Example code test.cpp
#include <array>
#include <string>
int main ()
{
// OK
const std::array<int, 2> array_int = {42, 1337};
std::array<float, array_int.size()> array_float_ok;
// Error
const std::array<std::string, 2> array_string = {"foo", "bar"};
std::array<float, array_string.size()> array_float_error;
return 0;
}
Compiling with g++ 4.8.4 (Ubuntu 14.04)
g++ -Wall -std=c++0x test.cpp -o test
Gives the following error message
test.cpp: In function ‘int main()’:
test.cpp:14:39: error: call to non-constexpr function ‘constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::basic_string<char>; long unsigned int _Nm = 2ul; std::array<_Tp, _Nm>::size_type = long unsigned int]’
std::array<float, array_string.size()> array_float_error;
^
In file included from test.cpp:1:0:
/usr/include/c++/4.8/array:162:7: note: ‘constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::basic_string<char>; long unsigned int _Nm = 2ul; std::array<_Tp, _Nm>::size_type = long unsigned int]’ is not usable as a constexpr function because:
size() const noexcept { return _Nm; }
^
/usr/include/c++/4.8/array:162:7: error: enclosing class of constexpr non-static member function ‘constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::basic_string<char>; long unsigned int _Nm = 2ul; std::array<_Tp, _Nm>::size_type = long unsigned int]’ is not a literal type
/usr/include/c++/4.8/array:81:12: note: ‘std::array<std::basic_string<char>, 2ul>’ is not literal because:
struct array
^
/usr/include/c++/4.8/array:81:12: note: ‘std::array<std::basic_string<char>, 2ul>’ has a non-trivial destructor
test.cpp:14:39: error: call to non-constexpr function ‘constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::basic_string<char>; long unsigned int _Nm = 2ul; std::array<_Tp, _Nm>::size_type = long unsigned int]’
std::array<float, array_string.size()> array_float_error;
^
test.cpp:14:40: note: in template argument for type ‘long unsigned int’
std::array<float, array_string.size()> array_float_error;
^
test.cpp:14:59: error: invalid type in declaration before ‘;’ token
std::array<float, array_string.size()> array_float_error;
^
test.cpp:9:39: warning: unused variable ‘array_float_ok’ [-Wunused-variable]
std::array<float, array_int.size()> array_float_ok;
^
test.cpp:14:42: warning: unused variable ‘array_float_error’ [-Wunused-variable]
std::array<float, array_string.size()> array_float_error;
^
Can someone explain this error? Why does the first example work while the second one does not compile?
The type std::string is not a literal type meaning that it cannot be manipulated as part of a constexpr function at compile time. At compile time, the compiler attempts to evaluate the array_string's size() function. The functions first type parameter as you can see in the first error is set to std::basic_string < char > (aka std::string); therefore, since std::string is not a literal type, the function cannot be evaluated as a constexpr function at compile time and you have an error.
I would refer you to the following to learn more about constexpr.
http://en.cppreference.com/w/cpp/language/constexpr
I would refer you to the following to learn about literal types.
http://en.cppreference.com/w/cpp/concept/LiteralType
Finally, try the following simple code and you will see that int and float are literal types and std::string is not. You can try it out with other types to see what are or aren't literal types.
#include <iostream>
int main(int argc, char** argv)
{
std::cout << std::is_literal_type<int>::value << std::endl;
std::cout << std::is_literal_type<float>::value << std::endl;
std::cout << std::is_literal_type<std::string>::value << std::endl;
return 0;
}
Hope that helps.
John
Related
I have a sorted vector of structures and I am trying to find the index of element that has the queried value in its member attribute. For this I am trying to use lower_bound, but I am having problems with the comparator lambda function. Here is a simplified version of my code:
#include <iostream>
#include <vector>
#include <algorithm>
struct Student
{
int id;
};
std::vector<Student> Students;
int main()
{
Student Ann = {1};
Student Bob = {3};
Student Alice = {4};
Students = {Ann, Bob, Alice};
int id_query = 3;
auto index_ptr = std::lower_bound(Students.begin(), Students.end(), id_query, [](const int a, const Student s) -> bool {return a < s.id;});
}
g++ with std=17 gives me the following
Starting build...
