Re: [Exim] remote delivery process count got out of step

ph10@??? (Philip Hazel) wrote on 28.08.02 in <Pine.SOL.4.44.0208281544070.26143-100000@???>:

> process xxx still exists: assume stolen by strace
>
> This seems to be a Linux "feature". The same thing does not happen under
> Solaris "truss", for example. I don't know about other OS.

It was recently discussed on linux-kernel. Several solutions were
discussed and rejected. Mainly, the problem seems to be bad design of the
ptrace() system, which reportedly other OSes have abandoned in favour of
better interfaces, though nobody explained how exactly others do their
work.

In any way, the real problem here is that a debugger (which strace counts
as) using ptrace does *also* expect to wait on the debugged processes as
if they were *its* children. There just does not seem to be any good
solution for making a process look like a child to two parents
simultaneously. Either one side breaks, or the kernel code gets ugly,
making the normal, undebugged case slower. Or at least that was the
consensus. And nobody really wants to slow down the undebugged case.

I expect a new interface will be developed in the end, but nobody knows
when.

MfG Kai