Agfdhyk

(/edit : added 3rd issue)

To parse easily primitive types from various origins (aswell as possibly some objects overloading operators << & >>), I have created this template function:

#include <sstream> 

#include <iostream> //only for the debug trace

#include <typeinfo> //only for the debug trace



template<typename T>

static inline bool getValue(std::string value, T& result)

{

    //std::cout << "getValue : '" << value << "'" << std::endl;

    //std::cout << "type : " << typeid(T).name() << std::endl;



    std::stringstream ss(value);

    ss >> std::boolalpha >> result;



    //std::cout << "result : " << result << std::endl;

    //std::cout << "ss.rdstate() : " << ss.rdstate() << std::endl;



    return ss.good();

}

This template function takes a string, and will try to extract a value according to the given reference type.

Output :

For a boolean, with value "true" :

getValue : 'true'

type : b

result : 1

ss.rdstate() : 0

For a boolean, with value "1" :

getValue : '1'

type : b

result : 0

ss.rdstate() : 4

For an int :

getValue : '60'

type : i

result : 60

ss.rdstate() : 2

For a string, with value "blah" : (yes i know, useless for a string, but if i want to loop on several parameters with various types...)

getValue : 'blah'

type : Ss

result : blah

ss.rdstate() : 2

I have here 2 issues :

1) boolean values

To be able to read value strings "true" or "false", I had to add this "std::boolalpha" manipulator. But then I am unable to successfully parse '0' or '1' for bool type.

2) rdstate flag

In case of error when parsing (with the boolean issue for instance) , rdstate returns 4 (ie "fail").
In case, it is correct, it will either return '0' (good) for the boolean ; or '2' (ie "eof") for integers or string.
Can someone explain why this difference between these types ? should i consider "eof" as correct ? I assume yes.

Documentation of operator>> (http://www.cplusplus.com/reference/istream/istream/operator%3E%3E/) states that failbit is set for :
"Either no characters were extracted, or the characters extracted could not be interpreted as a valid value of the appropriate type"

So the empty string will trigger a fail. (checked and rdstate is 6, ie eof|fail);

3) side question : value in case of error :

Testing it, I tried an error case when setting an int. I called my function with a zero-initialized int, and an empty string :

int timeout=0;

getValue ("", timeout);



getValue : ''

type : i

result : 60000

ss.rdstate() : 6

This 60 000 value looks very weird to me. it corresponds to the init value of another int.. WTF !
For sure I am checking the result, but some other users may not. Should I enable the exceptions on failbits?