Parag Gadkari <parag.gadkari <at> gmail.com> writes:

It could be the way you are trying to pass parameters. Can you post a snippet

of what you are trying to do? Well, I tried to obtain something smaller, but I failed. Due to copyright restrictions I post only excerpts to give you an idea. Most of the code passed gcc-4.0.3 (cygwin), so I blame the compiler until proven wrong (looking forward to Service Pack 1). #include <boost/multi_index_container.hpp> #include <boost/multi_index/member.hpp> #include <boost/multi_index/ordered_index.hpp> #include <boost/multi_index/sequenced_index.hpp> #include <boost/multi_index/key_extractors.hpp> [...] #include <string> #include <iostream> [...] struct int_string_pair { int i_; std::string s_; int_string_pair(int i, std::string const & s) : i_(i) , s_(s) {} }; struct as_int {}; struct as_string {}; // shortcut for typedef struct property_storage { typedef int_string_pair value_type; typedef boost::multi_index::multi_index_container< value_type, boost::multi_index::indexed_by <boost::multi_index::sequenced<>, boost::multi_index:: ordered_unique<boost::multi_index::tag<as_int>, boost::multi_index::member<value_type, int, &value_type::i_> >, boost::multi_index:: ordered_unique<boost::multi_index::tag<as_string>, boost::multi_index::member<value_type, std::string, &value_type::s_> > > > value_storage_t; }; template <typename T> class CConfigurationProperty { public: typedef property_storage::value_type value_type; typedef property_storage::value_storage_t value_storage_t; private: value_type m_Value; value_type GetCheckedValue(int i) { value_storage_t const & possible_values = T::PossibleValues(); value_storage_t::index_iterator<as_int>::type it = boost::multi_index::get<as_int>(possible_values).find(i); ENFORCE(it != boost::multi_index::get<as_int>(possible_values).end()) ("Trying to assign invalid value '")(i) ("' to configuration property."); return value_type(*it); } value_type GetCheckedValue(std::string const & s) { value_storage_t const & possible_values = T::PossibleValues(); value_storage_t::index_iterator<as_string>::type it = boost::multi_index::get<as_string>(possible_values).find(s); ENFORCE(it != boost::multi_index::get<as_string>(possible_values).end()) ("Trying to assign invalid value '")(s) ("' to configuration property."); return value_type(*it); } public: CConfigurationProperty() : m_Value(GetCheckedValue(T::DefaultValue())) {} CConfigurationProperty(int i) : m_Value(GetCheckedValue(i)) {} CConfigurationProperty(std::string const & s) : m_Value(GetCheckedValue(s)) {} CConfigurationProperty(char const * s) : m_Value(GetCheckedValue(std::string(s))) {} operator int() const { return m_Value.i_; } operator std::string() const { return m_Value.s_; } CConfigurationProperty<T> & operator=(int i) { m_Value = GetCheckedValue(i); return *this; } CConfigurationProperty<T> & operator=(std::string const & s) { m_Value = GetCheckedValue(s); return *this; } CConfigurationProperty<T> & operator=(char const * s) { m_Value = GetCheckedValue(std::string(s)); return *this; } }; [...] struct example_type : public property_storage { static value_storage_t const & PossibleValues() { static value_storage_t const possible_values = boost::assign::list_of<int_string_pair> (-1, "undefined") (0, "foo") (1, "bar"); return possible_values; } static int DefaultValue() { return -1; } }; typedef CConfigurationProperty<example_type> example_type_t; example_type_t test; test = "foo"; assert(test != "bar"); assert(test == 0);

Joaquín M López Munoz <joaquin <at> tid.es> writes:

My hunch is that this is indeed just a compiler quirk, specially since you've reported that the problem is not reproducible in GCC. You can try the following:

1. What Boost version are you using?

1.33.1

Line 445 of ordered_index.hpp in Boost 1.33.1 does not even contain a '.' character.

Yes I know ... - wait - BINGO! If someone else states what I already know it always helps. This was the hint I needed. Someone had defined MACRO "upper" and "lower" in a header I included before multiindex stuff (only with VC8, which explains why the error did not crop up with gcc). What a pity. So it is not the compiler to blame this time, though emitting the _preprocessed_ offending line would have helped a lot in this situation. Thanks a lot! Markus