Dangers for unprojectable layers

[warmerdam]

A client is encountering a problem where a tileindex raster layer is in an lcc projection and map request is issued in lat/long. If the lat/long request is from the other side of the world the reprojection into lcc will fail and currently msProjectRect() substitutes a "whole world" region due to the failure of the reprojection, but does return MS_FAILURE.

Unfortunately, the MS_FAILURE is ignored by the tileindex machinery of mapraster.c, and all tiles are selected. In turn it seems that attempts to select an appropriate region on the source files by reprojecting the map region fails and so the whole tile image is loaded as a fallback. Then much pointless work is done trying to sample the whole (huge in our case) tile, and nothing gets rendered.

So the point is that our effort to continue despite failure of msProjectRect() is causing serious performance problems in some configurations. Attached is a small patch to give up with a debug message but no error for layers into which the map request cannot be reprojected.

It is not entirely clear to me how wise it would be to change the default behavior in this way, but I offer the patch for consideration and possible future use.

[mko]

Adapted patch throwing an exception in stead of debug message only.

[sdlime]

Adding Daniel. Any reason not to patch 6.0? I'll defer to the raster guru's...

Steve

[dmorissette]

Raster guru, that would be Frank... adding him to CC...