On Sat, May 28, 2011 at 3:04 PM, Ingo Weinhold <ingo_weinhold@xxxxxx> wrote: > You might have confused that with the thread's kernel stack. The userland > stack default is 16 MB + one page for the main thread and 256 KB + one page > for any other thread. It had been discussed to reduce the stack size for > (certain?) app server threads -- which the kernel and the pthread API support > (there isn't a public Haiku API for that yet, though) -- but nobody has done > that. Oops, indeed, didn't realize that allocation was only the kernel side of it. Thanks for clearing that up. > The system-wide limit is currently not enforced (though that should be > changed, of course). There are other related limits that might be hit: At > least three semaphores are created per thread, so the semaphore limit takes > effect too (it is computed at boot time depending on memory, but the upper > hard limit is 65536). Since various memory allocation are made for each > thread, the possibly allocatable kernel heap imposes another limit, which in > turn is implied by physically available memory and available kernel address > space (and fragmentation). The same limits (address space fragmentation to a > far lesser degree, though) are relevant for the thread's kernel stack. > Fair enough. In any case though, it's still safe to say that one doesn't have to be quite as paranoid about blowing up the app_server via off-screen bitmaps + windows as one does on R5. Regards, Rene