Classified non 8bit rasters not displaying

[msmitherdc]

Raster info:

Band 1 Block=480x4 Type=Float32, ColorInterp=Gray
  Min=18.645 Max=22.925 
  Minimum=18.645, Maximum=22.925, Mean=19.321, StdDev=1.002
  NoData Value=-3.4028234663852886e+38
  Metadata:
    LAYER_TYPE=athematic
    STATISTICS_MINIMUM=18.644718170166
    STATISTICS_MAXIMUM=22.925430297852
    STATISTICS_MEAN=19.320643792494
    STATISTICS_STDDEV=1.001758947595

  PROCESSING "SCALE=0, 255"
  PROCESSING "SCALE_BUCKETS=255"
  CLASS
    NAME "red"
    EXPRESSION ([pixel] > 0)
    STYLE
      COLOR 255 0 0
    END
  END

This displays a red raster in 5.6.5 but displays empty with trunk (r10920)

[warmerdam]

Mike,

I have a similar test in the regression test suite (msautotest/gdal/class16_oddscale.map) and it seems to working ok. I think I will need a minimal map file and test data file I can use to reproduce this problem to go further.

[msmitherdc]

Here's the data link

 http://12.189.158.78:8081/dem3_final.zip

ANd the URL Request:

 http://server/cgi-bin/mapserv?BASEELEV=28.8&map=dem_test.map&mode=map&map_imagetype=png24&LAYERS=lst_depth_grids&mapext=-9033184.6834905+3256600.1750351+-8951836.5761526+3304526.1917612&map_size=1128+768

And the layer is attached

[msmitherdc]

layer section showing >16bit data classification problem

[msmitherdc]

Hmm. I tried the autotest map, class16_oddscale.map, and it displayed fine in mapserver 5.6.5 but not in trunk (i got a blank image)

Both are linked dynamically to trunk gdal.

[warmerdam]

It turns out I have the same problem here, I just didn't know because perceptualdiff was telling me the results matched the test image. perceptualdiff does not understand color tables and was thus of limited use for this comparison.

I am pursuing the bug...

[warmerdam]

This relates to the following code in msGetClass_FloatRGB():

          case(MS_EXPRESSION):
            /*
            ** Big, fat TODO!
            **  
            ** See msEvalExpression() in maputil.c for an example of how to do this. Basically need to define some raster bind 
            ** variables: namely [red], [green], [blue] and [pixel]. Need to update the lexer and msTokenizeExpression() to 
            ** recognize them...
            */

            return(-1);

[warmerdam]

So it turns out that raster classification was never implemented when the expression parser was upgraded. I have done so now in trunk (r10976).

[sdlime]

Oh crap, sorry. I did make the big fat TODO note at least. Thanks for covering my big, fat... I'll take a look at your implementation.

There should be massive speed increases with rasters, I'd be very interested to know if that's actually true.

Steve

[warmerdam]

Steve,

The TODO was very helpful when I stepped through the code. :-)

I haven't tried any performance comparison - perhaps I can construct a test case.