At 09:09 AM 3/30/2002 -0500, Dan Sugalski wrote: >At 1:03 AM -0500 3/30/02, Melvin Smith wrote: >>Frame stacks now keep their size, no use in freeing the chunks; if we >>reached a frame depth N once, we will typically reach N many more times. > >If someone's feeling ambitious, code to check the number of unused chunks >may be in order--that way if we have more than N unused chunks we can free >the excess. (For those cases where the stack gets massively, but rarely, blown)
I almost did this but decided to call it a night. I was thinking along the lines of a resource limit similar to the JVM where we can pass in optimal and maximum stack sizes. We'd just need a chunk counter in the chunk member and multiply it times the frames per chunk. For a tuning configurable it would be nice to modify FRAMES_PER_CHUNK at compile time. Max stack size is a good check against spiralling death syndrome because I can say, "Ok for this particular production app, there should never be a possibility of a depth of 1000 frames so if we ever get there dump a bytecode image before we get to 2000, 3000, 10000 and end up thrashing the system or dumping a 1Gb core the hard way." -Melvin