/bin/g++ -g /home/aram/dev/cpp/coverages/stack_overflow.cpp -std=c++17 -o /home/aram/dev/cpp/coverages/stack_overflow
In file included from /usr/include/c++/9/bits/stl_algobase.h:71,
from /usr/include/c++/9/bits/char_traits.h:39,
from /usr/include/c++/9/ios:40,
from /usr/include/c++/9/ostream:38,
from /usr/include/c++/9/iostream:39,
from /home/aram/dev/cpp/coverages/stack_overflow.cpp:1:
/usr/include/c++/9/bits/predefined_ops.h: In instantiation of ‘bool __gnu_cxx::__ops::_Iter_comp_val<_Compare>::operator()(_Iterator, _Value&) [with _Iterator = __gnu_cxx::__normal_iterator<Student*, std::vector<Student> >; _Value = const int; _Compare = main()::<lambda(int, Student)>]’:
/usr/include/c++/9/bits/stl_algobase.h:979:14: required from ‘_ForwardIterator std::__lower_bound(_ForwardIterator, _ForwardIterator, const _Tp&, _Compare) [with _ForwardIterator = __gnu_cxx::__normal_iterator<Student*, std::vector<Student> >; _Tp = int; _Compare = __gnu_cxx::__ops::_Iter_comp_val<main()::<lambda(int, Student)> >]’
/usr/include/c++/9/bits/stl_algo.h:2032:32: required from ‘_FIter std::lower_bound(_FIter, _FIter, const _Tp&, _Compare) [with _FIter = __gnu_cxx::__normal_iterator<Student*, std::vector<Student> >; _Tp = int; _Compare = main()::<lambda(int, Student)>]’
/home/aram/dev/cpp/coverages/stack_overflow.cpp:21:142: required from here
/usr/include/c++/9/bits/predefined_ops.h:177:11: error: no match for call to ‘(main()::<lambda(int, Student)>) (Student&, const int&)’
177 | { return bool(_M_comp(*__it, __val)); }
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~
/usr/include/c++/9/bits/predefined_ops.h:177:11: note: candidate: ‘bool (*)(int, Student)’ <conversion>
/usr/include/c++/9/bits/predefined_ops.h:177:11: note: candidate expects 3 arguments, 3 provided
/home/aram/dev/cpp/coverages/stack_overflow.cpp:21:83: note: candidate: ‘main()::<lambda(int, Student)>’
21 | auto index_ptr = std::lower_bound(Students.begin(), Students.end(), id_query, [](const int a, const Student s) -> bool {return a < s.id;});
| ^
/home/aram/dev/cpp/coverages/stack_overflow.cpp:21:83: note: no known conversion for argument 1 from ‘Student’ to ‘int’
Build finished with error(s).
The terminal process failed to launch (exit code: -1).
The comparator of std::lower_bound is supposed to take the object dereferenced from the iterator, i.e. the element as the 1st parameter, the value to be compared as the 2nd parameter.
The type Type1 must be such that an object of type ForwardIt can be dereferenced and then implicitly converted to Type1. The type Type2 must be such that an object of type T can be implicitly converted to Type2.
You need to change the order of the parameters, e.g.
auto index_ptr = std::lower_bound(Students.begin(), Students.end(), id_query, [](const Student s, const int a) -> bool {return s.id < a;});
LIVE
I have a small working code for finding a range of items that uses a special compare method. But when I try to rewrite it with lower_bound() and upper_bound() function, I get a strange error. I have written a small code to show my problem. Here is the code:
#include <iostream>
#include <vector>
#include <algorithm>
#include <cstring>
using namespace std;
int main() {
string str = "banana";
string keyword = "ana";
int sa_idx[] = {5, 3, 1, 0, 4, 2};
vector<int> sa(sa_idx, sa_idx + sizeof(sa_idx) / sizeof(int) );
auto cmp = [&str] (const int &a, const string &keyword) -> bool
{
return strncmp(str.c_str() + a, keyword.c_str(), keyword.length()) < 0;
};
cout << (upper_bound(sa.begin(), sa.end(), keyword, cmp) -
lower_bound(sa.begin(), sa.end(), keyword, cmp)) << endl;
return 0;
}
As you see, compare function uses keyword and value of sa array for compare decision. A standard says:
The type Type1 must be such that an object of type ForwardIt can be
dereferenced and then implicitly converted to Type1. The type Type2
must be such that an object of type T can be implicitly converted to
Type2.
