On Mon, 2017-06-05 at 07:53 -0700, Dave Hansen wrote:

On 06/05/2017 06:09 AM, Bastien Nocera wrote:

quoted

quoted

I agree with Dave. If there is no solution found in time for
-rc5, 
reverting to previous state would be the proper way to go.

I don't see how it's possible to retroactively fix user-space.

It's not possible to retroactively change userspace.  That why the
kernel tries so hard not to break it in the first place.  Although
this
is in "minor annoyance" territory for me at the moment, this patch
causes a clear, user-visible issue with new kernels.

The right way to do this is to have the kernel export the data in a
way
that does not confuse old userspace.  Perhaps we should separate out
"power supplies that run the system" from "power supplies in a
perihperal".

There's already such a property for it, "scope". I think that you don't
realise that it's this version of UPower you're using (one major API
version behind the current one) is buggy when it comes to handling
kernel-created "power_supply".

It's just that UPower used to do this itself, in user-space, and that
it gets thoroughly confused when it accesses both the battery from
user-space and kernel-space.

And, no, a config option isn't the right thing either.

Because...? It's the best way to avoid exposing the feature for ancient
user-spaces. The battery information will be gathered from user-space.
It doesn't regress either.