Segmentation fault using AGG/PNG in label placement

[kjelderg]

I will of course try to provide the necessary details to understanding the bug. First, the symptom as reported by shp2img.

shp2img -all_debug 1000 -m minimal.map -o /dev/null -e -12981074.6475 3613583.0651 -12079727.9116 4514929.80106 -s 1460 1460

[Tue Aug 3 16:13:33 2010].232124 msLoadMap(): 0.002s

[Tue Aug 3 16:13:33 2010].243300 msSearchDiskTree(): Search returned no results. Unable to open spatial index for /local-project/www/safetravelusa.com/mapserv/./NDOT_data_arc.qix. In most cases you can safely ignore this message, otherwise check file names and permissions.

[Tue Aug 3 16:13:33 2010].244699 msDrawMap(): Layer 0 (roadshields), 0.002s

[1] 4647 segmentation fault (core dumped) shp2img -all_debug 1000 -m minimal.map -o /dev/null -e -12981074.6475 -s

A bt on the core provides the following

#0 0x000000000044c8a2 in msOracleSpatialLayerOpen ()

#1 0x00000000004b044d in png_write_data_to_stream ()

#2 0x0000000000435242 in msOracleSpatialLayerOpen ()

#3 0x000000000043b3ee in msOracleSpatialLayerOpen ()

#4 0x000000000040bff8 in ?? ()

#5 0x000000000040baee in ?? ()

...

#1440 0x0000000000000061 in ?? ()

Cannot access memory at address 0x800000000000

So what did I do to achieve this result? The details I know to be relevant follow. The mapfile (but unfortunately not the data file) follow.

DRIVER "AGG/PNG" is relevant. Switching drivers causes valid images to render.

LAYER TYPE annotation is "relevant" in that this seems related to the auto-placement of symbols.

LABEL SIZE 0 END is relevant. If I set the size to any other value, a valid image is rendered but of course with unwanted text visible. This implies that LABELITEM is needed of course.

[kjelderg]

stripped down mapfile.

[kjelderg]

I neglected to mention, my application is producing a tiled map and only a few tiles trigger this segmentation fault. Also, at other zoom levels the corresponding area is generally rendered fine and a set of tiles that is not geographically related triggers the segmentation fault.

[sdlime]

Adding Thomas as a CC...

[tbonfort]

the backtrace is suspicious, as this isn't a normal succession of function calls. maybe there's a library conflict somewhere.

without a datafile, there's not much we can do to investigate.