Consider the following code, it compiles and runs:
#include <iostream>
#include <sstream>
struct Foo {};
void operator>>(std::istream &, Foo) {
}
int main() {
std::stringstream{} >> Foo{};
}
However, if I change std::istream
to std::stringstream
, I get an error:
c.cpp: In function 'int main()':
c.cpp:7:25: error: no match for 'operator>>' (operand types are 'std::stringstream' {aka 'std::__cxx11::basic_stringstream<char>'} and 'Foo')
7 | std::stringstream{} >> Foo{};
| ~~~~~~~~~~~~~~ ^~ ~~~~~
| | |
| | Foo
| std::stringstream {aka std::__cxx11::basic_stringstream<char>}
c.cpp:4:6: note: candidate: 'void operator>>(std::stringstream&, Foo)' (near match)
4 | void operator>>(std::stringstream &, Foo) {
| ^~~~~~~~
c.cpp:4:6: note: conversion of argument 1 would be ill-formed:
c.cpp:7:10: error: cannot bind non-const lvalue reference of type 'std::stringstream&' {aka 'std::__cxx11::basic_stringstream<char>&'} to an rvalue of type 'std::stringstream' {aka 'std::__cxx11::basic_stringstream<char>'}
7 | std::stringstream{} >> Foo{};
| ^~~~~~~~~~~~~~
which makes sense: I cannot bind a lvalue reference to a rvalue (temporary) object.
Why does the first code compile?
UPD: my compiler is
g (Rev2, Built by MSYS2 project) 10.3.0
Copyright (C) 2020 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
and flags are -std=gnu 17 -Wall -Wextra -Wshadow -O2
UPD2: at the very end there is an error regarding the operator which Brian Bi is talking about:
In file included from C:/Software/msys64/mingw64/include/c /10.3.0/iostream:40,
from c.cpp:1:
C:/Software/msys64/mingw64/include/c /10.3.0/istream:980:5: note: candidate: 'template<class _Istream, class _Tp> typename std::enable_if<std::__and_<std::__not_<std::is_lvalue_reference<_Tp> >, std::__is_convertible_to_basic_istream<_Istream>, std::__is_extractable<typename std::__is_convertible_to_basic_istream<_Tp>::__istream_type, _Tp&&, void> >::value, typename std::__is_convertible_to_basic_istream<_Tp>::__istream_type>::type std::operator>>(_Istream&&, _Tp&&)'
980 | operator>>(_Istream&& __is, _Tp&& __x)
| ^~~~~~~~
C:/Software/msys64/mingw64/include/c /10.3.0/istream:980:5: note: template argument deduction/substitution failed:
C:/Software/msys64/mingw64/include/c /10.3.0/istream: In substitution of 'template<class _Istream, class _Tp> typename std::enable_if<std::__and_<std::__not_<std::is_lvalue_reference<_Tp> >, std::__is_convertible_to_basic_istream<_Istream>, std::__is_extractable<typename std::__is_convertible_to_basic_istream<_Tp>::__istream_type, _Tp&&, void> >::value, typename std::__is_convertible_to_basic_istream<_Tp>::__istream_type>::type std::operator>>(_Istream&&, _Tp&&) [with _Istream = std::__cxx11::basic_stringstream<char>; _Tp = Foo]':
c.cpp:7:32: required from here
C:/Software/msys64/mingw64/include/c /10.3.0/istream:980:5: error: no type named 'type' in 'struct std::enable_if<false, std::basic_istream<char>&>'
CodePudding user response:
As strange as it might seem, this code is well-formed. It should always compile. See Godbolt. It should also compile when the operator>>
overload is changed to take std::stringstream&
.
The reason is that there exists a special rvalue operator>>
overload for classes derived from std::ios_base
, marked (3) here:
template< class Istream, class T >
Istream&& operator>>( Istream&& st, T&& value );
The effect is equivalent to:
st >> std::forward<T>(value);
return std::move(st);
This is the operator>>
overload that is called by your code. It delegates to the operator>>
that you've written. (Because Foo
is a member of the global namespace, unqualified name lookup will find your operator>>
from any context thanks to ADL.)
If you have a toolchain that doesn't have the rvalue stream extraction operator, then your code will not compile because, obviously, a non-const lvalue reference, Base&
, will never bind to an rvalue of type Derived
(where Derived
is derived from Base
).
I don't know exactly why the rvalue stream extraction operator exists in the standard library. I think it's because someone realized that there's no good reason for is >> x
not to work if is
happens to be an rvalue expression of stream type. But many of the existing operator>>
s were free functions taking an lvalue reference to basic_istream
as their first argument. So adding this rvalue operator>>
overload, which delegates to an lvalue one, was the solution to this problem. According to this point of view, the fact that your code compiles is not a bug: it's intentional that you can write an operator>>
that takes a non-const lvalue reference to stream type and have it work on rvalues too.