My compare function has int type for first argument(because of vector<int> of array) and string for the second argument type(as type of keyword). But I don't know why I get following error:
In file included from /usr/include/c++/6/bits/stl_algobase.h:71:0,
from /usr/include/c++/6/bits/char_traits.h:39,
from /usr/include/c++/6/ios:40,
from /usr/include/c++/6/ostream:38,
from /usr/include/c++/6/iostream:39,
from prog.cpp:1:
/usr/include/c++/6/bits/predefined_ops.h: In instantiation of ‘bool __gnu_cxx::__ops::_Val_comp_iter<_Compare>::operator()(_Value&, _Iterator) [with _Value = const std::__cxx11::basic_string<char>; _Iterator = __gnu_cxx::__normal_iterator<int*, std::vector<int> >; _Compare = main()::<lambda(const int&, const string&)>]’:
/usr/include/c++/6/bits/stl_algo.h:2049:14: required from ‘_ForwardIterator std::__upper_bound(_ForwardIterator, _ForwardIterator, const _Tp&, _Compare) [with _ForwardIterator = __gnu_cxx::__normal_iterator<int*, std::vector<int> >; _Tp = std::__cxx11::basic_string<char>; _Compare = __gnu_cxx::__ops::_Val_comp_iter<main()::<lambda(const int&, const string&)> >]’
/usr/include/c++/6/bits/stl_algo.h:2114:32: required from ‘_FIter std::upper_bound(_FIter, _FIter, const _Tp&, _Compare) [with _FIter = __gnu_cxx::__normal_iterator<int*, std::vector<int> >; _Tp = std::__cxx11::basic_string<char>; _Compare = main()::<lambda(const int&, const string&)>]’
prog.cpp:19:57: required from here
/usr/include/c++/6/bits/predefined_ops.h:173:11: error: no match for call to ‘(main()::<lambda(const int&, const string&)>) (const std::__cxx11::basic_string<char>&, int&)’
{ return bool(_M_comp(__val, *__it)); }
^~~~~~~~~~~~~~~~~~~~~~~~~~~
prog.cpp:14:61: note: candidate: main()::<lambda(const int&, const string&)>
auto cmp = [&str] (const int &a, const string &keyword) -> bool
^~~~
prog.cpp:14:61: note: no known conversion for argument 1 from ‘const std::__cxx11::basic_string<char>’ to ‘const int&’
Have I missed something super obvious? Because it seems compiler looks for string as first argument of compare function.
Your problem here is that std::uppper_bound requires the cmp be of the signature
bool(T, decltype(*Iterator))
std::lower_bound has the opposite requirement though and wants
bool(decltype(*Iterator), T)
so you can't use the same comparator with both functions. There are other ways to modify the code but just adding a second comparator like
auto cmp1 = [&str](const string &keyword, const int &a) -> bool
{
return strncmp(keyword.c_str(), str.c_str() + a, keyword.length()) < 0;
};
auto cmp2 = [&str] (const int &a, const string &keyword) -> bool
{
return strncmp(str.c_str() + a, keyword.c_str(), keyword.length()) < 0;
};
cout << (upper_bound(sa.begin(), sa.end(), keyword, cmp1) -
lower_bound(sa.begin(), sa.end(), keyword, cmp2)) << endl;
allows the code to compile.
The following code:
std::array<int, 4> arr1;
std::array<float, arr1.size()> arr2;
...compiles with both gcc and clang because std::array::size is considered constexpr.
But the following does not compile with gcc (version 5.3.0 20151204):
std::array<std::vector<int>, 4> arr1;
std::array<std::vector<double>, arr1.size()> arr2;
For me, there is no reason such code should fail to compile if the first one works but since I did not find a lot of post on this I don't know if it is a gcc bug or a clang extension?
The error from gcc (that I don't really understand... ):
main.cpp: In function 'int main()':
main.cpp:6:46: error: call to non-constexpr function 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]'
std::array<std::vector<double>, arr1.size()> arr2;
^
In file included from main.cpp:1:0:
/usr/local/include/c++/5.3.0/array:170:7: note: 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]' is not usable as a constexpr function because:
size() const noexcept { return _Nm; }
^
/usr/local/include/c++/5.3.0/array:170:7: error: enclosing class of constexpr non-static member function 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]' is not a literal type
/usr/local/include/c++/5.3.0/array:89:12: note: 'std::array<std::vector<int>, 4ul>' is not literal because:
struct array
^
/usr/local/include/c++/5.3.0/array:89:12: note: 'std::array<std::vector<int>, 4ul>' has a non-trivial destructor
main.cpp:6:46: error: call to non-constexpr function 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]'
std::array<std::vector<double>, arr1.size()> arr2;
^
main.cpp:6:48: note: in template argument for type 'long unsigned int'
std::array<std::vector<double>, arr1.size()> arr2;
^
I think this is a related to CWG issue 1684. Previously, the constexpr requirements included:
The class of which a constexpr function is a member shall be a literal type
And std::array<std::vector<int>, 4> is not a literal type, and hence its size() member function would not be constexpr. However, the new wording allows for a constexpr non-static member functions for non-literal types, assuming those functions meet all the other requirements of constexpr (which array<T,N>::size() clearly does).
Per the new wording, this is a gcc bug. Previously filed as 66297.
The following code:
std::array<int, 4> arr1;
std::array<float, arr1.size()> arr2;
...compiles with both gcc and clang because std::array::size is considered constexpr.
