tags:

views:

1478

answers:

7

The recursion is sort of a 'divide and conquer' style, it splits up while getting smaller (Tree data structure), and I want it to break completely if a violation is found, meaning break all the recursive paths, and return true. Is this possible?

A: 

You could return an error code, or modify some global variable so that each recursive instance knows to "kill itself".

Something of the sort.

int foo(bar){
     int to_the_next;

      if (go_recursive){
            to_the_next = foo(whisky_bar);

            if (to_the_next ==DIE) return DIE;
      }

      if (something_unexpected_happened) return DIE;

      process;//may include some other recursive calls, etc etc
}
Tom
Hmm thanks, I thought of something like that, but I just wanted to see if there was a way to just kill all recursive steps, but I'll just go with this.
Stasis
Well, technically it could be done. If you knew where your original function is located, you could jump to that stack frame, and pretend nothing happened. However, it requires assembly programming (cant think of any other way to do it)
Tom
Yeah I could possibly do it in x68, but this project has to be done in Java ><, and I wouldn't really know how to delve into stack frame's in Java.
Stasis
@Stasis: You always need to backtrack with recursion, it's how it works. Recursion works by use of the stack area of memory... you need to backtrack out of the function calls to resume execution of where you began recursing. This is the very nature/magic of recursion. It seems like you may need to understand recursion more.
Tom
@Tom: (the Tom 3 posts up... not me, 1 post up)... that is a gross optimization that is especially not even worthwhile of contemplating in Java. If you really can't afford the overhead, then do the "recursion" yourself in a loop with a stack. Also, if Java could do this, it would be through one of the memory management classes that is built-in, but that would probably add overhead using those classes. I would never recommend doing this in any high level programming language. Don't get in the habit of contemplating such premature optimizations that are not even good. Yuck :-P.
Tom
Hi (other) Tom that posted this answer... don't you think SO should implement a number system where an ID gets appended to a duplicate username in a post? So if I posted first, I'd be Tom. Then you post, and my name suddenly appears all over as Tom (1), and you appear as Tom (2)... something like that would make things less confusing... tell me if you think it's good and maybe I can make a feature request somehow.
Tom
Thanks to both Tom's. It seems that I'd forgotten the basis of recursion, and somehow got it in my head that it was simultaneous, hence my concern, and I wanted to limit the overhead, although I guess seeing as I'm not dealing with large red-black tree's (not necessarily, anyway) I won't need to go through so much precaution. Thanks again.
Stasis
the code is not breaking all recursive paths (as the original question stated)
Adeel Ansari
@Tom (the other): Yes, it is gross optimization. I was just saying that it could be done. I know its beyond java, and that java wise, its impossible.
Tom
@Tom (the other) Yes, some sort of control for duplicate username would be nice, but it would have to be clear.
Tom
+1  A: 

You can do something similar by storing a variable that tracks whether the recursions should break or not. You'd have to check it every recursion unfortunately but you can do it.

Joe Philllips
+3  A: 

If it's a single thread doing the recursion you could throw an exception. Bit ugly though - kind of using an exception as a goto.

boolean myPublicWrapperMethod(...) {
    try {
        return myPrivateRecursiveFunction(...);
    } catch (MySpecificException e) {
        return true;
    } 
}

A better approach would be to eliminate the recursion and use a Stack collection holding a class representing what would have been recursive state, iterate in a loop and just return true when you want.

Tom
A: 

Unless recursive calls are being evaluated in parallel, you probably just need to add some logic to check the first recursive call's value prior to making the second (and subsequent, if not a binary tree structure) recursive call.

public abstract class Tree {

    protected abstract boolean headIsViolation();

    protected abstract boolean isLeaf();

    public Tree getLeft();

    public Tree getRight();

    // Recursive
    public boolean checkViolation() {
        if(this.isLeaf()) {
            return this.headIsViolation();
        }

        // If needed, you could pass some sort of 'calculation state'
        // object through the recursive calls and evaluate the violation
        // through that object if the current node is insufficient
        if(this.headIsViolation()) {
            // Terminate the recursion
            return true;
        }

