Vote: making wayland=on default

Grzegorz Junka list1 at gjunka.com
Thu Dec 21 20:40:41 UTC 2017


On 21/12/2017 03:19, Michael Gmelin wrote:
>
>> On 21. Dec 2017, at 02:14, Chris H <portmaster at BSDforge.com> wrote:
>>
>> On Thu, 21 Dec 2017 00:29:40 +0100 "Michael Gmelin" <freebsd at grem.de> said
>>
>>>> On 20. Dec 2017, at 18:50, Chris H <portmaster at BSDforge.com> wrote:
>>>>> On Wed, 20 Dec 2017 17:13:43 +0000 <freebsd-ports-owner at freebsd.org> said
>>>>> On Wed, 20 Dec 2017 16:23:59 +0000 "Johannes Lundberg" <johalun0 at gmail.com>
>>>> said
>>>>>> On Wed, Dec 20, 2017 at 4:08 PM, Chris H <portmaster at bsdforge.com> wrote:
>>>>>> On Wed, 20 Dec 2017 09:20:20 +0000 "Johannes Lundberg"
>>>>> <johalun0 at gmail.com>
>>>>>> said
>>>>>>
>>>>>>> Hi
>>>>>>>
>>>>>>> I want to suggest that we enable wayland by default. In current state
>>>>>>> having some parts of wayland in ports is basically useless the
>>>>>>> end-users themselves re-build gtk30 and mesa-libs with wayland
>>>>>>> enabled.
>>>>>>>
>>>>>>> libwayland-egl.so from mesa-libs and the extra libraries and headers
>>>>>>> from gtk30 adds like a few KB, a drop in the ocean compared to xorg
>>>>>>> packages. (might be something more that I missed)
>>>>>>>
>>>>>>> Personally I see no reason not to make it default on, even with
>>>>>>> flavors coming up. For any Desktop user (as well as embedded devices
>>>>>>> like IVI-systems and whatnot), Wayland is the future. There's no
>>>>>>> escaping that.
>>>>>>>
>>>>>>> Wayland has been quite usable on FreeBSD for over a year now but
>>>>>>> access to it is limited due to the extra efforts required to use it.
>>>>>>>
>>>>>>> If we are to compare with the other guys, several Linux distros are
>>>>>>> already switching to wayland-based compositors as default window
>>>>>>> server.
>>>>>>>
>>>>>>> What do you think?
>>>>>> IMHO it's (still) too early. Too much other X(org) related work
>>>>>> still being completed. In fact, I just built a new dev box to
>>>>>> track 12 (CURRENT), and this was the first time I was not required
>>>>>> to pre generate a config file for Xorg. I was only required to
>>>>>> inform /usr/local/etc/X11/xorg.conf.d/nvidia-driver.conf that
>>>>>> the driver was "nvidia", not "nv". Everything work(s|ed) famously.
>>>>>> A real treat. I'm also a bit concerned about the progress (or lack
>>>>>> there of) on network transparency.
>>>>>> I (personally) could conceive it as a KERNEL OPTION, but would not
>>>>>> want to see it in the Default kernel.
>>>>>>
>>>>>> Well, those are *my* thoughts. Because you asked. :-)
>>>>>>
>>>>>> --Chris
>>>>>>
>>>>> Thanks for your feedback!
>>>>> Just to clarify, we're not talking about changing any defaults that
>>>>> would impact or change users' choice of desktop. We only want to
>>>>> enable Wayland compositors as an alternative to X (leaving X as is).
>>>>> This does not break or modify anything existing. It does not force you
>>>>> to do anything differently. It simply adds a couple of libraries that
>>>>> you won't use unless you run Wayland stuff (if you install qt5/gtk30
>>>>> and mesa-libs).
>>>>> The reference to Linux making it default might have been unclear.
>>>>> Since FreeBSD doesn't have a default desktop, it's hard to change. It
>>>>> is and will continue to be up to the end user what they choose to use,
>>>>> we only add more options :)
>>>> Thanks for the informative reply, Johannes.
>>>> So no kernel (libs/extensions)?
>>>> Hmm, gtk3. Why is it not possible to make the Wayland stuff a sub
>>>> package/option? I think this is the preferred track/policy anyway.
>>>> I do this for all the ports I currently maintain. IOW any DE related
>>>> stuff I install, that uses GNOME related material, will pull in gtk3,
>>>> which, as I understand you say, will ultimately pull in Weston,mesa,...
>>>> is that correct? While I understand, you indicate it's only a few Kb.
>>>> I think it's cruft/(unnecessary)overhead. Which, in and of itself
>>>> seems insignificant. But in the "big picture", and over many (100's)
>>>> of builds/installations, is *not* insignificant. This also dismisses
>>>> the security related work, maintaining extra un(used|needed) material.
>>>> I suppose some will think that I'm just being nit-picky. But IMHO
>>>> I'm not. This sort of thing, if overlooked, *does* affect the bottom
>>>> line.
>>>>> Thanks again, Johannes!
>>>>> P.S. I have nothing against Wayland. I'm just not ready to run it
>>>> on anything "production" related, just yet. :-)
>>>>> --Chris
>>>> The key is to have it in a state that easy to maintain and allows people to
>>> install it using pkg install without conflicting with X, so you can switch
>>> back and forth easily. I'm also not ready to switch to wayland yet (favorite
>>> window manager not available, so many custom configurations I came up with
>>> over the years etc.), but giving users an easy way to test it (or use it, as
>>> it's becoming more and more mainstream now) is a good thing. Having a modern, working, out of the box desktop (read: no custom kernel
>>> builds, no need to use ports, a laptop is the point of first contact for many
>>> potential users) is incredibly important for proliferation and compared to
>>> the total size of binaries required to run X, I think the usefulness of
>>> providing wayland easily outweighs the extra overhead.
>> I wouldn't argue, nor did I argue those points. Who would? But muddying up
>> the individual ports (gtk3 for example) doesn't make anything lighter, or
>> better. Quite the contrary. IMHO Wayland should probably be added. Who
>> doesn't like more options? But, if it's coming to FreeBSD, and the ports
>> tree. It should isolate itself as it's own port(s), and include those
>> dependencies it requires. This is supposed to be policy. IOW if I decide
>> to include gtk3 as an option to one of the ports I'm installing as a run/
>> build depends, I don't want it installing Wayland, mesa, and a bunch of
>> other things I don't need -- no matter how small they might be.
> It's supposed to be insignificant in size, a small price compared to maintaining multiple versions of these ports. Just for comparison, on my system, gtk3 from pkg.freebsd.org requires about 450MiB including dependencies.
>
>> Doesn't that just make sense for *any* port? That's really my only possible
>> gripe. :-)
>>
> I understand the ambition to keep things lean and in general I agree with this approach of only installing what you need, but having multiple versions/flavors of certain libraries doesn't seem viable (multiple flavors? Multiple versions of the same package? Building applications twice?). That's why I'm in favor of it.
>
> Anyway, I think the positions are clear and that's all Johannes asked about.
>
> Yours,
> Michael
>

Why Wayland can not be enabled by default the same way that byhyve was 
enabled by default, even before it was production ready, so that people 
could test how it works for them? Wayland isn't going anywhere, rather 
the opposite, and having it on by default will allow to catch issues 
earlier. If it doesn't break the existing stuff then isn't it a win-win 
situation?

GrzegorzJ


More information about the freebsd-ports mailing list