuations and in statically typed languages, Exceptions are more powerful than continuations, so, if you need continuations but your language doesn't have them, you can use Exceptions to implement them.

Well, actually, if you need continuations and your language doesn't have them, you chose the wrong language and you should rather be using a different one. But sometimes you don't have a choice: client-side web programming is the prime example – there's just no way to get around JavaScript.

An example: Microsoft Volta is a project to allow writing web applications in straight-forward .NET, and let the framework take care of figuring out which bits need to run where. One consequence of this is that Volta needs to be able to compile CIL to JavaScript, so that you can run code on the client. However, there is a problem: .NET has multithreading, JavaScript doesn't. So, Volta implements continuations in JavaScript using JavaScript Exceptions, then implements .NET Threads using those continuations. That way, Volta applications that use threads can be compiled to run in an unmodified browser – no Silverlight needed.

I feel that there is nothing wrong with your example. On the contrary, it would be a sin to ignore the exception thrown by the called function.

In the JVM, throwing an exception is not that expensive, only creating the exception with new xyzException(...), because the latter involves a stack walk. So if you have some exceptions created in advance, you may throw them many times without costs. Of course, this way you can't pass data along with the exception, but I think that is a bad thing to do anyway.

But you won't always know what happens in the Method/s that you call. You won't know exactly where the exception was thrown. Without examining the exception object in greater detail....

There are a few general mechanisms via which a language could allow for a method to exit without returning a value and unwind to the next "catch" block:

• Have the method examine the stack frame to determine the call site, and use the metadata for the call site to find either information about a try block within the calling method, or the location where the calling method stored the address of its caller; in the latter situation, examine metadata for the caller's caller to determine in the same fashion as the immediate caller, repeating until one finds a try block or the stack is empty. This approach adds very little overhead to the no-exception case (it does preclude some optimizations) but is expensive when an exception occurs.

• Have the method return a "hidden" flag which distinguishes a normal return from an exception, and have the caller check that flag and branch to an "exception" routine if it's set. This routine adds 1-2 instructions to the no-exception case, but relatively little overhead when an exception occurs.

• Have the caller place exception-handling information or code at a fixed address relative to the stacked return address. For example, with the ARM, instead of using the instruction "BL subroutine", one could use the sequence:

        adr lr,next_instr
        b subroutine
        b handle_exception
    next_instr:
  

To exit normally, the subroutine would simply do bx lr or pop {pc}; in case of an abnormal exit, the subroutine would either subtract 4 from LR before performing the return or use sub lr,#4,pc (depending upon the ARM variation, execution mode, etc.) This approach will malfunction very badly if the caller is not designed to accommodate it.

A language or framework which uses checked exceptions might benefit from having those handled with a mechanism like #2 or #3 above, while unchecked exceptions are handled using #1. Although the implementation of checked exceptions in Java is rather nuisancesome, they would not be a bad concept if there were a means by which a call site could say, essentially, "This method is declared as throwing XX, but I don't expect it ever to do so; if it does, rethrow as an "unchecked" exception. In a framework where checked exceptions were handled in such fashion, they could be an effective means of flow control for things like parsing methods which in some contexts may have a high likelihood of failure, but where failure should return fundamentally different information than success. I'm unaware of any frameworks that use such a pattern, however. Instead, the more common pattern is to use the first approach above (minimal cost for the no-exception case, but high cost when exceptions are thrown) for all exceptions.

One aesthetic reason:

A try always comes with a catch, whereas an if doesn't have to come with an else.

if (PerformCheckSucceeded())
   DoSomething();

With try/catch, it becomes much more verbose.

try
{
   PerformCheckSucceeded();
   DoSomething();
}
catch
{
}

That's 6 lines of code too many.