Problem
As I am rewriting the documentation for the property map system, I find the current way to navigate headers rather complex:
| Header |
Contents |
<boost/graph/properties.hpp> |
Internal property tags (vertex_index, edge_weight, ...), get(tag, g) |
<boost/property_map/property_map.hpp> |
iterator_property_map, static_property_map, associative_property_map, dummy_property_map |
<boost/property_map/vector_property_map.hpp> |
vector_property_map |
<boost/property_map/function_property_map.hpp> |
make_function_property_map |
<boost/property_map/transform_value_property_map.hpp> |
make_transform_value_property_map |
<boost/graph/property_maps/constant_property_map.hpp> |
constant_property_map, make_constant_property |
<boost/graph/property_maps/null_property_map.hpp> |
null_property_map, make_null_property |
Property map types are split across two libraries: boost/graph/ (internal properties, constant/null maps) and boost/property_map/ (the generic adaptors). This split is an implementation detail users should not have to think about.
Solution
I am thinking about a convenience header that pulls those in an uniformize the namespaces. That would reduce the friction for newcomers, and we can leave this previous table in the documentation if more experimented users want to trim dependencies ?
#include <boost/graph/properties.hpp> // 374 lines pulls the PM library and its BGL extensions
// boost/property_map/property_map.hpp (iterator_, static_, assoc_, dummy_)
// boost/graph/property_maps/constant_property_map.hpp
// boost/graph/property_maps/null_property_map.hpp
#include <boost/property_map/vector_property_map.hpp> // 96 lines
#include <boost/property_map/function_property_map.hpp> // 66 lines
#include <boost/property_map/transform_value_property_map.hpp> // 67 lines
If I understood correctly, all of the underlying types live in the boost:: namespace so we are good namespace-wise.
Problem
As I am rewriting the documentation for the property map system, I find the current way to navigate headers rather complex:
<boost/graph/properties.hpp>vertex_index,edge_weight, ...),get(tag, g)<boost/property_map/property_map.hpp>iterator_property_map,static_property_map,associative_property_map,dummy_property_map<boost/property_map/vector_property_map.hpp>vector_property_map<boost/property_map/function_property_map.hpp>make_function_property_map<boost/property_map/transform_value_property_map.hpp>make_transform_value_property_map<boost/graph/property_maps/constant_property_map.hpp>constant_property_map,make_constant_property<boost/graph/property_maps/null_property_map.hpp>null_property_map,make_null_propertyProperty map types are split across two libraries:
boost/graph/(internal properties, constant/null maps) andboost/property_map/(the generic adaptors). This split is an implementation detail users should not have to think about.Solution
I am thinking about a convenience header that pulls those in an uniformize the namespaces. That would reduce the friction for newcomers, and we can leave this previous table in the documentation if more experimented users want to trim dependencies ?
If I understood correctly, all of the underlying types live in the
boost::namespace so we are good namespace-wise.