Help with pointers

[slippy]

Why does the address &ra[0] keep changing?

void overlappingPermutations(Distribution& D, int* ra) {

  // overlappingPermutations
  //
  //      Take five consecutively-generated RNs at a time and
  //      determine their order. There are 5P5 = 120 different
  //      orderings.
  //
  //
  // Example:
  //
  //
  //
  //
  // Expected distribution: Flat
  //
  // Complexity: O(N)
 vector<int> pattern(120);
  int is[5] = {0};

  while(ra) {
    int i = 0;
    while(i < 5){
      is[i++] = *ra++;
    }
    if(ra - &ra[0] == 0) cout << "bang!\n"; // Always written!
    rankPermutations(pattern, is, 5);
    }

// Rest of implementation comes later

}

// ... 

int main(void) {

  Distribution* D = new Distribution[12];
  bool running = true;
  int* ra;


  try {

    ra = new int[10000000];

  } catch (bad_alloc) { 

    cout << "Memory allocation failed\n";
    return 127;

  }

  long int count = 0;

  seedRNG();

  while(running) {


    ra[count++] = rand();

    if(count % 1000000 == 0) cout << count << " values so far.\n";
    if(count == 10000000) running = false;
  }

  diehard(ra); // Diehard just calls functions like `overlappingPermutations', 
                      // which all have the same prototype.
}

Thanks.

[anon]

ra and &ra[0] should always be equal. ra is not a magic array whose start won't change and a magic pointer that you can make point elsewhere at the same time.

[laserlight]

I suggest that you post the smallest and simplest compilable program that demonstrates the problem.

I noticed a few problems:

1. You use new[] in main(), but it lacks a matching delete[].

2. You rather randomly return 127 from main(). You should return 0, EXIT_SUCCESS or perhaps in this case, EXIT_FAILURE. The latter two are found in <cstddef>.

3. You might want to indent by 4 spaces instead. 2 spaces (or at one point, it was one space) can be rather... insignificant.

Also, I noticed that:

while(running) {


  ra[count++] = rand();

  if(count &#37; 1000000 == 0) cout << count << " values so far.\n";
  if(count == 10000000) running = false;
}

Is far more simply and directly expressed as:

while (count < 10000000) {
    ra[count++] = rand();

    if(count % 1000000 == 0) cout << count << " values so far.\n";
}

[slippy]

ra and &ra[0] should always be equal. ra is not a magic array whose start won't change and a magic pointer that you can make point elsewhere at the same time.

But doesn't ra++ move the pointer up by sizeof(int)? I thought &ra[0] points to the zeroth element (always), and ra is the current position. Does ra++ move &ra[0] up too? Why?

[Codeplug]

ra[0] points to the zeroth element of the array starting at ra.
ra++ does increment the pointer - and ra[0] points the zero'th element starting at ra.

ra is just a pointer sized variable. The compiler does not generate code to keep track of ra's initial value - as the "fixed zero'th element".

gg

[grumpy]

ra is a pointer, which (in this context) should be equivalent to &ra[0], no matter what you've done to ra previously. Which means that ra-&ra[0] should always compare equal to zero.

If you do this;

    int *rb = new int[5];
    int *ra = rb;
    ++ra;

then ra will be equal to rb+1. Manipulating ra[0] (or, equivalently, *ra) will therefore manipulate rb[1]. So &ra[0] == &rb[1] will always compare true.

[slippy]

Thanks for your help!

I traced the problem to the way values are assigned to an array in a structure:

#include <iostream>
#include <vector>
#include <string>
#include <cstdlib>
#include <ctime>
#include <cstddef>

using namespace std;

typedef struct {

  string d;   // Statistical distribution

  int n;      // Number of entries

  double* dx; // x-values
  double* dy; // y-values
  double* dz; // z-values

} Distribution;

void seedRNG() {

  srand(static_cast<unsigned>(time(0)));

}

void fill(Distribution& D, double x) {

  *(D.dx + D.n) = x; // crash!!!!!

  D.n++;
}

double l2Norm(double p1, double p2) {

  return p1 - p2 >= 0 ? p1 - p2 : p2 - p1;

}

void poissonL2Norm(Distribution& D, int* ra) {

  // poissonL2Norm
  //
  //      Throw beans on a closed interval of width RAND_MAX+1
  //      and measure the distance between adjacent beans:
  //
  // Example:
  //           0                       RAND_MAX
  //           [--B---BB--- ... ---B--B]
  //              |< >||<         >|<>|
  //
  //    Width:    3   1  RAND_MAX-7 2
  //
  // Expected distribution: Poisson
  //
  // Efficiency: O(n)

  //    while(*(1+ra++));
  fill(D, l2Norm(*ra, *(ra+1)) );

}

void diehard(int* ra, Distribution* D  =0) {

  poissonL2Norm(D[0], ra);
  

}

int main(void) { // Thanks, laserlight

  Distribution* D = new Distribution[12];
  int* ra;


  try {
    ra = new int[10000000];

  } catch (bad_alloc) { // Handling not mandatory, but safe
    cout << "Memory exhausted\n";
    return EXIT_FAILURE;
  }

  long int count = 0;

  seedRNG();

  cout << "start: " << ra << '\n';
  while(count < 10000000) {

    ra[count++] = rand();
    if(count % 1000000 == 0) cout << count << " values so far.\n";
  }


  diehard(ra, D);

  delete[] ra;
  delete[] D;

}

This program goes up in flames at runtime because of the way I "fill" the distribution. I just want to add one number at a time to D.dx, but I'm having trouble doing this. Any ideas? This is my first 1000-liner in C++, and it looks like I'll be 100 by the time it works completely : (

Also: Emacs' syntax highlighter for C++ is only indenting two deep. How do you tell it to go four deep? I'm so, so lost here : (

[grumpy]

You are incorrectly assuming, when you create a pointer, that it points at something valid.

For example;

int main()
{
    int *p;
    *(p + 4) = 5;   // undefined behaviour here
}

yields undefined behaviour, for which a common (but not guaranteed) symptom is a program crash, as p is uninitialised. It does NOT point at the first element of any array.

Furthermore, initialising to NULL (or zero), also yield undefined behaviour.

#include <stdlib.h>
int main()
{
    int *p = NULL;
    *(p + 4) = 5;   // undefined behaviour
}

The bottom line is that YOU need to ensure the pointer points at something valid.

int main()
{
    int *p = new int[5];
    *(p + 4) = 5;    //  OK, as p points at an array of 5 elements
    delete [] p;
    *(p + 4) = 10;   // undefined behaviour here, as the memory pointed at by p has been released by the program
}

[slippy]

It does NOT point at the first element of any array.
...
The bottom line is that YOU need to ensure the pointer points at something valid.

Thanks!