关于临时人员销毁的编译器政策

Question

I've been playing with the following piece of code. file_string returns a temporary string that should only "live" until the end of the statement. In Visual Studio 2008, when you use pTempFolder , it contains rubbish as expected. In Linux though, with Intel compiler 11.0, pTempFolder still points to a valid string. Do compilers have different policies regarding the destruction of temporaries, kind of eager (Visual) versus lazy (Intel)? Or maybe this is just a coincidence?

boost::filesystem wpathTempFolder("/tmp");
const wchar_t* const pTempFolder = wpathTempFolder.file_string().c_str();
// use pTempFolder

BTW, that is boost filesystem version 2. I've also seen that file_string is being deprecated in boost filesystem version 3. And that there is a new c_str method that operates over a string&, instead of over a temporary string.

/*filesystem 2*/
const string_type file_string() const;
/*filesystem 3*/
const string_type&  native() const;  // native format, encoding
const value_type*   c_str() const;   // native().c_str()

Answer 1

Likely, the string is still invalid, it just so happens that that section of memory hasn't yet been de-allocated at the operating system level and it "happens" to work. This program exhibits undefined behaviour- which always includes "may continue to work as if nothing went wrong". Visual Studio is completely correct here to crash your program or pretty much anything.

Answer 2

The lifetime of a temporary (with a few exceptions) is until the end of the full expression. The lifetime of the array object pointed to by the return value of std::string::c_str() does not excede that of the string object itself (and may be shorter, if any non-const functions are called on the string). Accessing an object after its lifetime has ended is undefined behavior, so you cannot draw any conclusions from what the compiler does.