[GRASS-dev] Re: [GRASS GIS] #1131: Global LFS for wingrass

[GRASS GIS]

#1131: Global LFS for wingrass
--------------------------+-------------------------------------------------
 Reporter:  mmetz         |       Owner:  grass-dev@…              
     Type:  defect        |      Status:  new                      
 Priority:  critical      |   Milestone:  7.0.0                    
Component:  libgis        |     Version:  svn-trunk                
 Keywords:  LFS,wingrass  |    Platform:  MSWindows 7              
      Cpu:  All           |  
--------------------------+-------------------------------------------------

Comment(by mmetz):

 Replying to [comment:11 glynn]:
 > Makefile substitutions and config.h macro definitions are completely
 separate.

 But should be in sync in this case, i.e. _FILE_OFFSET_BITS must be set
 according to USE_LARGEFILES.

 There are issues with wingrass, with raster processing. Using nc_spm_08,

 {{{
 g.region -p rast=elev_state_500m at PERMANENT res=100
 r.resample input=elev_state_500m at PERMANENT output=elev_state_100m
 method=bicubic
 }}}

 produces different results on Windows than on Linux. On Windows, the
 results for grass64 and grass7 with LFS are identical, but unfortunately
 different from Linux, where again results from grass64 and grass7 are
 identical. A difference map has (rounded) min = -7.73e-12  max = 5.91e-1,
 not much but not zero.

 One of the ctypes errors, the one about float('inf'), is caused by missing
 python support for inf and nan as valid string literals on some platforms.
 This seems to be independent of the python version. Python doesn't really
 know inf and nan, e.g. 0.0/0.0 produces nan in C but an error in python. A
 workaround I found is replacing float('inf') with e.g. float(1e1000) which
 should become DBL_MAX. Mathematically this is not correct, but in this
 case (lib/python/ctypes/ctypesgencore/expressions.py) it might do the job?

 The good news is (apart from the passed diglib test) that r.watershed in
 disk swap mode produced a >2GB temp file and completed successfully (only
 possible with LFS).

 Markus M

-- 
Ticket URL: <http://trac.osgeo.org/grass/ticket/1131#comment:12>
GRASS GIS <http://grass.osgeo.org>