Created attachment 7701 [details] xwd|xwdtopnm|pnmtopng of x11alpha display window The x11alpha device’s background has obtained some black boxes this week. I suspect the clipping changes; I will bisect to be sure. The command line I used for the example screen dump attached below is: ./language_switch/obj/pspcl6 -r72 -sDEVICE=x11alpha gs/examples/colorcir.ps (-r72 just to keep it small enough to send over my tiny link.)
I see that a debug build produces gray instead of black anomalies. I’ll attach an example of that, too. alphabet.ps, chess.ps, doretree.ps also show the bug. None of the other x11 devices do, w/ or w/o -d{Text,Graphics}AlphaBits=4.
Created attachment 7702 [details] xwd|xwdtopnm|pnmtopng of x11alpha display window from debug build ./language_switch/debugobj/pspcl6 -r36 -sDEVICE=x11alpha gs/examples/colorcir.ps (Even the 100K png from -r72 was an iffy upload.)
[SIGH] It was not the clipping, but rather the fix for bug #692234: :; git bisect bad a46123967e919fdd557411d2b0d177e6be2722e3 is the first bad commit commit a46123967e919fdd557411d2b0d177e6be2722e3 Author: Henry Stiles <henry.stiles@artifex.com> Date: Tue Jul 19 21:57:11 2011 -0600 Fixes 692234 premature erasepage. The page was being erased before the postscript interpreter was properly initialized. In particular for the crash in this bug a default halftone has to be installed before erasepage. :040000 040000 bc3f8c80274a95ed0f422cbb14f1b33689856ac9 d039777c7ff2af7c4bd11bfb6f5089a016cf8f9c M psi And revering that does fix this bug. Something must be left unititialized w/o the erasepage???
(In reply to comment #3) > [SIGH] > > It was not the clipping, but rather the fix for bug #692234: > > :; git bisect bad > a46123967e919fdd557411d2b0d177e6be2722e3 is the first bad commit > commit a46123967e919fdd557411d2b0d177e6be2722e3 > Author: Henry Stiles <henry.stiles@artifex.com> > Date: Tue Jul 19 21:57:11 2011 -0600 > > Fixes 692234 premature erasepage. > > The page was being erased before the postscript interpreter was > properly initialized. In particular for the crash in this bug a > default halftone has to be installed before erasepage. > > :040000 040000 bc3f8c80274a95ed0f422cbb14f1b33689856ac9 > d039777c7ff2af7c4bd11bfb6f5089a016cf8f9c M psi > > And revering that does fix this bug. > > Something must be left unititialized w/o the erasepage??? Yes right I thought the postscript setup code would do this but apparently it isn't. This is a patch I'm playing with, it basically moves the erasepage after the screen setup which should address the uninitialized value associated with the halftone you identified and erase the background so alpha works properly. git diff diff --git a/psi/psitop.c b/psi/psitop.c index e46c182..d146534 100644 --- a/psi/psitop.c +++ b/psi/psitop.c @@ -301,11 +301,12 @@ ps_impl_set_device( code = gsicc_init_device_profile_struct(device, NULL, 0); if (code < 0) return code; + /* Set the device into the gstate */ code = gs_setdevice_no_erase(pgs, device); - if (code < 0) return code; + /* install a screen appropriate for the device */ { const char *screen_str = ".setdefaultscreen\n"; @@ -319,6 +320,12 @@ ps_impl_set_device( if (code < 0) return code; } + + /* this shouldn't fail */ + code = gs_erasepage(pgs); + if (code < 0) + return code; + return exit_code; }
The rendering looks OK with that patch. I tested a range of ps, pdf and pcl file and have not been able to break it.
Fixed in http://git.ghostscript.com/?p=ghostpdl.git;a=commit;h=7e4051f