        // Fortunately, Java short-circuits ||
        // So if the left child is in violation, the right child will
        // not even be checked
        return this.getLeft().checkViolation() || this.getRight().checkViolation();
    }
}
Greg Case
+10  A: 

No matter what you do, you are going to have to unwind the stack. This leaves two options:

  1. Magic return value (as described by one of the Toms)
  2. Throw an exception (as mentioned by thaggie)

If the case where you want things to die is rare, this may be one of those situations where throwing an exception might be a viable choice. And before everyone jumps down my throat on this, remember that one of the most important rules of programming is knowing when it's appropriate to break the rule.

As it turns out, I spent today evaluating the zxing library from google code. They actually use exception throws for a lot of control structures. My first impression when I saw it was horror. They were literally calling methods tens of thousands of times with different parameters until the method doesn't throw an exception.

This certainly looked like a performance problem, so I made some adjustments to change things over to using a magic return value. And you know what? The code was 40% faster when running in a debugger. But when I switched to non-debugging, the code was less than 1% faster.

I'm still not crazy about the decision to use exceptions for flow control in this case (I mean, the exceptions get thrown all the time). But it's certainly not worth my time to re-implement it given the almost immeasurable performance difference.

If your condition that triggers death of the iteration is not a fundamental part of the algorithm, using an exception may make your code a lot cleaner. For me, the point where I'd make this decision is if the entire recursion needs to be unwound, then I'd use an exception. IF only part of the recursion needs to be unwound, use a magic return value.

Kevin Day
The OP writes "I want it to break completely if a violation is found", which sounds to me like an "exceptional condition", and is exactly what exceptions are for.
DJClayworth
(I'm the author of the zxing code)If a method's contract is "return the barcode found in this row", what would you like it to return when one doesn't exist? 'null' is a cop-out, since it is returning nothing. An exception is appropriate.But your point was performance. The other author challenged me on this initially. If you dig in the code you'll see we use a singleton exception to get around performance issues.
Sean Owen
I would always start from the position that the performance in an exceptional condition doesn't matter. If profiling showed me otherwise I would change my mind.
DJClayworth
Sean - very good to hear from you! I probably would have made the return result contain a marker for found/not-found (instead of returning null or throwing an exception). But like I said, performance is performance, right?
Kevin Day
DJ - agreed for the OP. In the case of zxing, the 'exceptional condition' isn't very exceptional (it's actually more of the norm than the exception). If I were designing it from scratch, I'd do it differently, but the performance measurements show that the current approach doesn't significantly impact speed.
Kevin Day
A: 

What you are asking is the definition of recursion.

At some point all recursive paths should break. Otherwise it will be an infinite recursion and stack overflow exception occurs.

So you should design the recursion function like that. An example binary search in a sorted array:

BinarySearch(A[0..N-1], value, low, high) {
       if (high < low)
           return -1 // not found
       mid = low + ((high - low) / 2)  // Note: not (low + high) / 2 !!
       if (A[mid] > value)
           return BinarySearch(A, value, low, mid-1)
       else if (A[mid] < value)
           return BinarySearch(A, value, mid+1, high)
       else
           return mid // found
}
Emre
A: 

I'd recommend an exception handling. That makes clear, that the recursion was aborted because of some violation (or other exception):

public void outer() {
  try {
    int i = recurse(0);
  } catch (OuchException ouch) {
    // handle the exception here
  }
}

private int recurse(int i) throws OuchException {

    // for tree-like structures
    if (thisIsALeaf) {
       return i;
    }

    // the violation test
    if (itHurts)
       throw new OuchException("That really hurts");

    // we also can encapsulate other exceptions
    try {
       someMethod();
    } catch (Exception oops) {
       throw new OuchException(oops);
    }

    // recurse
    return recurse(i++);
}

Sure, it violates the initial requirement to return 'true' upon abortion. But I prefer a clean separation of return values and notification on abnormal behaviour.

Andreas_D