Re: is Wayland mature enough to be the default desktop choice in Buster?

2019-04-21 Thread Simon McVittie
On Sat, 20 Apr 2019 at 23:17:15 +0200, Adam Borowski wrote:
> But there's also the technical matter of "GNOME and/or Wayland don't work
> at all on machine XYZ".  This _is_ relevant.

Whether GNOME works on machine XYZ (at all) is relevant when deciding
whether the default desktop environment should be GNOME or something
else, but is not relevant when deciding whether GNOME should default to
Wayland or X11 mode.

The decision of whether GNOME defaults to Wayland or X11 mode is only
affected by whether GNOME-on-Xorg works better than GNOME-as-Wayland.

(Another option is to default to GNOME-as-Wayland, but expand the rules
in gnome-session-bin that force X11 mode on certain hardware and drivers,
which are already used for things like the proprietary nvidia driver.)

> [1]. Hard-hangs during boot with systemd (even a purely text install),
> works fine with sysvinit+xfce all the way.

This is certainly not a bug in GNOME's Wayland mode (a purely text
install doesn't have any X11 or Wayland), and making GNOME default to
Wayland mode would not do anything to solve it.

smcv



Re: is Wayland mature enough to be the default desktop choice in Buster?

2019-04-21 Thread Simon McVittie
On Sun, 07 Apr 2019 at 17:59:38 +0200, Adam Borowski wrote:
> * an amd64 desktop:
>   * nouveau: way too crashy to be considered "working".

Please report this as a bug in the nouveau driver stack (sorry, I'm
not sure whether the kernel or Mesa is the right place).

If GNOME-on-Xorg doesn't work any better on this hardware than
GNOME-as-Wayland, then this bug isn't relevant to deciding whether GNOME
should default to Wayland mode or X11 mode.

> With xfce, disabling the compositor makes it work.  With GNOME, it's
> AFAIK not an option (the fallback is gone, right?).

GNOME Flashback (packaged as gnome-session-flashback in Debian) is the
closest equivalent of the fallback mode in stretch, but isn't really the
GNOME desktop environment: it's more like another GNOME fork alongside
Cinnamon and MATE.

>   * nvidia proprietary: doesn't work with new kernels.

My experience has been that it mostly does (SteamOS uses a recent
kernel and nvidia proprietary driver, via DKMS, on a system otherwise
heavily based on Debian 8), but I don't think the user-space part of
the proprietary driver supports Wayland. GNOME is meant to detect the
proprietary driver and fall back to Xorg mode automatically (if it
doesn't, please report that as a bug in gnome-session-bin, which is
where the supported/unsupported GPU detection lives).

> * Pine{64,book}:
>   simplefb.  GNOME no workie.
> 
> * RockPro64, used as a desktop (I'm typing these words on it):
>   armsoc.  GNOME no workie.
> 
> * N900:
>   didn't try.  I don't suspect it could work, though.
> 
> * Gemini:
>   libhybris.  No way to run Wayland I guess, X GNOME probably either.

GNOME is designed for a desktop or laptop with a working GPU, and these
devices probably aren't that (the N900 certainly isn't). If the Clutter
toolkit used in GNOME Shell doesn't work on this hardware (likely),
then GNOME-on-Xorg won't work there either, which means making GNOME
default to X11 mode wouldn't improve its ability to run on these devices.

> * Omega OAN133:
>   crashes with a black screen (although it's been a while since I tried).

Sorry, I don't know what that device is. If it's a desktop or laptop
from the last few years with an otherwise working and supportable GPU,
please report a bug.

> * an i386 desktop (used as a pedestal for RockPro):
>   i915 [910GL].  Might or might not run, although the mandatory compositor
>   on hardware this old would cause such a slideshow on 2560x1600 that it
>   wouldn't be usable.

Wikipedia tells me this is a 2004 Pentium 4 chipset, which I
suspect is much too old for the Clutter toolkit used by GNOME Shell to
work: last time I tried running GNOME on an Intel GPU of comparable age,
I think the problem was that Clutter requires driver/hardware support for
non-power-of-two textures, which wasn't present.

