[openbeos] Re: Kernel/app_server req #768 - suspend/resume
- From: "Manuel Jesus Petit de Gabriel" <freston@xxxxxxxxxxx>
- To: <openbeos@xxxxxxxxxxxxx>
- Date: Tue, 2 Apr 2002 20:26:53 -0800
----- Original Message -----
From: "Michael Phipps" <mphipps1@xxxxxxxxxxxxxxxx>
To: <openbeos@xxxxxxxxxxxxx>
Sent: Tuesday, April 02, 2002 8:02 PM
Subject: [openbeos] Re: Kernel/app_server req #768 - suspend/resume
> This is an R2 thing.
>
> But I will talk a little bit, here, about the technical side of it...
>
> Unix apps, as many of you know, can be sent a SIGSTOP (iirc).
> The app stops and is not scheduled, much as Zenja suggests, here.
Pedantic note: BeOS kernel sends a SIGSTOP to a thread as result
of a suspend_thread(tid); as it sends a SIGCONT as a result of a
resume_thread(tid).
You probably are confusing SIGSTOP with the usual Ctrl-Z bussines. That
it is not SIGSTOP but SIGTSTP. The latter requires all the session and job
control stuff that it is not implemented in the BeOS tty layer.
> >From what I understand, and if someone knows differently, please say so,
> signals under R5 are on a per-thread basis. And that makes sense, given
that
> really, in the kernel, the threads of a process are a lot less bound
together than
> in a traditional Unix style threaded environment.
Signals, sig handlers, sig masks and (of course) pending signals, are on a
per-thread
basis. However kill(-pid, ...) will send a signal to all threads in process
pid if i remember
correctly.
Note also that some of the signal stuff it is fixed in Dano (for instance
SIGWINCH
was not properly handled, that was the reason why some version of ssh died
when
you resized the window).
> Also, you would have to decide what to do with allocated resources. If the
app opens a window,
> it could easily get confusing, since the app_server side controls would
still function, but the messages
> to the application would stack up...
In general it is a bad idea to stop applications threads indiscriminately.
Also note that there is
a lot of application code on the net that does not check the return code of
things like
acquire_semaphore().
manuel,
Other related posts:
