Package: acpid
Version: 1.0.6-3
Severity: serious
Tags: d-i

For a long time, at least since sarge, d-i has installed acpid by
default on acpi-capable systems. The reason is simple: Users have a very
basic expectation, that pressing the power button will turn the machine
off. This is a basic expectation, both because it's very basic to the
design of the machine, that it have a button you can use to turn it off;
and because users don't have a really refined concept of what the power
button does. Any intiatation of an action that ends up with the machine
turning off is broadly acceptable to most users. 

If this basic expectation is not met, all users will consider the system
to be broken or not support their hardware well. While some users will
be able to log in and shut it down by hand, others will mash on the
power button until a hard power cutoff happens, potentially to their
detriment. Note that users expect this to work whether or not a GUI
environment is working. Indeed, being able to hit the power button to
shut a headless server down cleanly is often even more important.

This is my impression based on reading thousands of installation
reports, as well as multiple user lists and forums, as well as observing
users in RL. Perhaps I'm wrong.

acpid's implementation of shutdown on power button press was clearly
suboptimal. The problems were:

#285779: Systems running desktops that don't handle power events are
         shut down on power button.
#354445: Spurious power events on resume taking some machines down.
         (Also #401414)
#386026: gnome-power-manager didn't get to do its thing since acpid ran
         shutdown. (Also #400409, #417306, #425291)

#386026 seems easily fixable (after all, the script already does a
similar test for KDE, which must work, based on the lack of bugs). The
bug even includs an example of how to test for a running
gnome-power-manager and let it do its thing.

#354445 seems fixable. Apparently Matthew Garret has some robust fix for
it. I suspect looking at Ubuntu might be helpful here. Fixing it by hand
doesn't seem too tricky.

#285779 is a weak bug. You hits the power button, and you takes your
chances. Losing unsaved files is not an unexpected consequence, and
should be especially expected by users who are proficient enough not to
use gnome and kde. Calling this "data loss" smells of hyperbole. Surely
the real data loss is when the power button fails to work, is mashed
down for a hard power cut, and an unclean shutdown hoses the filesystem?
The fact that gnome and kde go a step further with fancy shutdown
support that tries to save all files and double-check that "you are
ready to shut down your computer" a-la-windows doesn't make this
behavior needed for ion and console users.

acpid should fix #386026, put in the best available guard for #354445,
and re-enable the behavior of shutting down when the power button is
pressed.

I have filed this bug report as severity serious because this is a
serious reversion in functionality, that if shipped with lenny, will
flood out support channels with users confused as to why their machine
has stopped shutting down as they expect after the upgrade, as well as
new installs trying to figure out why debian can't support a power
button. I have tagged this bug d-i because I expect the d-i team will
take the brunt of this support load, and because, if it's not fixed
in the acpid package, we will have to find some other package to install
that does fix it.

-- 
see shy jo

Attachment: signature.asc
Description: Digital signature

Reply via email to