C++: Different statement scopes vs. Same statement scope

Question

I can't find any practical use for this, but I'm still curious as to why I'm getting different results when run in the same statement scope as apposed to different statement scopes?

Variation #1

int num = 3;

vector<int> ivec;
vector<int*> ipvec;

for (int i = 0; i != num; ++i)
{
    ivec.push_back(i);
    ipvec.push_back(&ivec[i]);
}

for (int i = 0; i != num; ++i)
{
    cout << "ivec:\t" << ivec[i] << endl;
    cout << "ipvec:\t" << *ipvec[i] << endl;
}

Output

ivec:   0
*ipvec: -572662307
ivec:   1
*ipvec: -572662307
ivec:   2
*ipvec: 2

Variation #2

int num = 3;

vector<int> ivec;
vector<int*> ipvec;

for (int i = 0; i != num; ++i)
    ivec.push_back(i);

for (int i = 0; i != num; ++i)
    ipvec.push_back(&ivec[i]);

for (int i = 0; i != num; ++i)
{
    cout << "ivec:\t" << ivec[i] << endl;
    cout << "*ipvec:\t" << *ipvec[i] << endl;
}

Output

ivec:   0
*ipvec: 0
ivec:   1
*ipvec: 1
ivec:   2
*ipvec: 2

**Variation #1** may exhibit undefined behavior, as `ivec.push_back(i)` may trigger reallocation and invalidate pointers to elements of `ivec`, leaving `ipvec` full of dangling pointers. — Igor Tandetnik, Nov 10 '17 at 21:54
Hint: What does `push_back` do if the vector doesn't have space reserved? — StoryTeller - Unslander Monica, Nov 10 '17 at 21:54
Also, that number you are getting isn't completely random. It's `0xffffffffdddddddd`. Almost as if something wanted to mark a memory location as no longer accessible... — StoryTeller - Unslander Monica, Nov 10 '17 at 21:56

Garrigan Stafford · Accepted Answer · 2017-11-10T22:34:34.407

4

So std::vector allocates memory in and resizes based off input. So in variation 1 you cross over the first allocation size (2 in this case). So your pointers for [0] and [1] are no longer valid since the array inside the vector has been reallocated and now has a different address. In variation 2 you don't have this problem since you have already allocated the full list of the vector, so all the pointers you add are valid. This is not an issue of scope, but instead from the fact std::vector allocates the internal array based on powers of 2 when the list is growing. ( hence why [2] works in both iterations (since the internal array is of size 4 at the end).

edited Nov 10 '17 at 22:34

answered Nov 10 '17 at 22:00

Garrigan Stafford

1,331
9
20

2

allocations in sizes of powers of 2 is common in implementations but not really an essential feature of `std::vector`. It could be powers of 3 as well – 463035818_is_not_an_ai Nov 10 '17 at 22:01
1

I've also seen approximations of a constant times powers of 1.5. – aschepler Nov 10 '17 at 22:03
...not saying that your answer is wrong, just that one might understand it as `std::vector` always using powers of 2 for allocations, which is not the case – 463035818_is_not_an_ai Nov 10 '17 at 22:03
As is explained here: https://stackoverflow.com/questions/11560904/vector-memory-allocation-strategy , Your statement about `std::vector` always allocating in powers of 2 is incorrect. – Algirdas Preidžius Nov 10 '17 at 22:04
I edited above to remove the power of 2 specification – Garrigan Stafford Nov 10 '17 at 22:35
Also I would like to note that code should not depend on the allocation size of vector. The base idea of resizing off of exponent (like powers of 2) is to get O(1) amortized insertion time. – Garrigan Stafford Nov 10 '17 at 22:37

C++: Different statement scopes vs. Same statement scope

Variation #1

Output

Variation #2

Output

1 Answers1