On Dec 2,  3:03pm, manu@netbsd.org (Emmanuel Dreyfus) wrote:
-- Subject: Re: proposal for Linux exit_group emulation

| > This looks ugly to me. Why trap both the syscall and tne userret?
| 
| In userret to catch processes sleeping in the kernel
| In syscall to catch processes running in userland. I realize this is
| poor since the process could spend quite some time before going in 
| the kernel. Sending it a signal is not possible because it will be 
| reported to wait4(). What else can be done?

You have control over who gets notified in the wait call. I.e. you can
check a flag in the wait call and not notify the parent, right?

| > Have you tested the above with openoffice or java?
| 
| It works fine with java. I don't have the OpenOffice monster at hand.

You should install the monster, because this is the application that
usually breaks.

christos