weirdest bug ever - int definition causes sigsegv. stack overflow?

Question

I'm using the BSPlib and when adding a simple definition of "int i" on the function running on multiple threads (along with many other) I get a message like "Process 2 caught SIGNAL 11 Segmantation fault ". it's important to note that I checked alot and without it I don't get the segmentation fault and with it I get it almost all the time. how could int definition cause it? is there a stack overflow I might have caused? thanks.

int P;
int main( int argc, char* argv[] )
{
    /** sequentail - process 0 */
    P=bsp_nprocs(); /// maximum number of process avialble (must do that on sequential part ,need for bsp begin)
    bsp_begin(P);
    char* str1;
    char* str2;
    int n;
    int** table;
    int thread=bsp_pid();
    int num_threads=bsp_nprocs();
    if(thread == 0)
    {
        ifstream file1(argv[1]);
        ifstream file2(argv[2]);
        // check if the strings are the same size RDBG
        string string1((istreambuf_iterator<char>(file1)), istreambuf_iterator<char>());
        string string2((istreambuf_iterator<char>(file2)), istreambuf_iterator<char>());
        n=string1.length();
        str1= (char*)malloc(sizeof(char)*(n+1));
        str2= (char*)malloc(sizeof(char)*(n+1));
        strcpy(str1,string1.c_str());
        strcpy(str2,string2.c_str());
    }
    if (thread!=0)
    {
        str1= (char*)malloc(sizeof(char)*(n+1));
        str2= (char*)malloc(sizeof(char)*(n+1));
    }
    bsp_push_reg(&n,SZINT);
    bsp_sync();
    bsp_get(0,&n,0,&n,SZINT);
    bsp_sync();
    if (thread==0)
    {
        table=(int**)malloc(sizeof(int)*(n+1));
        for (int i=0; i<n+1; i++)
            table[i]=(int*)malloc(sizeof(int)*(n+1));
    }
    bsp_push_reg(str1,SZCHAR*(n+1));
    bsp_push_reg(str2,SZCHAR*(n+1));
    bsp_push_reg(table,n*n*SZINT);
    bsp_sync();
    if (thread==0)
    {
        for(int t=1; t<num_threads; t++)
            for (int k=0; k<=n; k++)
            {
                bsp_put(t,str1+k,str1,k*SZCHAR,SZCHAR);
                bsp_put(t,str2+k,str2,k*SZCHAR,SZCHAR);
            }
    }
    bsp_sync();
    cout << thread << "!!!" << str1 << ";" << str2 << endl;
    int i;
    bsp_sync();
    bsp_pop_reg(table);
    bsp_pop_reg(str2);
    bsp_pop_reg(str1);
    bsp_pop_reg(&n);
    bsp_sync();
    free(str1);
    free(str2);
    bsp_sync();
    bsp_end();
    return 0;
}

Answer 1

Your declaration/initialization of the table variable is incorrect. You are initializing it as an array of arrays (i.e. as n+1 distinct memory blocks), whereas you are telling bsplib that it is a a contiguous memory block of n*n ints. You either need to change your allocation or the registration.

As a consequence, bsplib overwrites memory that isn't initialized at all.

Answer 2

In the vast majority of cases where a semmingly innocuous change causes or fixes a problem, you have what's known as a Heisenbug. In that case, it's not the actual change that's the root cause, the change is just a catalyst to cause the true bug to surface.

I'm not entirely certain how BSPlib does its threads but it appears to me that the n value is not initialised for the case where thread is non-zero.

In other words, the value is set to the length of string1 only for thread == 0 yet it is used to malloc space for thread !=0, said space being dependent on whatever rubbish happened to be on the stack.

Answer 3

    string string1((istreambuf_iterator<char>(file1)), istreambuf_iterator<char>());
    string string2((istreambuf_iterator<char>(file2)), istreambuf_iterator<char>());
    n=string1.length();
    str1= (char*)malloc(sizeof(char)*(n+1));
    str2= (char*)malloc(sizeof(char)*(n+1));
    strcpy(str1,string1.c_str());
    strcpy(str2,string2.c_str());

What happens when string2 is longer than string1? Is this ever the case? You are using malloc to allocate the size of string1 in str2. If string2 is longer than string1, you will buffer overflow, probably clobbering all kinds of stuff in memory.

Should you be doing n1 = string1.length(); n2 = string2.length();?