Starting with r8668 Ghostscript output for 148-11.ps differs in page mode vs. banding (see screenshot.png, attached). The command lines I'm using: bin/gs -sDEVICE=ppmraw -sOutputFile=test1.ppm -r300 -dMaxBitmap=30000000 ./148-11.ps bin/gs -sDEVICE=ppmraw -sOutputFile=test2.ppm -r300 -dMaxBitmap=10000 ./148-11.ps
Created attachment 4243 [details] screenshot.png
This looks pretty bad, and we need to investigate in and either fix it or revert the relevant change. The log message for 8668 doesn't indicate expected differences: ------------------------------------------------------------------------------- r8668 | mvrhel | 2008-04-27 23:47:28 -0700 (Sun, 27 Apr 2008) | 1 line Changed paths: M /trunk/gs/src/gsicc.c Undo accidental commitment of ENABLE_CUSTOM_COLOR_CALLBACK (has callback) code. -------------------------------------------------------------------------------
Created attachment 4244 [details] Interp_Bug_Patch.diff This patch fixes the issue. Issue was caused by me thinking that we were always indexing in the source data by one byte forward when we were dealing with and indexed image. If the geometric transformation matrix had a -xx value we end up going backwards through the data.
Thanks for the prompt response; I'm testing the patch now.
Problem is fixed (patch also fixes big vs. little-endian issue I noticed as was about to add to the bug report). I'm closing the bug.