gdal_rasterize doesn't always create the proper sized output

[Kyle Shannon]

When gdal_rasterize creates a dataset, it can miss some rasterization:

kyle@kyle-bsu-workstation:~/Desktop/tmp$ cat in.csv 
WKT,value
"LINESTRING (0 0, 5 5, 10 0, 10 10)",1
kyle@kyle-bsu-workstation:~/Desktop/tmp$ cat in.vrt 
<OGRVRTDataSource>
    <OGRVRTLayer name="in">
        <SrcDataSource>in.csv</SrcDataSource>
        <GeometryType>wkbLineString</GeometryType>
        <GeometryField encoding="wkt" field="WKT"/>
    </OGRVRTLayer>
</OGRVRTDataSource>
kyle@kyle-bsu-workstation:~/Desktop/tmp$ gdal_rasterize -burn 1 -init 0 in.csv -tr 1 1 nobuff.tif -at
0...10...20...30...40...50...60...70...80...90...100 - done.
kyle@kyle-bsu-workstation:~/Desktop/tmp$ gdal_translate -of aaigrid nobuff.tif nobuff.asc
Input file size is 10, 10
0...10...20...30...40...50...60...70...80...90...100 - done.
kyle@kyle-bsu-workstation:~/Desktop/tmp$ cat nobuff.asc
ncols        10
nrows        10
xllcorner    0.000000000000
yllcorner    0.000000000000
cellsize     1.000000000000
 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 0 0 0 0 0 0
 0 0 0 0 1 1 0 0 0 0
 0 0 0 1 1 0 1 0 0 0
 0 0 1 1 0 0 0 1 0 0
 0 1 1 0 0 0 0 0 1 0
 1 1 0 0 0 0 0 0 0 1

I would expect the far right column to be burnt. If the input is POINT, the raster is created with a small buffer (+cellsize/2 for each dimension). I think this should be for all geometries.

The proposed patch:

Index: apps/gdal_rasterize.cpp
===================================================================
--- apps/gdal_rasterize.cpp	(revision 29611)
+++ apps/gdal_rasterize.cpp	(working copy)
@@ -362,8 +362,7 @@
             /* When rasterizing point layers and that the bounds have */
             /* not been explicitly set, voluntary increase the extent by */
             /* a half-pixel size to avoid missing points on the border */
-            if (wkbFlatten(OGR_L_GetGeomType(hLayer)) == wkbPoint &&
-                !bTargetAlignedPixels && dfXRes != 0 && dfYRes != 0)
+            if (!bTargetAlignedPixels && dfXRes != 0 && dfYRes != 0)
             {
                 sLayerEnvelop.MinX -= dfXRes / 2;
                 sLayerEnvelop.MaxX += dfXRes / 2;

fixes this. It breaks an test by changing some output raster sizes, but that is expected. I have a patch for the autotests too. Looking for some feedback, in case I misses something. The output with the fix (all steps the same):

kyle@kyle-bsu-workstation:~/Desktop/tmp$ cat nobuff.asc
ncols        11
nrows        11
xllcorner    -0.500000000000
yllcorner    -0.500000000000
cellsize     1.000000000000
 0 0 0 0 0 0 0 0 0 0 1
 0 0 0 0 0 0 0 0 0 0 1
 0 0 0 0 0 0 0 0 0 0 1
 0 0 0 0 0 0 0 0 0 0 1
 0 0 0 0 0 0 0 0 0 0 1
 0 0 0 0 0 1 0 0 0 0 1
 0 0 0 0 1 1 1 0 0 0 1
 0 0 0 1 1 0 0 1 0 0 1
 0 0 1 1 0 0 0 0 1 0 1
 0 1 1 0 0 0 0 0 0 1 1
 1 1 0 0 0 0 0 0 0 0 1

Which I think is expected. I don't think that the larger output will affect anyone too much.

[Even Rouault]

Looks good to me

[Kyle Shannon]

Done. r29617, rasterize: always create output with 1/2 cell buffer of input geometry envelope, #6058.

[Even Rouault]

trunk r29618 "Fix test". Fix pyflakes errors (​https://travis-ci.org/rouault/gdal_coverage/builds/74421633), and hopefully mingw/mingw-w64 failures due to line-feeds in command line + use of ReadAsArray() that requires numpy which isn't always there