MS RFC 25: Align MapServer pixel and extent models with OGC models¶
steve.lime at DNR.STATE.MN.US
At present MapServer uses different pixel and extent model than defined by OGC services such as WCS and WMS. MapServer uses the center of a pixel to represent its unique coordinate value. An extent is interpreted as the bounding box that runs from the center of the UL pixel in an image to the center of the LR pixel in an image. Why? Well, it goes back to companion software that existed along side MapServer to display satellite data stored in ERDAS that used the center to center extent model. The math is simple and there is a certain logic in having the extent actually represent pixel values - that is, if you render the extent as a polygon you get the exact edge of the image as one might expect.
On the other hand, OGC service specifications define an extent (BBOX) to refer to the dimensions of the outside edges of the image being requested. This appears to be a far more common means of expressing an area of interest.
I've not been able to ascertain where the coordinate of an individual pixel is located from various OGC specifications. MapServer '''could''' retain a center-based pixel model. That does add complexity to the map <=> image coordinate transformations since you have to offset things by one-half cellsize. Since that computation is done many times I would expect a performance hit. We could optimize things by computing and storing the one-half cellsize value once (as cellsize is now), but that complicates the C APIs and requires huge amounts of change. I propose moving to a upper-left-based pixel model to simplify these conversions.
Note: In looking at the code there were past efforts to go to the OGC extent in 4.8 and 4.10, but it was not universally applied. This RFC would ensure that the same extent and pixel model is in use throughout MapServer and that it is consistent with OGC.
Affected files (relative to what is in main development trunk):
map.h: change definition of MS_CELLSIZE
#define MS_CELLSIZE(min,max,d) ((max - min)/d)
map.h: change coordinate conversion macros (no change just add comments)
** These macros work relative to the UL corner of the UL pixel of a map exent. Pixel
** model is (as of 5.0) the UL corner of a pixel. UL pixel = minx,maxy.
#define MS_MAP2IMAGE_X(x,minx,cx) (MS_NINT((x - minx)/cx))
#define MS_MAP2IMAGE_Y(y,maxy,cy) (MS_NINT((maxy - y)/cy))
#define MS_IMAGE2MAP_X(x,minx,cx) (minx + cx*x)
#define MS_IMAGE2MAP_Y(y,maxy,cy) (maxy - cy*y)
maputil.c: update msAdjustExtent()
ox = MS_MAX((width - (rect->maxx - rect->minx)/cellsize)/2,0);
oy = MS_MAX((height - (rect->maxy - rect->miny)/cellsize)/2,0);
mapwcs.c: remove code to convert between MapServer extent and OWS extent
mapwms.c: remove code to convert between MapServer extent and OWS extent
mapwmsclient.c remove code to convert between MapServer extent and OWS extent
various raster layer handlers: Need Frank's comments here.
mapscript ...zoom functions: I don't believe these will need any change.
The affect of these changes on end-users should be minimal since it is unlikely they the are aware of differences in extent interpretation. The larger impact may be on 3rd party applications like dBox, Chameleon and QGIS that manage extents and call MapServer. Efforts must be made to make those folks aware of the change.
None, these are internal changes only.
None, these are internal changes only.
The models for a pixel and an extent need to documented in a couple of places: mapfile reference, the related OWS service how-to's and perhaps a new how-to pertaining to just this topic.
Need to develop some tests to somehow validate the math. The WCS interface or output drivers using GDAL are excellent candidates since they produce georeferenced output. Some testing has already been done to verify the existence of the issue (first discovered via WCS) and the proposed fix.
Mini-images (e.g. 7x7) can be used to verify rendering makes sense. For example, if you draw the extent as a polygon you would expect to see lines for the left and top edges, but not for the bottom and right. That is because the maxx,miny extent values won't represent a pixel in the output (rather the next tile to the right and/or below).
Staffing / Timeline¶
Changes to maputil.h, map.h, mapwcs.c and mapwms.c would be done by Steve Lime. Changes to other portions of MapServer would be coordinated with the various component owners. This RFC would be completed for the 5.0 release.