Avoiding using same road twice in pgrouting?

Question

I am currently developing an application to create tours for motorbikes. At the current stage of development there is a start/endpoint and several points which outline the tour. Between these points the pgr_dijkstra function is used (with a customized cost function). The problem is best described by the picture attached.

Is there a way to avoid an edge getting traveled twice? I have tried to use the reverse cost column with a high cost but as the graph is not directed I get odd results. At least the pgrouting documentation states that the graph has to be directed to use the reverse cost if I understood that correctly.

sql := 'SELECT DISTINCT ON (seq) 
        seq, id, the_geom, id1 AS node, id2 AS edge, cost, km 
            FROM pgr_dijkstra(''
                 SELECT id,
                 source::integer,
                 target::integer,
                 my_cost::double precision AS cost,
                     FROM network 
                     WHERE network.the_geom && bbox
            '',' || source || ',' || target || ', false, false), 
            network WHERE id2 = network.id';

This is the statement to create a part of the route. It is nested inside a loop which passes new source and target in every iteration as long as there are some. This code is executed and returned record-wise.

Answer 1

First thing I see is (folowing documentation) if you want to include rewerse_cost in your function you have to include it in select, so:

sql := 'SELECT DISTINCT ON (seq) 
        seq, id, the_geom, id1 AS node, id2 AS edge, cost, km 
            FROM pgr_dijkstra(''
                 SELECT id,
                 source::integer,
                 target::integer,
                 my_cost::double precision AS cost,
                 my_reverse_cost::double precision AS reverse_cost 
                     FROM network 
                     WHERE network.the_geom && bbox
            '',' || source || ',' || target || ', false, false), 
            network WHERE id2 = network.id';

But I think It's not the resolution. Maybe you should try to create stored procedure as (pseudo-code):

1. Create tmp table tmp_network as select * from network n where n.way && st_expand(st_envelope(st_union(search_points)))
2. create tmp table tmp_resolution as select 1 as segment, * from first part of route using tmp_network
3. update tmp_network set cost = cost*2(4,8,100??) where exists (select 1 from tmp_resolution where id=id and segment = 1)
4. insert into tmp_resolution select 2 as segment, * from second part of route using tmp_network.
   ....

This (after tuning) will prevent routing via once selected edge if another way is possible and resonable.

If you need hardcore version just delete edges instead updating cost.