Mask layer failure when nonsquare pixels

[AsgerPetersen]

This (anonymized) URL
http://host/fcgi-bin/mapserv.fcgi?map=/mapfile_with_mask_layer.map&request=GetMap&bbox=562968.936898,6123011.53890,614428.41406,6154345.17839&crs=EPSG:25832&layers=masked_layer&styles=&width=1281&height=781&format=image/png&transparent=TRUE&bgcolor=0xFFFFFF&exceptions=XML&version=1.3.0 returns a nicely masked map.

Whereas this URL
http://host/fcgi-bin/mapserv.fcgi?map=/mapfile_with_mask_layer.map&request=GetMap&bbox=562968.936896,6123011.5389,614428.41406,6154345.17839&crs=EPSG:25832&layers=masked_layer&styles=&width=1281&height=781&format=image/png&transparent=TRUE&bgcolor=0xFFFFFF&exceptions=XML&version=1.3.0 returns a blank image.

Note that the only difference between the URLs is on the 6th decimal of xmin in the bbox.

Looking at the debug output from the failing request these lines stand out:

...
msDrawMap(): kicking into non-square pixel preserving mode.
...
msPostGISLayerWhichShapes query: select encode(ST_AsBinary(ST_Force2D("geom"),'NDR'),'hex') as geom,"uid" from (select g.uid, g.geom
from public.filter_polygon g WHERE uid in (5,6 )) as subquery where geom && 
ST_GeomFromText('POLYGON((0 0,0 781,1281 781,1281 0,0 0))',find_srid('','public.filter_polygon','geom'))
...

Note that the BBOX used for the postgresql query seems to be the image size rather than the bounding coordinates of the request.

The debug output from the successful request doesn't show the non-square pixel preserving mode-notice and the postgres query looks like this:

...
msPostGISLayerWhichShapes query: select encode(ST_AsBinary(ST_Force2D("geom"),'NDR'),'hex') as geom,"uid" from (select g.uid, g.geom 
from public.filter_polygon g WHERE uid in (5,6 )) as subquery where geom && 
ST_GeomFromText('POLYGON((562989.02256434 6123011.53890013,562989.02256434 6154345.17838987,614408.32839366 6154345.17838987,614408.32839366 6123011.53890013,562989.02256434 6123011.53890013))',find_srid('','public.filter_polygon','geom'))
...

I am using mapserver 6.4.1.

[AsgerPetersen]

For future reference:
Explicitly setting a PROJECTION block on the mask layer works around this issue.

[jratike80]

No similar reports since 2015. Explicitly setting a PROJECTION block imho is not a workaround, leaving it unset is hazardous.

[AsgerPetersen]

No similar reports for eight years. Time flies by...

A great thank you to everyone working on this project. It is much appreciated!