4 of the pdf's tests_private/comparefiles, Bug688778.pdf Bug689581.pdf Bug690379.pdf Bug689982.pdf either ends with /VMerror, segfault (killed by kernel oom killer?), or takes a long time to process through pxlcolor at the default -r600. They all work okay with ppmraw at -r600, and also either "-r600 -dNOTRANSPARENCY" (with some rendering mistakes) or "-r72" (one or two or them works at -r150 on my machine also). I file this against vectors, because of the -dNOTRANSPARENCY "workaround", and also the fact that -sDEVICE=x11/x11alpha also locks up my machine. Note that the files have extremely large dimensions (in pt - 3000 pt is almost 42 inches, or 3.5 foot): Bug688778.pdf 1950.54 x 2385.74 pdf 1.4 Bug689581.pdf 3003.99 x 5382 pdf 1.4 Bug690379.pdf 4535.43 x 8503.94 pdf 1.6 Bug689982.pdf - 1887.87 x 2681.57 pdf 1.5
The pxl device (as well as other gdevvec devices, and x11* and display do not use the clist device, so transparency can use large memory. I think there is another bug about this, I'll see about combining these.
I think the main culprit is pdfwrite, the bug #689805 is a place holder for a number of issues relating to pdfwrite not using the clist when flattening transparency.
I tried -K500000k -Zv and various resolutions; at least with Bug689581.pdf, it is just allocating 4 channels on pdf_open() and then again on seeing the first graphic object, I think, so that's 8 x width x height bytes right away.
The resource usage is not a problem anymore. They all works within reasonable time with x11 (just plain 'gs *.pdf' without arguments) now. timing "gs -r600 -sDEVICE=pxlcolor -o /dev/null ..." gives Bug688778.pdf 3:14.72 Bug689581.pdf 7:40.99 Bug689982.pdf 2:23.52 Bug690379.pdf 2:28.20 or 2 minutes to 8 minutes with maximum resident set size ~ 56MB. This seems to fit the description of bug 689805 . Ken's suggestion of this being a duplicate is good enough for me. There are a few rendering issues AFAIK with clist-based transparency but they are tracked elsewhere. *** This bug has been marked as a duplicate of bug 689805 ***