Dmitry Torokhov wrote:
> On Sun, Mar 08, 2009 at 09:56:45PM +0100, Rafael J. Wysocki wrote:
>> On Sunday 08 March 2009, Matthew Garrett wrote:
>> > On Sun, Mar 08, 2009 at 08:45:59AM +0000, Richard Hughes wrote:
>> > > On Sat, Mar 7, 2009 at 8:39 PM, Matthew Garrett <[email protected]>
>> > > wrote:
>> > > > We don't have to at all - as far as I've been able to tell, the
>> > > > kernel is utterly consistent in its current usage. The only drivers
>> > > > that emit KEY_SLEEP are either embedded-specific (where it's
>> > > > clearly suspend to RAM and not hibernate), the ACPI driver (where
>> > > > usage in other operating systems is consistent with it being
>> > > > suspent to RAM) and the panasonic and thinkpad drivers which use it
>> > > > consistently. If there's any confusion, it's over the fact that
>> > > > KEY_SUSPEND is is used for suspend to RAM in a (smaller) number of
>> > > > places.
>> > >
>> > > The fact that we're mapping x->y and y->x is the reason people keep
>> > > getting it wrong.
>> >
>> > Sure, doing things differently would have made sense several years ago
>> > when nobody was relying on this behaviour. We don't have that option
>> > now - making this change will break things, and we've got no idea how
>> > much it'll break.
>>
>> Which is a good enough reason to avoid it.
>>
>> Alternatively, we can add completely new definitions _along_ _with_ the
>> old ones, mark the old ones as obsolete (after some time) and try to make
>> the user space start using the new ones only (that may be difficult,
>> though).
>>
>> I said I liked the names, but I didn't realize that changing them would
>> break things.
>>
>
> I don't think we want to break anything if we can help it. The problem
> with Richard's patch is that it changes meaning of KEY_SUSPEND from STD
> to STR. I would prefer if we could do the following:
>
> - KEY_SLEEP - leave the keycode, the action should be the default
> system state defined by either platform or user. I expect that the vast
> majority of system have default state similar to S3 so there should not
> be anysurprises.
>
> - KEY_SUSPEND - provide better comment for its intended usage and maybe
> add KEY_HIBERNATE alias.
>
> - KEY_SUSPEND2RAM - add a new definition.
>
> Do you think this would this work?
>
> I intend to back out the patch in question for the time being.
>
What about systems that have clear indication of STR and STD on respective
keys? Should they continue to return KEY_SLEEP for STR? I think this
conflicts with definition above. So in the long run those should be changed to
return KEY_SUSPEND2RAM it seems.
On Mon, Mar 09, 2009 at 11:31:36AM +0300, Andrey Borzenkov wrote:
> What about systems that have clear indication of STR and STD on respective
> keys? Should they continue to return KEY_SLEEP for STR? I think this
> conflicts with definition above. So in the long run those should be changed to
> return KEY_SUSPEND2RAM it seems.
Perhaps, but at present they KEY_SLEEP is probably adequate. KEY_SLEEP
is always going to default to suspend to RAM, so really the only
distinction between it and a dedicated suspend to RAM key is whether a
UI element affects its behaviour or not.
--
Matthew Garrett | [email protected]