This simple code:
#include <iostream>
#include <sstream>
int main()
{
float x = 0.0;
std::stringstream ss("NA");
ss >> x;
std::cout << ( ss.eof() ? "is" : "is not") << " at eof; x is " << x << std::endl;
}
returns different results, depending on which library I choose (I'm running clang from xcode 5 on OSX 10.9):
clang++ -std=c++11 -stdlib=libstdc++ -> not at eof
clang++ -stdlib=libstdc++ -> not at eof
/usr/bin/g++-4.2 -> not at eof
clang++ -std=c++11 -stdlib=libc++ -> at eof
clang++ -stdlib=libc+ -> at eof
It seems to me that if I try to read an alphabetic character into a float, the operation should fail, but it shouldn't eat the invalid characters - so I should get fail() but not eof(), so this looks like a bug in libc++.
Is there a c++ standard somewhere that describes what the behavior should be?
p.s. I've expanded the original test program like so:
#include <iostream>
#include <sstream>
int main()
{
float x = 0.0;
std::stringstream ss("NA");
ss >> x;
std::cout << "stream " << ( ss.eof() ? "is" : "is not") << " at eof and " << (ss.fail() ? "is" : "is not") << " in fail; x is " << x << std::endl;
if (ss.fail())
{
std::cout << "Clearing fail flag" << std::endl;
ss.clear();
}
char c = 'X';
ss >> c;
std::cout << "Read character: \'" << c << "\'" << std::endl;
}
and here's what I see:
With libc++:
stream is at eof and is in fail; x is 0
Clearing fail flag
Read character: 'X'
with stdlibc++:
stream is not at eof and is in fail; x is 0
Clearing fail flag
Read character: 'N'
p.p.s. As described in the issue that n.m. links to, the problem doesn't occur if the stringstream is set to "MA" instead of "NA". Apparently libc++ starts parsing the string, thinking it's going to get "NAN" and then when it doesn't, it gets all upset.
