The customer reports and I've confirmed that the attached PDF file causes Ghostscript to use a large amount of memory when rendered at 300 DPI. This happens with gs8.60 and gshead (r8317). The customer reported 2 gigs of memory usage; my PowerBook core dumps after allocating 1.8 gigs. The command line I used: bin/gs -sDEVICE=tiff24nc -sOutputFile=test.tif -r300 ./47540xtesthela.pdf
Created attachment 3498 [details] 47540xtesthela.pdf.gz
This is probably a duplicate of bug 689492. Both trip over logic that results in transparency buffers being leaked. Running non-banded mode, the GC will locate and free these chunks, but when banding is used, the GC never runs during the 'print_page' so the leaks accumulate.
I now get "/undefined in G" with HEAD. Assigning back to support; maybe Alex should take a look?
Never mind, I missed that the attachment was compressed.
reassigning to Leo to see if this bug should have been closed with 689492.
Marcos will re-test
Memory usage for this file with gshead (r8693) is down to 743 megs.
With HEAD (rev 8700) this file completes in about 80Mb (adding -dBufferSpace=32000000 to the options for speed). Note there is a brief spike of memory usage up to about 160Mb (probably a duplicate of the SMask issue), but the primary large leak is fixed.
Created attachment 3987 [details] leak_fix.patch Attaching a patch for gs8.60 that fixes 8.60 to use the same amount of memory as head. This patch also fixes the 'leak' issue with bug 689492 on 8.60