Segfault on C++ map iterator not out of bounds

Question

I have the following code :

iterator = u.U.begin(); //u.U is a map<bitset,long>
int sizeofIterator = u.U.size();

for ( int pp = 0; pp<sizeofIterator; pp++ ){

    double *faux = new double[q];

    for (int k = 0 ; k<q ; k++){
        faux[k]=1;
    }

    for (int k=0; k<q ; k++){

        for (int i=0; i<I ; i++){
            faux[k]= faux[k]*successProbability(theta[k],a[i],b[i],c[i]); 
        }
        faux[k] = faux[k]*nodeWeight[k];

    }

    double sum = 0;

    for (int k=0; k<q ; k++){
        sum += faux[k];
    }

    for (int k=0; k<q; k++){

        faux[k] = faux[k]/sum;
        faux[k] = iterator->second * faux[k];

        f[k] = faux[k]+f[k];

        for(int i = 0; i<I; i++){
            if(iterator->first[i]){
                double aux = faux[k];
                r[k*I+i]=r[k*I+i]+aux; 
            }
        }
    }
     iterator++;
}

When that code is run, I've got a segmentation fault. I've got some conclusions and I know that the problem is in the line that's got

    iterator++;

The iterator has to iterate 32 times (the size of the map) but the segfault occurs in the 18th iteration

TIP: When I comment the aux in line: r[k*I+i]=r[k*I+i]+aux;

i.e. I put the following r[k*I+i]=r[k*I+i]+0;//+aux;

All works (But I need to perform that sum, surely!)

¿Do you know why? ¿Can you help me please?

Thanks in advance

EDIT : We could not replicate the error on less code, it simply works, but is not the code we want to work :(

Can you print out the value of `aux` during each iteration, does that work? — Chris O, Nov 22 '13 at 03:06
My guess is: Look for the field `r`! I think the segment violation occurs there. The reasoning is that if you comment `aux` out then there is nothing happening in the loop and maybe the compiler optimizes something out. Furthermore, the corresponding block can be ignored depending on `iterator`. There is the connection with `iterator`. I guess the index is out of range. Print `k*I+i` before you use it and compare it to the bounds of `r`. — Tobias, Nov 22 '13 at 03:21

score 0 · Accepted Answer · answered Nov 22 '13 at 03:44

TIP: When I comment the aux in line: r[k*I+i]=r[k*I+i]+aux;

i.e. I put the following r[k*I+i]=r[k*I+i]+0;//+aux; All works (But I need to perform that sum, surely!)

This is called voodoo debugging, and is also associated with your inability to geenrate a smaller example. Basically you have, at some point, corrupted either your stack or heap (or both). Sometimes after corrupting these, you can continue, but have other function call be fragile. You error is not in the map or the iterator. There is little cause to believe the error is even in this function.

This leads to spurious and misleadding errors (serach for voodoo debugging on the web to find out more). I recommend a memory tool like valgrind, which might help.

Else, another apporach might be to put a full map traversal loop into a single line statement, and see at which point your structure becomes invalid, since above you are finding out its invalid when you go to use it, but its probably been invalid for a while.

note in the above code, you haven't given us I, which will break the code if too big, and you havent told us above nodeWeight, theta or F.

btw, if I had to bet on stack vs heap corruption, I would bet stack. Adding extra terms to an equation (as per your tip) creates and destroys temporaries, which is usually a stack thing. So if doing it makes a difference, they are often being create in the wrong place or similar.

Many thanks, i think this is hiighly plausible , will look into it. — Juan Liberato, Nov 22 '13 at 11:43

score 0 · Answer 2 · answered Nov 22 '13 at 20:14

0

SOLVED: Error was in the line

f[k] = faux[k]+f[k];

Because the f vector was not initialized.

THANKS ALL!

answered Nov 22 '13 at 20:14

As Gest user

1

So, writing to f was invalid, but you were unlucky enough that it merely corrupted your state rather than crashing immediately. Corrupted state has weird behavior and crashes in unrelated locations and in response to small seemingly unrelated changes in code. – RichardPlunkett Nov 23 '13 at 23:59

Segfault on C++ map iterator not out of bounds

The iterator has to iterate 32 times (the size of the map) but the segfault occurs in the 18th iteration

All works (But I need to perform that sum, surely!)

2 Answers2