[GRASS-dev] [GRASS GIS] #1761: v.net.steiner broken

[GRASS GIS]

#1761: v.net.steiner broken
---------------------------+------------------------------------------------
 Reporter:  cmbarton       |       Owner:  grass-dev@…              
     Type:  defect         |      Status:  new                      
 Priority:  minor          |   Milestone:  7.0.0                    
Component:  Vector         |     Version:  svn-trunk                
 Keywords:  v.net.steiner  |    Platform:  All                      
      Cpu:  All            |  
---------------------------+------------------------------------------------
Changes (by cmbarton):

  * type:  enhancement => defect


Comment:

 Replying to [comment:2 mmetz]:
 > The Steiner tree problem [0] is NP-hard and a heuristic algorithm is
 used in this module so the result may be suboptimal. The implemented
 algorithm needs quite a lot of memory in order to determine a good (not
 necessarily optimal) solution. This is not a bug but inherent to the
 algorithm used by the module. There may be less memory-intensive
 solutions. Changing type to enhancement.
 >
 > [0] http://en.wikipedia.org/wiki/Steiner_tree_problem


 In GRASS 7, it simply doesn't do anything. No message. Nothing. It
 finishes with no output. That is a problem. In GRASS 6.4.3 it generates a
 memory error. These are clearly errors. If there is not enough memory,
 then it should simply give that as a message, not generate error messages
 or fail silently. So at least this needs to be fixed. I'm leaving this at
 minor, even though it will probably fail for many uses and users won't
 know why. But I'm changing this back to a defect because it needs to fail
 in a more graceful and informative manner at a minimum.

 I have 8GB RAM and sufficient disk space, and this fails on the
 streets_wake map with 3 nodes. it seems like a serious problem if it
 cannot handle this limited real-world network with the resources I have.
 Some suggestion in the docs to a realistic limit in the size of the
 network this can handle would be helpful until the algorithm can be better
 optimized.

-- 
Ticket URL: <http://trac.osgeo.org/grass/ticket/1761#comment:3>
GRASS GIS <http://grass.osgeo.org>