But the following does not compile with gcc (version 5.3.0 20151204):
std::array<std::vector<int>, 4> arr1;
std::array<std::vector<double>, arr1.size()> arr2;
For me, there is no reason such code should fail to compile if the first one works but since I did not find a lot of post on this I don't know if it is a gcc bug or a clang extension?
The error from gcc (that I don't really understand... ):
main.cpp: In function 'int main()':
main.cpp:6:46: error: call to non-constexpr function 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]'
std::array<std::vector<double>, arr1.size()> arr2;
^
In file included from main.cpp:1:0:
/usr/local/include/c++/5.3.0/array:170:7: note: 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]' is not usable as a constexpr function because:
size() const noexcept { return _Nm; }
^
/usr/local/include/c++/5.3.0/array:170:7: error: enclosing class of constexpr non-static member function 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]' is not a literal type
/usr/local/include/c++/5.3.0/array:89:12: note: 'std::array<std::vector<int>, 4ul>' is not literal because:
struct array
^
/usr/local/include/c++/5.3.0/array:89:12: note: 'std::array<std::vector<int>, 4ul>' has a non-trivial destructor
main.cpp:6:46: error: call to non-constexpr function 'constexpr std::array<_Tp, _Nm>::size_type std::array<_Tp, _Nm>::size() const [with _Tp = std::vector<int>; long unsigned int _Nm = 4ul; std::array<_Tp, _Nm>::size_type = long unsigned int]'
std::array<std::vector<double>, arr1.size()> arr2;
^
main.cpp:6:48: note: in template argument for type 'long unsigned int'
std::array<std::vector<double>, arr1.size()> arr2;
^
I think this is a related to CWG issue 1684. Previously, the constexpr requirements included:
The class of which a constexpr function is a member shall be a literal type
And std::array<std::vector<int>, 4> is not a literal type, and hence its size() member function would not be constexpr. However, the new wording allows for a constexpr non-static member functions for non-literal types, assuming those functions meet all the other requirements of constexpr (which array<T,N>::size() clearly does).
Per the new wording, this is a gcc bug. Previously filed as 66297.
If I compile the code below:
#include <list>
using iter_t = std::list<const unsigned>::const_iterator;
int main(int, char**) {
return 0;
}
I get the following compile error:
In file included from /usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/x86_64-apple-darwin13.4.0/bits/c++allocator.h:33:0,
from /usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/bits/allocator.h:46,
from /usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/list:61,
from main.cpp:1:
/usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/ext/new_allocator.h: In instantiation of 'class __gnu_cxx::new_allocator<const unsigned int>':
/usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/bits/allocator.h:92:11: required from 'class std::allocator<const unsigned int>'
/usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/bits/stl_list.h:315:9: required from 'class std::_List_base<const unsigned int, std::allocator<const unsigned int> >'
/usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/bits/stl_list.h:447:11: required from 'class std::list<const unsigned int>'
main.cpp:3:41: required from here
/usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/ext/new_allocator.h:93:7: error: 'const _Tp* __gnu_cxx::new_allocator<_Tp>::address(__gnu_cxx::new_allocator<_Tp>::const_reference) const [with _Tp = const unsigned int; __gnu_cxx::new_allocator<_Tp>::const_pointer = const unsigned int*; __gnu_cxx::new_allocator<_Tp>::const_reference = const unsigned int&]' cannot be overloaded
address(const_reference __x) const _GLIBCXX_NOEXCEPT
^
/usr/local/Cellar/gcc/4.9.2_1/include/c++/4.9.2/ext/new_allocator.h:89:7: error: with '_Tp* __gnu_cxx::new_allocator<_Tp>::address(__gnu_cxx::new_allocator<_Tp>::reference) const [with _Tp = const unsigned int; __gnu_cxx::new_allocator<_Tp>::pointer = const unsigned int*; __gnu_cxx::new_allocator<_Tp>::reference = const unsigned int&]'
address(reference __x) const _GLIBCXX_NOEXCEPT
^
I get roughly the same error message if I change the container type from std::list to std::vector, std::deque or std::forward_list. If I change the template argument from const unsigned to unsigned it compiles fine.
What's going on here? I can't see why it wouldn't compile.
Visual C++ 2015 gives the clarifying diagnostic
” The C++ Standard forbids containers of const elements because allocator<const T> is ill-formed.
As I understand it, this is because an allocator's allocate method returns a pointer to an array of uninitialized T items, and if T is const then these items can never be initialized.
I can't find any explicit statement that allocator<const T> is ill-formed, so it is presumably a consequence of the semantics.
As Dieter Lücking notes in a comment, there is a direct clash for const type T between the two member functions that provide address of a reference and const_reference, because these are equal when the type is const.