Variable reuse in C

Question

The code I'm looking at is this:

for (i = 0; i < linesToFree; ++i ){
    printf("Parsing line[%d]\n", i);

    memset( &line, 0x00, 65 );
    strcpy( line, lines[i] );

    //get Number of words:
    int numWords = 0;

    tok = strtok(line , " \t");
    while (tok != NULL) {
        ++numWords;
        printf("Number of words is:  %d\n", numWords);
        println(tok);

        tok = strtok(NULL, " \t");
    }
}

My question centers around the use of numWords. Does the runtime system reuse this variable or does it allocate a new int every time it runs through the for loop? If you're wondering why I'm asking this, I'm a Java programmer by trade who wants to get into HPC and am therefore trying to learn C. Typically I know you want to avoid code like this, so this question is really exploratory.

I'm aware the answer is probably reliant upon the compiler... I'm looking for a deeper explanation than that. Assume the compiler of your choice.

Answer 1

This is what is called "block", "automatic" or "local" scope in C. It is a form of lexical scoping, i.e., a name refers to its local environment. In C, it is top down, meaning that it happens as the file is parsed and compiled and visible only after defined in the program. When the variable goes out of scope, the lexical name is no longer valid (visible) and the memory may be reused.

The variable is declared in a local scope or a block defined by curly braces { /* block */ }. This defines a whole group of C and C99 idioms, such as:

for(int i=0; i<10; ++i){    // C99 only. int i is local to the loop
    // do something with i 
}  // i goes out of scope here...

There are subtleties, such as:

 int x = 5;
 int y = x + 10; // this works

 int x = y + 10;
 int y = 5;      // compiler error     

and:

int g;        // static by default and init to 0
extern int x; // defined and allocated elsewhere - resolved by the linker 
int main (int argc, const char * argv[])
{
    int j=0;  // automatic by default  
    while (++j<=2) {
        int i=1,j=22,k=3;    // j from outer scope is lexically redefined
        for (int i=0; i<10; i++){
            int j=i+10,k=0;
            k++;             // k will always be 1 when printed below
            printf("INNER: i=%i, j=%i, k=%i\n",i,j,k);
        }
        printf("MIDDLE: i=%i, j=%i, k=%i\n",i,j,k);   // prints middle j
    }

    // printf("i=%i, j=%i, k=%i\n",i,j,k); compiler error
    return 0;
}

There are idiosyncrasies:

1. In K&R C, ANSI C89, and Visual Studio, All variables must be declared at the beginning of the function or compound statement (i.e., before the first statement)
2. In gcc, Variables may be declared anywhere in the function or compound statement and is only visible from that point on.
3. In C99 and C++, Loop variables may be declared in for statement and are visible until end of loop body.
4. In a loop block, the allocation is performed ONCE and the RH assignment (if any) is performed each time.

In the particular example you posted, you enquired about int numWords = 0; and if a new int is allocated each time through the loop. No, there is only one int allocated in a loop block, but the right hand side of the = is executed every time. This can be demonstrated so:

#include <stdio.h>
#include <time.h>
#include <unistd.h>

volatile time_t ti(void){
    return time(NULL);
}

void t1(void){
    time_t t1;
    for(int i=0; i<=10; i++){
        time_t t2=ti();    // The allocation once, the assignment every time
        sleep(1);
        printf("t1=%ld:%p t2=%ld:%p\n",t1,(void *)&t1,t2,(void *)&t2);
    }
}

Compile that with any gcc (clang, eclipse, etc) compatible compiler with optimizations off (-O0) or on. The address of t2 will always be the same.

Now compare with a recursive function:

int factorial(int n) {
    if(n <= 1)
        return 1;
    printf("n=%i:%p\n",n,(void *)&n);
    return n * factorial(n - 1);
}

The address of n will be different each time because a new automatic n is allocated with each recursive call.

Compare with an iterative version of factorial forced to used a loop-block allocation:

int fac2(int num) {
    int r=0;   // needed because 'result' goes out of scope
    for (unsigned int i=1; i<=num; i++) {
        int result=result*i;   // only RH is executed after the first time through
        r=result;
        printf("result=%i:%p\n",result,(void *)&result);  // address is always the same
    }
    return r;
}

In conclusion, you asked about int numWords = 0; inside the for loop. The variable is reused in this example.

The way the code is written, the programmer is relying on the RH of int numWords = 0; after the first to be executed and resetting the variable to 0 for use in the while loop that follows.

Answer 2

It will be allocated every time through the loop (the compiler can optimize out that allocation)

for (i = 0; i < 100; i++) {
    int n = 0;
    printf("%d : %p\n", i, (void*)&n);
}

No guarantees all 100 lines will have the same address (though probably they will).

Edit: The C99 Standard, in 6.2.4/5 says: "[the object] lifetime extends from entry into the block with which it is associated until execution of that block ends in any way." and, in 6.8.5/5, it says that the body of a for statement is in fact a block ... so the paragraph 6.2.4/5 applies.

Answer 3

Please note the difference between automatic and dynamic memory allocation. In Java only the latter exists.

This is automatic allocation:

int numWords = 0;

This is dynamic allocation:

int *pNumWords = malloc(sizeof(int));
*pNumWords = 0;

The dynamic allocation in C only happens explicitly (when you call malloc or its derivatives).

In your code, only the value is set to your variable, no new one is allocated.

Answer 4

From a performance standpoint, it's not going to matter. (Variables map to registers or memory locations, so it has to be reused.)

From a logical standpoint, yes, it will be reused because you declared it outside the loop.

From a logical standpoint:

• numWords will not be reused in the outer loop because it is declared inside it.
• numWords will be reused in the inner loop because it isn't declared inside.

Answer 5

The scope of the numWords variable is inside the for loop. Just as Java, you can only use the variable inside the loop, so theoretically its memory would have to be freed on exit - since it is also on the stack in your case.

Any good compiler however would use the same memory and simply re-set the variable to 0 on each iteration.

If you were using a class instead of an int, you would see the destructor being called every time the for loops.

Even consider this:

class A;

A* pA = new A;
delete pA;
pA = new A;

The two objects created here will probably reside at the same memory.

Answer 6

Your conception about how this works in Java might be misinformed - Java doesn't "allocate" a new int every time through a loop like that either. Primitive type variables like int aren't allocated on the Java heap, and the compiler will reuse the same local storage for each loop iteration.

On the other hand, if you call new anything in Java every time through a loop, then yes, a new object will be allocated every time. However, you're not doing that in this case. C also won't allocate anything from the heap unless you call malloc or similar (or in C++, new).