Created attachment 14429 [details] The source file The attached PS file produces horizontal lines in 9.22 that are not present when rasterized in 9.21. The command used is: /opt/local/bin/gs -sOutputFile=output92X.png -sDEVICE=png48 -dTextAlphaBits=4 -dGraphicsAlphaBits=4 -q -dBATCH -dNOPAUSE -r170 -f outputPS2.ps
Created attachment 14430 [details] The 9.21 output
Created attachment 14431 [details] The 9.22 output
Almost certainly this issue: https://bugs.ghostscript.com/show_bug.cgi?id=698672 *** This bug has been marked as a duplicate of bug 698672 ***
This appears to be yet another weird interaction of GraphicsAlphaBits and the news scan converter. You can restore the old behaviour by setting -dSCANCONVERTERTYPE=0 For ease of use debugging on Windows, the png16m device exhibits the same behaviour, as do other devices such as tiff24nc.
(In reply to Chris Liddell (chrisl) from comment #3) > Almost certainly this issue: I don't think so, its still exhibits with HEAD for me.....
Thanks for locating the (potential) duplicate bug. Of note, the search engine cannot find the duplicate when searching for "GraphicsAlphaBits", which is why I didn't find it in my initial search.
(In reply to Eric from comment #6) > Thanks for locating the (potential) duplicate bug. Of note, the search > engine cannot find the duplicate when searching for "GraphicsAlphaBits", > which is why I didn't find it in my initial search. Yeh, sorry, as Ken says, it still shows in the current code, so it's not the same problem - I must have mis-typed the command line when I tried it.
Fixed with: commit 1626f36c9b6ae68abad7818be7cfb2a95f4ecffe Author: Robin Watts <robin.watts@artifex.com> Date: Thu Nov 2 17:54:57 2017 +0000 Bug 698713: Further graphicsAlphaBits + new scan converter fix. Previous I ensured that when using graphicsAlphaBits, the new scan converter would restrict its output rectangles/traps to fit within y % max_fill_band pixels at a time as required. I forgot that the base 'y' value taken for this was not necessarily aligned to a max_fill_band multiple. Fixed here. Thanks for the report.