It seems to me like people dislike a goto statement a lot, so I felt the need to straighten this out a bit.
I believe the 'emotions' people have about goto eventually boil down to understanding of code and (misconceptions) about possible performance implications. Before answering the question, I will therefore first go into some of the details on how it's compiled.
As we all know, C# is compiled to IL, which is then compiled to assembler using an SSA compiler. I'll give a bit of insights into how this all works, and then try to answer the question itself.
From C# to IL
First we need a piece of C# code. Let's start simple:
foreach (var item in array)
{
// ...
break;
// ...
}
I'll do this step by step to give you a good idea of what happens under the hood.
First translation: from foreach to the equivalent for loop (Note: I'm using an array here, because I don't want to get into details of IDisposable -- in which case I'd also have to use an IEnumerable):
for (int i=0; i<array.Length; ++i)
{
var item = array[i];
// ...
break;
// ...
}
Second translation: the for and break is translated into an easier equivalent:
int i=0;
while (i < array.Length)
{
var item = array[i];
// ...
break;
// ...
++i;
}
And third translation (this is the equivalent of the IL code): we change break and while into a branch:
int i=0; // for initialization
startLoop:
if (i >= array.Length) // for condition
{
goto exitLoop;
}
var item = array[i];
// ...
goto exitLoop; // break
// ...
++i; // for post-expression
goto startLoop;
While the compiler does these things in a single step, it gives you insight into the process. The IL code that evolves from the C# program is the literal translation of the last C# code. You can see for yourself here: https://dotnetfiddle.net/QaiLRz (click 'view IL')
Now, one thing you have observed here is that during the process, the code becomes more complex. The easiest way to observe this is by the fact that we needed more and more code to ackomplish the same thing. You might also argue that foreach, for, while and break are actually short-hands for goto, which is partly true.
From IL to Assembler
The .NET JIT compiler is an SSA compiler. I won't go into all the details of SSA form here and how to create an optimizing compiler, it's just too much, but can give a basic understanding about what will happen. For a deeper understanding, it's best to start reading up on optimizing compilers (I do like this book for a brief introduction: http://ssabook.gforge.inria.fr/latest/book.pdf ) and LLVM (llvm.org).
Every optimizing compiler relies on the fact that code is easy and follows predictable patterns. In the case of FOR loops, we use graph theory to analyze branches, and then optimize things like cycli in our branches (e.g. branches backwards).
However, we now have forward branches to implement our loops. As you might have guessed, this is actually one of the first steps the JIT is going to fix, like this:
int i=0; // for initialization
if (i >= array.Length) // for condition
{
goto endOfLoop;
}
startLoop:
var item = array[i];
// ...
goto endOfLoop; // break
// ...
++i; // for post-expression
if (i >= array.Length) // for condition
{
goto startLoop;
}
endOfLoop:
// ...
As you can see, we now have a backward branch, which is our little loop. The only thing that's still nasty here is the branch that we ended up with due to our break statement. In some cases, we can move this in the same way, but in others it's there to stay.
So why does the compiler do this? Well, if we can unroll the loop, we might be able to vectorize it. We might even be able to proof that there's just constants being added, which means our whole loop could vanish into thin air. To summarize: by making the patterns predictable (by making the branches predictable), we can proof that certain conditions hold in our loop, which means we can do magic during the JIT optimization.
However, branches tend to break those nice predictable patterns, which is something optimizers therefore kind-a dislike. Break, continue, goto - they all intend to break these predictable patterns- and are therefore not really 'nice'.
You should also realize at this point that a simple foreach is more predictable then a bunch of goto statements that go all over the place. In terms of (1) readability and (2) from an optimizer perspective, it's both the better solution.
Another thing worth mentioning is that it's very relevant for optimizing compilers to assign registers to variables (a process called register allocation). As you might know, there's only a finite number of registers in your CPU and they are by far the fastest pieces of memory in your hardware. Variables used in code that's in the inner-most loop, are more likely to get a register assigned, while variables outside of your loop are less important (because this code is probably hit less).
Help, too much complexity... what should I do?
The bottom line is that you should always use the language constructs you have at your disposal, which will usually (implictly) build predictable patterns for your compiler. Try to avoid strange branches if possible (specifically: break, continue, goto or a return in the middle of nothing).
The good news here is that these predictable patterns are both easy to read (for humans) and easy to spot (for compilers).
One of those patterns is called SESE, which stands for Single Entry Single Exit.
And now we get to the real question.
Imagine that you have something like this:
// a is a variable.
for (int i=0; i<100; ++i)
{
for (int j=0; j<100; ++j)
{
// ...
if (i*j > a)
{
// break everything
}
}
}
The easiest way to make this a predictable pattern is to simply eliminate the if completely:
int i, j;
for (i=0; i<100 && i*j <= a; ++i)
{
for (j=0; j<100 && i*j <= a; ++j)
{
// ...
}
}
In other cases you can also split the method into 2 methods:
// Outer loop in method 1:
for (i=0; i<100 && processInner(i); ++i)
{
}
private bool processInner(int i)
{
int j;
for (j=0; j<100 && i*j <= a; ++j)
{
// ...
}
return i*j<=a;
}
Temporary variables? Good, bad or ugly?
You might even decide to return a boolean from within the loop (but I personally prefer the SESE form because that's how the compiler will see it and I think it's cleaner to read).
Some people think it's cleaner to use a temporary variable, and propose a solution like this:
bool more = true;
for (int i=0; i<100; ++i)
{
for (int j=0; j<100; ++j)
{
// ...
if (i*j > a) { more = false; break; } // yuck.
// ...
}
if (!more) { break; } // yuck.
// ...
}
// ...
I personally am opposed to this approach. Look again on how the code is compiled. Now think about what this will do with these nice, predictable patterns. Get the picture?
Right, let me spell it out. What will happen is that:
- The compiler will write out everything as branches.
- As an optimization step, the compiler will do data flow analysis in an attempt to remove the strange
more variable that only happens to be used in control flow.
- If succesful, the variable
more will be eliminated from the program, and only branches remain. These branches will be optimized, so you will get only a single branch out of the inner loop.
- If unsuccesful, the variable
more is definitely used in the inner-most loop, so if the compiler won't optimize it away, it has a high chance to be allocated to a register (which eats up valuable register memory).
So, to summarize: the optimizer in your compiler will go into a hell of a lot of trouble to figure out that more is only used for the control flow, and in the best case scenario will translate it to a single branch outside of the outer for loop.
In other words, the best case scenario is that it will end up with the equivalent of this:
for (int i=0; i<100; ++i)
{
for (int j=0; j<100; ++j)
{
// ...
if (i*j > a) { goto exitLoop; } // perhaps add a comment
// ...
}
// ...
}
exitLoop:
// ...
My personal opinion on this is quite simple: if this is what we intended all along, let's make the world easier for both the compiler and readability, and write that right away.
tl;dr:
Bottom line:
- Use a simple condition in your for loop if possible. Stick to the high-level language constructs you have at your disposal as much as possible.
- If everything fails and you're left with either
goto or bool more, prefer the former.
