Return the container, or take the container and assign to it?

I have a function which is for retrieving a result set from a MySQL db here. The idea is that I want to isolate the retrieval of the data and the formating of it from each other so I've figured that I should have a function - DBI::got_ims() - that does one of two things: either it should create a vector of vector<string>s (ie. a vector of rows from the db) and then return it, or it should take a pointer (or possibly reference) to a vector< vector<string> > and assign that and return a bool as to whether it worked or not.

The first method is below (this works OK, but looks horrible):

Code:

---

vector< vector<string> >
            DBI::got_ims( const string& to,
                          const string& from/*= NULL*/ )
{
/* code that uses to and from and queries the db
    (it's messy and irrelevant so ommitted) */

    rslt = mysql_store_result(&mysql);

    if( !rslt )        this->fail(F_GOT_IMS);

    if( mysql_num_rows(rslt) == 0 ){
        mysql_free_result(rslt);
        // return an empty set
        return vector<vector<string> >();
    }else{
        vector<vector<string> > msgs(mysql_num_rows(rslt));

        unsigned int j = 0;
        unsigned int num_fields = mysql_num_fields(rslt);

        while( row = mysql_fetch_row(rslt) ){
            for( unsigned int k = 0; k < num_fields; k++ ){
                msgs[j].push_back(row[k]);
            }
            ++j;
        }
        mysql_free_result(rslt);
        return msgs;
    }

---

I can't get the other method to work, because the msgs container is only working when it has a specified size - I thought vectors were supposed to automatically resize themselves, so is there a problem in my code?

Code:

---

// my debugger points to here in the <vector> header as
// where a nullReferenceException occurs (only when no size for msgs is specified at allocation).
        size_type size() const
                {        // return length of sequence
                return (_Myfirst == 0 ? 0 : _Mylast - _Myfirst);
                }

---

Anyway it might seem like I'm not asking anything, but I thought that even though this is done in c++, the act is so common on the internet that I hoped that someone could turn around and go: 'stop being stupid, we always do that this way,' and point me to a cleaner solution.

I would pass by reference, although returning it is also ok (but a possible pessimization).

The code should work the same for passing by reference, except for the call to construct msgs. Since msgs would be a function parameter, it will already be constructed. What you want to do is use resize it to set it to the correct size.

If you are worried about the passed in vector having junk data in it, one nifty technique is to use your current code, then just before returning from the function swap the local msgs vector with the passed in vector (e.g. local_msgs.swap(msgs)). This will ensure the passed in vector contains only the most recent values. This would be unnecessary if you place the assumption upon the calling code to pass in an empty vector.