Subject: [Boost-users] transform_iterator causing strange problems
From: Lindley M French (lfrench1_at_[hidden])
Date: 2009-05-26 19:09:59


I'm not sure if this is the fault of transform_iterator----I really don't see how it could be----but I'm really confused here.

I've got an unordered_map<GroupID, Group> (typedefed to GroupMap) in my class. The specifics of those types don't matter except that GroupID is an integral type, and Group contains a std::vector vec.

I'm trying to use a transform_iterator so that my class may be viewed as a group container:

typedef boost::transform_iterator<GroupSelector, GroupMap::const_iterator> const_iterator;
    const_iterator begin() const
    {
        return boost::make_transform_iterator(currgroups.begin(),GroupSelector());
    }
    const_iterator end() const
    {
        return boost::make_transform_iterator(currgroups.end(),GroupSelector());
    }
    std::pair<const_iterator, const_iterator> groups() const
    {
        const Group &grp = *(begin());
        return std::make_pair(begin(),end());
    }

Where my GroupSelector class is defined as

struct GroupSelector
{
    typedef const Group& result_type;
    const Group& operator()(const std::pair<GroupID, Group> &p) const
    {
        std::cout << p.second.vec.size() << " " << &p.second << std::endl;
        return p.second;
    }
};

Note the output statement. Now, the usage in my main function is

for (tie(iter,end) = final.groups(); iter != end; ++iter)
    {
        const Group& grp = *iter;
        cerr << &grp << endl;
        cerr << grp.segs.size() << endl;
...

My output is
10 0012F50C
10 0012F65C
0012F65C
1243176

which is strange for two reasons. The first line comes from GroupSelector when I dereference begin() in the groups() function. The second comes from GroupSelector when I dereference iter in main. As these should both correspond to the same iterator, I can't understand why the address of the returned Group reference is different.

The 3rd and 4th lines are outputted in main. Here's the second confusion: The address for the Group object is the same as in the most recent GroupSelector output, yet now the size of the vector within the Group comes out wrong. HUH?