Ways std::stringstream can set fail/bad bit?

问题

A common piece of code I use for simple string splitting looks like this:

inline std::vector<std::string> split(const std::string &s, char delim) {
    std::vector<std::string> elems;
    std::stringstream ss(s);
    std::string item;
    while(std::getline(ss, item, delim)) {
        elems.push_back(item);
    }
    return elems;
}

Someone mentioned that this will silently "swallow" errors occurring in std::getline. And of course I agree that's the case. But it occurred to me, what could possibly go wrong here in practice that I would need to worry about. basically it all boils down to this:

inline std::vector<std::string> split(const std::string &s, char delim) {
    std::vector<std::string> elems;
    std::stringstream ss(s);
    std::string item;
    while(std::getline(ss, item, delim)) {
        elems.push_back(item);
    }

    if(/* what error can I catch here? */) {
        // *** How did we get here!? ***
    }

    return elems;
}

A stringstream is backed by a string, so we don't have to worry about any of the issues associated with reading from a file. There is no type conversion going on here since getline simply reads until it sees the line delimeter or EOF. So we can't get any of the errors that something like boost::lexical_cast has to worry about.

I simply can't think of something besides failing to allocate enough memory that could go wrong, but that'll just throw a std::bad_alloc well before the std::getline even takes place. What am I missing?

回答1:

I can't imagine what errors this person thinks might happen, and you should ask them to explain. Nothing can go wrong except allocation errors, as you mentioned, which are thrown and not swallowed.

The only thing I see that you're directly missing is that ss.fail() is guaranteed to be true after the while loop, because that's the condition being tested. (bool(stream) is equivalent to !stream.fail(), not stream.good().) As expected, ss.eof() will also be true, indicating the failure was due to EOF.

However, there might be some confusion over what is actually happening. Because getline uses delim-terminated fields rather than delim-separated fields, input data such as "a\nb\n" has two instead of three fields, and this might be surprising. For lines this makes complete sense (and is POSIX standard), but how many fields, with a delim of '-', would you expect to find in "a-b-" after splitting?

Incidentally, here's how I'd write split:

template<class OutIter>
OutIter split(std::string const& s, char delim, OutIter dest) {
  std::string::size_type begin = 0, end;
  while ((end = s.find(delim, begin)) != s.npos) {
    *dest++ = s.substr(begin, end - begin);
    begin = end + 1;
  }
  *dest++ = s.substr(begin);
  return dest;
}

This avoids all of the problems with iostreams in the first place, avoids extra copies (the stringstream's backing string; plus the temp returned by substr can even use a C++0x rvalue reference for move semantics if supported, as written), has the behavior I expect from split (different from yours), and works with any container.

deque<string> c;
split("a-b-", '-', back_inserter(c));
// c == {"a", "b", ""}

来源：https://stackoverflow.com/questions/2562906/ways-stdstringstream-can-set-fail-bad-bit

标签

c++

stringstream

getline