Sven Neumann wrote: > > One more question related to size limits: How many bytes per pixel does the > > pixel cache use? I have a sneaking suspicion that the answer is 5 based on > > the image sizes where I'm running into problems. That's what ImageMagick > > uses for its pixel cache, and I run into problem at almost exactly the same > > image size. > > the answer is 1 for grayscale/indexed, 3 for RGB. Add another byte > if your image has an alpha channel and another 4 bytes for the > projection. The latter used to allocated copy-on-write at some > point. Unfortunately this feature has been disabled (at least to my > knowledge) I got the projection to a state where COWing from a simple source image worked but came across a showstopper bug, which is why it should be disabled until someone can fix the COW tile system's inability to (IIRC) do the right thing when COW-ing lazy-validated tiles that haven't been validated yet. I looked at it for ages and could never find the bug. As far as I could tell the projection tilemanager was the only source of tiles that we were potentially trying to COW which actually used the tile system's ability to validate lazily (tiles allow callbacks for populating their pixels the first time the tiles are ref'd). Well that's how it was last time I looked into tile management, for all I know the 1.3 tree uses Calvin's COW-manager which was looking pretty good a hundred years ago but possibly daunting to adapt to the trunk (even back then). -- Adam D. Moss . ,,^^ adam@xxxxxxxx http://www.foxbox.org/ co:3 *camZZZ> Aspirin: weirdly, last night I dreamed that you moved to the * US and spent all your time smoking dope and playing this * strange game involving prime numbers and chestnuts.
