Bug#347680: Xorg breaks acpid

[Alan Hourihane]

On Tue, 2006-01-24 at 11:56 +0100, Michel Dänzer wrote:
> On Mon, 2006-01-23 at 13:15 +0100, Lex Spoon wrote:
> > Marcus's analysis looks right to me:
> > > > But who can tell which is the grand-piece-of-sw that has the exclusive
> > > > right to open /proc/acpi/event?
> > > The one which has been designed to multiplex the events to make them
> > > available to other programs, and it's acpid.
> > 
> > The standard Debian setup is to use acpid to multiplex the event stream.
> >  Even if it is possible to run with acpid, that should be a different,
> > non-standard option.
> > 
> > There are lots of reasons to multiplex in userspace instead of in the
> > kernel.  Among these is that a userspace multiplexer can filter and
> > modify the stream as it goes by, and a userspace multiplexer can pull
> > events from multiple sources including virtual events injected by other
> > sources than the raw hardware.  The earlier comments that obviously the
> > kernel should multiplex just aren't right.  Doing it in userspace makes
> > sense, and if it is done in userspace, then additionally doing it in the
> > kernel would be a bad thing.
> > 
> > At any rate, the current strategy isn't right.  If the system is using
> > acpid--the most common configuration--then it is a bug for X to get in
> > the way of acpid.  
> > 
> > It is laudable that Xorg currently tries to support ACPI even when acpid
> > is not running.  However, this is an unusual and only marginally useful
> > configuration, isn't it?  Thus, if nothing else, it would seem to make
> > sense to simply remove the non-acpid ACPI support when compiling for
> > Debian.  In that case, the party line would be, if you want to use ACPI,
> > then you install acpid and programs talk to that.
> > 
> > If it is still desirable to make X work without acpid around, then that
> > should surely be a non-standard configuration which requires an
> > *explicit* configuration option on the X server.  This does not appear
> > worthwhile for Debian, since we do have acpid easily installable, but
> > maybe it is sufficiently worthwhile for other Linux distributions that
> > it is worth including.
> 
> I tend to agree, but I think that this discussion belongs upstream, so
> I'm CC'ing the xorg list.
> 
> Background for the xorg list: Debian is currently using a patch which
> makes the X server open /proc/acpi/event if acpid goes away, to avoid
> excessive attempts to access the acpid socket. The problem is that this
> prevents acpid from restarting.

I'm not sure I understand Debian's patch.

The thing is, the current code in X.Org tries to
open /var/run/acpid.socket which acpid offers already.

The Xserver only goes to open /proc/acpi/event when acpid isn't running.

Alan.