However, if GNOME-on-Xorg doesn't work any better on this hardware than
GNOME-as-Wayland, then it isn't relevant to deciding whether GNOME
should default to Wayland mode or X11 mode.

smcv



Re: is Wayland mature enough to be the default desktop choice in Buster?

2019-04-20 Thread Adam Borowski
On Sat, Apr 20, 2019 at 09:06:22PM +0100, Simon McVittie wrote:
> I would like to request that people who dislike GNOME, and would not
> use it regardless of what we do in its downstream maintenance, should
> not reiterate that opinion in the discussion of that bug (or in this
> thread, for that matter).

That's a very reasonable request for "I don't like GNOME because {poor
ergonomics,CSD,being counterintuitive,personal preference}" type of
opinions.  Those would be appropriate in a tasksel/d-i discussion, but
not here.  Apologies for even mentioning those reasons earlier.

But there's also the technical matter of "GNOME and/or Wayland don't work
at all on machine XYZ".  This _is_ relevant.

A good part of a Debian person's duties is testing stuff -- especially
shortly before a release.  If feature X doesn't work, it is up to us to
find that, regular users would complain only after the release when it's
way too late.  And here, we report that things have regressed.

Earlier in the thread, I posted a list of all machines I own, and a
report stating that I was unsuccessful attempting to even start GNOME
on any of them.  But, you may dismiss that as "fancy odd stuff" (although
a nVidia card in an amd64 machine is not that odd).  Thus, I just bought
a new boring ordinary machine with all pieces being mainstream, in part
specifically to be able to test mainstream stuff.  And it did not work,
in this case because of a dependency[1] -- but the result is the same.

Options that don't work on a substantial part of machines must either:
a) not be the default, or
b) autodetect and disable themselves

So while you're right to protest mixing personal preferences with
technical reasons, it's better to not bury reports that things do not
work.


Meow!

[1]. Hard-hangs during boot with systemd (even a purely text install),
works fine with sysvinit+xfce all the way.  As a detractor of systemd, it
is still my duty to substantiate the blame -- I've ordered a serial
console card to debug this, but received it only thursday evening then
departed for holidays friday right after work, thus didn't get around to
checking why it hangs.  But, a regular person would have no chance to even
try to troubleshoot, and conclude that "Debian sucks, it doesn't work".
-- 
⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Did ya know that typing "test -j8" instead of "ctest -j8"
⢿⡄⠘⠷⠚⠋⠀ will make your testsuite pass much faster, and fix bugs?
⠈⠳⣄



Re: is Wayland mature enough to be the default desktop choice in Buster?

2019-04-20 Thread Simon McVittie
On Mon, 08 Apr 2019 at 14:26:04 +0100, Jonathan Dowland wrote:
> Would the GNOME team kindly share with this thread the criteria that you folks
> use to make your decision as to whether to default to Wayland in Debian?

I didn't make that decision, so I can't cite any specific criteria.
Note that I don't consider myself to be a core member of the GNOME team:
I joined to help with maintenance of the lower-level bits like GLib, and
while I do help out with transitions and RC bug fixing in the higher-level
layers of the stack, anything I say should not be interpreted as some
sort of team policy or consensus.

The debian-devel mailing list is not an ideal way to reach the GNOME
team. I suspect that many team members might be deliberately avoiding the
-devel mailing list in an attempt to avoid their motivation being drained
by messages like many of the responses to this thread.

I've opened a bug against the gnome metapackage, cc'ing the -gtk-gnome
and -desktop mailing lists (and Jonathan), which I hope will result in
either a positive decision to follow upstream in keeping Wayland as the
default display protocol, or the revert that Jonathan advocated.

I would like to request that people who dislike GNOME, and would not
use it regardless of what we do in its downstream maintenance, should
not reiterate that opinion in the discussion of that bug (or in this
thread, for that matter). It isn't constructive and won't make Debian,
GNOME or GNOME-on-Debian better.

smcv



Re: is Wayland mature enough to be the default desktop choice in Buster?

