Thanks to some segmentation faults and warnings in valgrind, I found that this code is incorrect and has some sort of dangling reference in the for-range loop.
#include<numeric>
#include<vector>
auto f(){
std::vector<std::vector<double>> v(10, std::vector<double>(3));
iota(v[5].begin(), v[5].end(), 0);
return v;
}
int main(){
for(auto e : f()[5])
std::cout << e << std::endl;
return 0;
}
It looks as if the begin
and end
is taken from a temporary and lost in the loop.
Of course, a way around is to do
auto r = f()[5];
for(auto e : r)
std::cout << e << std::endl;
However, I wonder exactly why for(auto e : f()[5])
is an error and also if there is a better way around or some way to design f
or the even the container (std::vector
) to avoid this pitfall.
With iterator loops is more obvious why this problem happens (begin
and end
come from different temporary objects)
for(auto it = f()[5].begin(); it != f()[5].end(); ++it)
But in a for-range loop, as in the first example, it seems very easy to make this mistake.
Note that using a temporary as the range expression directly is fine, its lefetime will be extended. But for f()[5]
, what f()
returns is the temporary and it's constructed within the expression, and it'll be destroyed after the whole expression where it's constructed.
From C++20, you can use init-statement for range-based for loop to solve such problems.
(emphasis mine)
If range_expression returns a temporary, its lifetime is extended until the end of the loop, as indicated by binding to the rvalue reference __range, but beware that the lifetime of any temporary within range_expression is not extended.
This problem may be worked around using init-statement:
for (auto& x : foo().items()) { /* .. */ } // undefined behavior if foo() returns by value for (T thing = foo(); auto& x : thing.items()) { /* ... */ } // OK
e.g.
for(auto thing = f(); auto e : thing[5])
std::cout << e << std::endl;