[GRASS-dev] Re: [GRASS GIS] #1082: r.random cover= gets stray data in first and last rows

[GRASS GIS]

#1082: r.random cover= gets stray data in first and last rows
----------------------+-----------------------------------------------------
 Reporter:  hamish    |       Owner:  grass-dev@…              
     Type:  defect    |      Status:  new                      
 Priority:  normal    |   Milestone:  6.4.1                    
Component:  Default   |     Version:  svn-develbranch6         
 Keywords:  r.random  |    Platform:  All                      
      Cpu:  All       |  
----------------------+-----------------------------------------------------
Changes (by glynn):

  * platform:  Linux => All
  * cpu:  x86-32 => All
  * milestone:  6.4.0 => 6.4.1


Comment:

 Replying to [ticket:1082 hamish]:

 > the resulting "randout" map is nicely filled with random dots but on the
 first and last row of the output raster map there are chunks of solid dots
 which are created no matter what you do.

 This caused by the loop "continue"ing on null, without setting the output
 to null. So if the input cell is null but the cover isn't null, it ends up
 unconditionally writing the cover value to the output.

 This would be more obvious except that elevation.dem has so few nulls.
 (BTW, I think that base and cover are the wrong way around in the example.
 Either that or both r.random and r.statistics have them the wrong way
 around; I don't actually know which is correct).

 Merging the null tests into the main conditional gets around this issue,
 although the nc-- needs to also be moved into the body of the test. This
 should be fixed with r43716 (trunk).

 > also, for cover= it would be nice if random.c looped while(nt) so that
 it ran in multiple passes and you got as many points as you asked for, not
 less.

 The shortage of points is caused by it ignoring the number of nulls in the
 cover map when working out the probability of a given cell being chosen.
 The problem is, it can't work out the right answer because it counts the
 number of nulls individually for each of the two maps, regardless of
 whether they are disjoint or overlapping.

 This one is a bit more problematic to fix, and if I start, I'm probably
 going to end up trying to fix everything else which is wrong with r.random
 (e.g. the code is far more complex than it has a right to be, due to
 trying to process the data in its native format rather than just using
 DCELL internally).

 Also, the way that input= and cover= work is illogical; if you specify
 cover=, input= behaves as a base map; if you don't specify cover=, input=
 behaves as a cover map. It would make more sense for the options to be
 named base= and cover= or base= and input=.

-- 
Ticket URL: <http://trac.osgeo.org/grass/ticket/1082#comment:1>
GRASS GIS <http://grass.osgeo.org>