Positioning bug with gdal support compiled in

[wbr@…]

If i compile mapserver-4.6.0 with gdal support, positioning of a point (Postgis
0.9.2) over a raster layer (GeoTiff) produces different positions of that point
if i pan the window rectangle in different directions (I have zoomed the raster
layer very deep, so i can see the point in a pixel group) This is reproduceable
with mapserver-4.0/4.25/4.4 with gdal-1.1.9/1.2.6/1.3.0 .

My Platform is RedHat Enterprise Linux 3.0 (in real CentOS-3.5 wich is a free
RHEL 3.0 with all updates)

I can send a few hardcopy's which shows this error !

Wolfgang Brungert

[wbr@…]

This is correct ! (without gdal support compiled in)

[wbr@…]

This is false (with gdal support compiled in)

[wbr@…]

This is also false, after a few mouse clicks to pan around (with gdal support compiled in)

[fwarmerdam]

By default MapServer's GDAL renderer stretches a particular pixel region
from the file to a particular pixel region in the result raster.  It does not
attempt to address subpixel resolution issues. 

That means when you are zoomed in far past 1:1 raster resolution you will 
never see just part of a disk pixel.  The pixels will always be evenly spread
over the canvas even though this is clearly a misrepresentation.  It is done
this way since at normal resolutions it is essentially accurate and it is 
faster to do this sort of bit blitting. 

If you were able to force the layer to go through the resampler, you should
see correct results.  Possibly MapServer ought to force use of the resampling
code when images are requested at much more than their native resolution.

I can't really address this just now, but I will leave the bug open.  Let me
know if you need information on how to tweak things to use the resampler.