Sebastien Fortier ha escrito:

I will try to make this as simple as possible, I have the following struct

struct meteo_record { public: [...ctor...] [...getters and setters...] private: string var_name; string grid_type; float ip1; float ip2; float ip3; float ig1; float ig2; float ig3; int ip1Kind; int data_key; };

All these attributes are non-unique except for the data_key which is not a sorting criteria.

What I need to do is to be able to iterate over the whole set by certain criteria. [...] my criterias are sort by same gryd_type group by var_name with same ip2 and same ip1kind ordered on ip1 [...]

Hello Sebastien, I'm not sure I've fully understood all the details of your problem, but seems to me that composite keys ( http://tinyurl.com/hdf3q ) might be of help here. A composite key accepts several attributes and presents the records sorted in lexicographical order with respect to them. In your particular case, suppose you've got an index (non-unique) whose key is composite_key< meteo_record, const_mem_fun<meteo_record,const string&,&meteo_record::get_grid_type>, const_mem_fun<meteo_record,const string&,&meteo_record::get_var_name>, const_mem_fun<meteo_record,float,&meteo_record::get_ip2>, const_mem_fun<meteo_record,int,&meteo_record::get_ip1Kind>, const_mem_fun<meteo_record,float,&meteo_record::get_ip1>

(where I've used some imaginary names for the attribute getters). This index sorts records lexicographically by grid_type,var_name,ip2,ip1Kind,ip1 which looks like what you're after, from what I understand about the description of your current query process. The different groups can be then traversed like follows (for simplicity I assume the composite key is used in index #0, you'll probably want to have it in an additional index): for(recordIter=set.begin();recordIter!=set.end();) { recordIter2=set.upper_range( boost::make_tuple( it->get_grid_type(), it->get_var_name(), it->get_ip2(), it->get_ipKind(), ) ); // [recorditer1,recorditer2) is a range of elements with equal // grid_type,var_name,ip2 and ipKind, sorted by ip1 process(recordIter1,recordIter2); recordIter1=recordIter2; } Is this what you were looking for? Thank you for using Boost.MultiIndex, Joaquín M López Muñoz Telefónica, Investigación y Desarrollo