switch to modsetting on intel
Closed, ResolvedPublic

Description

The intel drivers is crappy crap. @sebas and @graesslin suggest to use the builtin modeset instead.

Testing suggests this is as trivial as simply having the driver not seeded (as mentioned in package description).
Unseeding should be done in such a way that a user can choose to install the driver manually. It should however not be pulled in by updates, nor installed by default.

@sitter says removing the driver switches to builtin kms which on an xps 13 solves all sorts of weird bugs to do with overlay painting (video rendering)

https://packages.debian.org/sid/xserver-xorg-video-intel

sitter created this task.Sep 14 2016, 9:57 AM
sitter moved this task from Ready To Do to Doing on the Neon board.Sep 21 2016, 1:46 PM
sitter claimed this task.

new thingy https://packaging.neon.kde.org/neon/xserver-xorg-video-intel-native-modesetting.git/commit/?id=85a3c0924b7ae32e3f96d5e148fc2e63269cc187

produces

  • xserver-xorg-video-intel-arbiter: an or-depends redirect (intel | intel-modeset)
  • xserver-xorg-video-intel-native-modesetting: or-dependet on by arbiter as first choice; conflicts and breaks the actual intel package; extra priority

new install workflow is thusly:

  • iso gets built
  • installs neon-desktop
  • neon-desktop depends abriter
  • arbiter defaults to modeset
  • iso gets neon-desktop & arbiter & modeset (intel driver not installed because modeset hates it and modeset has the higher dependency score as it resolves via a depends change rather than a recommends chain)

OPEN QUESTIONS

@graesslin @sebas should we forcefully remove the driver from existing installations? Or maybe display a notification prompting the user to let us do it (needs VDG and i18n involvement)? Or maybe just leave the driver installed?

Another thought that just occured to me. We could technically wire this up with the potential driver manager (T2844) to let the user get a notification via that and pick and choose via the manager. That entails vastly more work though as the manager needs a VDG overhaul, a port to packagekit, possibly custom code to make sure our modeset override is preferred (actually I have no clue how this magic works so I have no idea how one would make this work at all).

I wouldn't mind forcefully removing it, at least for dev/unstable.

Well, unstable eventually lands in user edition, so at some point we'd have to decide where to go with this. We could throw out a warning on social media and wait 30 days until we force modesetting. It's trivial enough to switch between the two packages, so if people have a bad experience it should be easily fixed.

I would forcefully remove it. We have so many issues with Intel driver, it's not funny

Pushed a change to unstable which should coerce existing systems from intel driver to modesetting for testing.

sitter moved this task from Doing to Review on the Neon board.Oct 27 2016, 9:29 AM

Landed in user edition. Waiting for some feedback. But should be implemented.

(for future reference: since neon-desktop needs to depend on the modeset you can't switch from modeset to intel without breaking the dependency. I presently have no good solution for this, so perhaps it'd be best if we simply lift the dependency in a month hoping the majority of systems got an upgrade and switched to the modesetting package)