Consider the following code: // ---- CODE ------ #include <iostream> #include <vector> class S { public: S(const std::vector<char>& v_) : v{v_} {} void undefined() { std::cout << v.front() << std::endl; } private: const std::vector<char>& v; }; int main(){ std::vector<char> foo {'f', 'a', 'i', 'l'}; std::cout << foo.front() << std::endl; S s{foo}; s.undefined(); return 0; } // ---- END CODE ------ Compiled with: g++ -std=c++11 main.cpp s.undefined(); prints invalid characters or crashes the executable. I think the result of the problem is that, the: v{v_} initialization creates a new temporary from the vector that is destroyed after the execution leaves the scope of the constructor. ( This would only be the intended behaviour in case v would be initialized from initializer list, but {v_} is clearly not an initializer list here.) If I replace the uniform initialization with regular one: v(v_) the snippet above works as intended. The very same snippet does not compile with gcc 4.5.2. Slightly related report on that issue: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50025. I guess the origin of this problem is the incomplete fix of the error above.
(In reply to comment #0) > Consider the following code: > > // ---- CODE ------ > > #include <iostream> > #include <vector> > > > class S { > public: > S(const std::vector<char>& v_) : v{v_} {} > void undefined() { > std::cout << v.front() << std::endl; > } > private: > const std::vector<char>& v; > }; > > int main(){ > std::vector<char> foo {'f', 'a', 'i', 'l'}; > std::cout << foo.front() << std::endl; > S s{foo}; > s.undefined(); > > return 0; > } > > // ---- END CODE ------ > > Compiled with: g++ -std=c++11 main.cpp > > s.undefined(); prints invalid characters or crashes the executable. > > I think the result of the problem is that, the: v{v_} > initialization creates a new temporary from the vector that is destroyed after > the execution leaves the scope of the constructor. ( This would only be the > intended behaviour in case v would be initialized from initializer list, but > {v_} is clearly not an initializer list here.) > > If I replace the uniform initialization with regular one: v(v_) > the snippet above works as intended. > > The very same snippet does not compile with gcc 4.5.2. Slightly related report > on that issue: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50025. > > I guess the origin of this problem is the incomplete fix of the error above. - I think the result of the problem is that, the: v{v_} + I think the source of the problem is that, the: v{v_}
(In reply to comment #0) > I guess the origin of this problem is the incomplete fix of the error above. There is no fix, PR 50025 is still open and this is just a dup of it. N.B. the C++11 standard actually required this behaviour, but that's a defect that's been fixed in the latest draft. *** This bug has been marked as a duplicate of bug 50025 ***
(In reply to comment #2) > (In reply to comment #0) > > I guess the origin of this problem is the incomplete fix of the error above. > > There is no fix, PR 50025 is still open and this is just a dup of it. I added a new report because 50025 is about a compilation error, and with gcc 4.7 the code compiles, but fails to work as expected, but probably there issues are closely related. > > N.B. the C++11 standard actually required this behaviour, but that's a defect > that's been fixed in the latest draft. > Could you point me to which part of the draft required this behaviour? Thanks for your response, Gábor
(In reply to comment #3) > I added a new report because 50025 is about a compilation error, and with gcc > 4.7 the code compiles, but fails to work as expected, but probably there issues > are closely related. The difference is not caused by GCC 4.7, it's because you used a const-reference and PR 50025 uses a non-const reference. If you change your code to use non-cosnt references you get a compilation error too. > Could you point me to which part of the draft required this behaviour? See the comments for PR 50025, there's a link to DR 1288