2019-04-08 Thread Jonathan Dowland

Dear Simon

On Sat, Apr 06, 2019 at 10:20:26PM +0100, Simon McVittie wrote:

It's perhaps important to point out before this thread gets much further
that Wayland is not like Xorg


Apologies for not being clearer in my original message. Thank you for clearing
that up.


GNOME in buster has defaulted to Wayland mode since August 2017. The
default could presumably be swapped back to X11, as we did for stretch,


Apparently Ubuntu decided that it was not suitable (yet) as the default
for 18.04, which was an LTS release, and still stuck with Xorg for 18.10.

I don't know what criteria they used to make that decision, but I would imagine
it should be very similar to ours. Especially for an LTS release. I also don't
know what the status is for 19.04 which is presumably expected soon.


but I'm not sure whether post-hard-freeze is necessarily an appropriate
time to do that.


The freeze is a tool we use to try and ensure a quality and orderly release.
If we collectively agree that this change is necessary for the quality of our
release, then I'd hope the release team would also agree that such a change was
worthy of a freeze exception. But both of these hypotheticals are a few steps
ahead of where we are now.


If I understand correctly, the pattern that led to synaptic's removal is
that it runs its full GUI as root, which isn't supported by the way many
(all?) Wayland environments set up Xwayland.


I think that's right. I appreciate that this is has been considered a bad
approach to the problem for a long time, long before Wayland. I suspect,
but have not confirmed, that it is not the only program that will fail to
work properly in a Wayland environment. It's probably one of the more high
profile programs to break in Debian, though. Another I'd like to verify is
gparted.

Would the GNOME team kindly share with this thread the criteria that you folks
use to make your decision as to whether to default to Wayland in Debian?


Best wishes

--

⢀⣴⠾⠻⢶⣦⠀
⣾⠁⢠⠒⠀⣿⡁ Jonathan Dowland
⢿⡄⠘⠷⠚⠋⠀ https://jmtd.net
⠈⠳⣄ Please do not CC me, I am subscribed to the list.



Re: is Wayland mature enough to be the default desktop choice in Buster?

2019-04-06 Thread Simon McVittie
On Sat, 06 Apr 2019 at 20:47:51 +0200, Guillem Jover wrote:
> On Fri, 2019-04-05 at 16:12:22 +0100, Jonathan Dowland wrote:
> > I was surprised to learn — by way of synaptic being autoremoved — that
> > the default desktop in Buster will be GNOME/Wayland.

It's perhaps important to point out before this thread gets much further
that Wayland is not like Xorg: it's a protocol, not a program. GNOME
Shell, Weston and sway are all (separate) Wayland implementations: they
share some library code (and they share Xwayland as a compatibility
layer for X11 apps), but GNOME in Wayland mode and KDE in Wayland mode
have less in common than GNOME on X11 and KDE on X11.

In particular, when GNOME Shell runs in Wayland mode, there is no Weston
involvement.

Weston is definitely not a candidate for the default desktop in buster.
It's the reference implementation of a Wayland compositor, and doesn't
provide a full desktop environment - think of it as more like an
equivalent of Xorg + openbox (or some similarly minimal window manager)
than an equivalent of GNOME or KDE.

GNOME in buster has defaulted to Wayland mode since August 2017. The
default could presumably be swapped back to X11, as we did for stretch,
but I'm not sure whether post-hard-freeze is necessarily an appropriate
time to do that. The GNOME team made the corresponding change in stretch
(for which we had been hoping to default to Wayland, but decided that
it wasn't ready) 2 months before the transition freeze.

If I understand correctly, the pattern that led to synaptic's removal is
that it runs its full GUI as root, which isn't supported by the way many
(all?) Wayland environments set up Xwayland. The reason that Wayland
environment developers don't want to support this is that it makes it
very likely that the GUI that runs as root can be subverted by other
X11 apps connected to the same display.

> I don't use GNOME at all, but I tried to switch to Wayland last month
> (from i3 to sway), and sadly the experience lasted only a couple of days.

sway is not GNOME any more than Weston is, and doesn't necessarily
implement all the same Wayland interfaces as GNOME Shell. I would
recommend assessing Wayland implementations (compositors) on their
own merits.

smcv