Take the 2-minute tour ×
Programmers Stack Exchange is a question and answer site for professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.

Now I'm only a novice programmer, but what my friends that are studying programming tell me is that recursive code is a good thing, it leads to less code duplication and it looks more elegant.

Now it may be so, but everytime I've tried my hand with recursive code I've always found it to be significantly slower than non-recursive code, I'm not sure if it's something about the ways I've used it but I highly doubt it, I've used it for fairly simply thing's like Collatz Conjecture and Fibonacci number generation but whenever I've compared it against normal iterative code the recursive approach will consistently clock at around twice the time of the iterative solutions.

share|improve this question
    
well if you're calculating Fibonacci using something like return f(n-1) + f(n-2), you're actually doing twice the work. You can use memoization to make sure you're only calculating each term once. –  Lescai Ionel May 4 at 13:43
4  
@LescaiIonel You're actually doing exponentially more work, since each of those recursive calls itself duplicates work. If it was just a constant factor of two, memoization might not necessarily be a win. OP should give some examples of "the" recursive approach (side by side with the iterative version that it was compared to) to determine whether the two algorithms are different of whether it's really just recursion vs. iteration. –  delnan May 4 at 13:45
    
I've found traversing a node tree performs faster recursively, then non-recursively. Non-recursive while loops require more work to get around what can be done easily recursively. It seems counter intuitive but I implemented a non-recursive tree walker thinking it would be faster to only revert my changes after benchmarking showed it was slower. –  Mathew Foscarini May 4 at 15:51
    
Fibonacci is actually a nice example of how to do recursion and then in the next lesson learn how to do it faster non-recursive. –  Pieter B May 4 at 16:58
    
@PieterB: There is also the faster recursive version of Fibonacci. No need to avoid recursion to have a fast implementation. –  Giorgio May 4 at 21:03

2 Answers 2

up vote 8 down vote accepted

It is good to at least understand how recursion works, because some algorithms are naturally recursive and thus much easier to express using recursion.

Also, recursive algorithms (or implementations) are not inherently slower than iterative ones. In fact, every recursive algorithm can be implemented by an equivalent iterative implementation at the cost of having to keep track some intermediate/temporary values yourself, where the recursive version automatically keeps track of those in the function call stack.

One example of a naturally recursive algorithm is if you want to apply an operation to all the nodes of a tree. The most natural implementation here is to have a function that performs the operation on one node and calls itself recursively for each of the children of the node.


For some algorithms, like calculating a Fibonacci sequence, recursion seems natural, but a naive implementation will be much slower than an iterative implementation, because the recursive version keeps re-doing the same work over and over again. This just means that for those particular algorithms, recursion may not be the best choice, or that you need to use some techniques to remember intermediate values that might be needed again elsewhere in the algorithm.

For other algorithms, in particular those that use divide-and-conquer tactics, you will find that you need a stack to keep track of some thing in the iterative solution. In those cases, a recursive version is much cleaner, because the stack handling becomes implicit.

share|improve this answer

Recursion is slower and it consumes more memory since it can fill up the stack. But there is a work-around called tail-call optimization which requires a little more complex code (since you need another parameter to the function to pass around) but is more efficient since it doesn't fill the stack.

Unfortunately C# does not support this so I would suggest you to use recursion with care, because you might get stack overflow for large values.

As a conclusion, in functional languages you only have recursion and have to deal with it; but in imperative languages iteration is more natural and more efficient. The majority of the most-used imperative languages either does not support (Java, C#, Python) or does not guarantee (C, C++) tail call optimization.

share|improve this answer
    
Yes, I've ran into a couple of stack overflows because of using recursion "carelessly" .. –  Overly Excessive May 4 at 14:14
3  
As far as I know, a clever implementation of tail-call optimization will run as fast as a while loop using local variables. The difference is that using recursion you have to specify inputs and outputs to each iteration explicitly, which makes it easier to write correct code. Also, while the underlying implementation can still use side-effects for efficiency, at the level of your programming language semantics you can write purely functional (side-effect free) code. –  Giorgio May 4 at 14:40
1  
@m3th0dman, in this, the second decade of the 21st century, pretty much ALL production-quality compilers support tail-call optimization. There was a StackOverflow question about strange 8051 code generation (may have been PIC), that turned out to be tail-call optimization. Read the various "Lambda: The Ultimate ..." papers from the MIT AI Lab, from decades ago. Tail-call optimization is in fact a very straightforward code generator optimization. –  John R. Strohm May 5 at 2:51
2  
@JohnR.Strohm What claim exactly do you want to support with that reference? I did not say TCO can't happen or can't be very effective. I supported m3th0dman's claim that many imperative languages don't have TCO by given numerous examples. Regardless of how effective TCO is, or how easy it might be to implement, if it isn't actually implemented users of those languages don't benefit from TCO. I see nothing in the Steele paper arguing against that. –  delnan May 5 at 6:17
1  
@Giorgio Tail recursion elimination is easy enough, and indeed implemented. TCO for arbitrary callees would require full trampolining everywhere, which kills Java interop (extremely important for both) and is very slow. You can't pull local tricks like jumping to the target code without setting up another stack frame because the JVM won't allow it. There's no cross-method jump, only the invoke* opcodes which always set up the call frame for you. –  delnan May 6 at 19:20

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.