Hello Bill, Mathieu The maxmemory and JPEGMEM settings can only work when libjpeg has an alternative to use temporary files instead of RAM. But on modern systems libjpeg is usually installed with the "jmemnobs.c" (no backing store) memory module, and then these settings are not regarded, because libjpeg has no alternative.
You would have to configure libjpeg with something like "jmemansi.c" to make use of these settings. But this would have a disadvantage and thus isn't recommended for normal systems because actually the libjpeg memory management isn't capable to use all of available RAM when the backing store mechanism is enabled - you would always be limited to a predefined maximum setting. Unfortunately there is no benefit of libjpeg8 over libjpeg62 here - nothing was changed there. I know about this deficiency, but actually couldn't revise the memory management because there are currently more important things to do for me.
$ cjpeg -progressive -maxmemory 4m -outfile huge1.jpg huge.ppm while $ cjpeg -maxmemory 4m -outfile huge1.jpg huge.ppm does not use much memory
Of course this is a huge difference for libjpeg processing because "-progressive" has to buffer the whole image as DCT coefficient array as opposed to the sequential processing.
$ cjpeg -progressive -arithmetic -rgb -block 1 -outfile huge1.jpg -quality 100 huge.ppm
And this is even worse because in the current implementation "-block 1" treats every sample as a full 8x8 DCT coefficient block. If you want to push libjpeg to its extreme regarding memory consumption then something like this is indeed the way to go. This is also the reason why I always recommend "-arithmetic" together with "-rgb -block 1" for lossless coding. Without the "-arithmetic", cjpeg would use Huffman optimization (default for progressive and for block sizes < 8) with full image DCT buffer. The only practical reason why one would want to use above command would be to benefit from the successive approximation which is used for the DC coefficients in the default progressive script. But as said this has a huge effect on memory consumption. Ciao Guido -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org