What could be reason it crashes when I use vector::erase? - c++

Question

I am trying to do some operation on vector. And calling erase on vector only at some case.
here is my code
while(myQueue.size() != 1)
{
vector<pair<int,int>>::iterator itr = myQueue.begin();
while(itr != myQueue.end())
{
if(itr->first%2 != 0)
myQueue.erase(itr);
else
{
itr->second = itr->second/2;
itr++;
}
}
}
I am getting crash in 2nd iteration.And I am getting this crash with message vector iterator incompatible .
What could be the reason of crash?

Answer 1

If erase() is called the iterator is invalidated and that iterator is then accessed on the next iteration of the loop. std::vector::erase() returns the next iterator after the erased iterator:
itr = myQueue.erase(itr);

Answer 2

Given an iterator range [b, e) where b is the beginning and e one past the end of the range for a vector an erase operation on an iterator i somewhere in the range will invalidate all iterators from i upto e. Which is why you need to be very careful when calling erase. The erase member does return a new iterator which you can you for subsequent operations and you ought to use it:
itr = myQueue.erase( itr );
Another way would be to swap the i element and the last element and then erase the last. This is more efficient since less number of moves of elements beyond i are necessary.
myQueue.swap( i, myQueue.back() );
myQueue.pop_back();
Also, from the looks of it, why are you using vector? If you need a queue you might as well use std::queue.

Answer 3

That is undefined behavior. In particular, once you erase an iterator, it becomes invalid and you can no longer use it for anything. The idiomatic way of unrolling the loop would be something like:
for ( auto it = v.begin(); it != v.end(); ) {
if ( it->first % 2 != 0 )
it = v.erase(it);
else {
it->second /= 2;
++it;
}
}
But then again, it will be more efficient and idiomatic not to roll your own loop and rather use the algorithms:
v.erase( std::remove_if( v.begin(),
v.end(),
[]( std::pair<int,int> const & p ) {
return p.first % 2 != 0;
}),
v.end() );
std::transform( v.begin(), v.end(), v.begin(),
[]( std::pair<int,int> const & p ) {
return std::make_pair(p.first, p.second/2);
} );
The advantage of this approach is that there is a lesser number of copies of the elements while erasing (each valid element left in the range will have been copied no more than once), and it is harder to get it wrong (i.e. misuse an invalidated iterator...) The disadvantage is that there is no remove_if_and_transform so this is a two pass algorithm, which might be less efficient if there is a large number of elements.

Answer 4

Iterating while modifying a loop is generally tricky.
Therefore, there is a specific C++ idiom usable with non-associative sequences: the erase-remove idiom.
It combines the use of the remove_if algorithm with the range overload of the erase method:
myQueue.erase(
std::remove_if(myQueue.begin(), myQueue.end(), /* predicate */),
myQueue.end());
where the predicate is expressed either as a typical functor object or using the new C++11 lambda syntax.
// Functor
struct OddKey {
bool operator()(std::pair<int, int> const& p) const {
return p.first % 2 != 0;
}
};
/* predicate */ = OddKey()
// Lambda
/* predicate */ = [](std::pair<int, int> const& p) { return p.first % 2 != 0; }
The lambda form is more concise but may less self-documenting (no name) and only available in C++11. Depending on your tastes and constraints, pick the one that suits you most.
It is possible to elevate your way of writing code: use Boost.Range.
typedef std::vector< std::pair<int, int> > PairVector;
void pass(PairVector& pv) {
auto const filter = [](std::pair<int, int> const& p) {
return p.first % 2 != 0;
};
auto const transformer = [](std::pair<int, int> const& p) {
return std::make_pair(p.first, p.second / 2);
};
pv.erase(
boost::transform(pv | boost::adaptors::filtered( filter ),
std::back_inserter(pv),
transformer),
pv.end()
);
}
You can find transform and the filtered adaptor in the documentation, along with many others.