Re: [gentoo-project] Call for Agenda Items -- Council Meeting 2015-09-13

[Alexandre Rostovtsev]

On Wed, 2015-09-09 at 15:05 -0400, Rich Freeman wrote:
> While I don't know that we'll get to a vote, I think that the whole
> qt4/5, gtk2/3, etc USE flag matter could really use some
> standardization.
>
> My personal sense of the best ideas of the thread would be to have a
> proposal like this:
>
> Have a generic GUI USE flag which should be used where appropriate.
> I'm hesitant to just suggest using X11 for this since Wayland is on
> its way.  Maintainers should use this flag if all they're doing is
> enabling/disabling any GUI, regardless of how it was implemented.
> Profiles and users are encouraged to manipulate this flag globally as
> appropriate.  This is considered a completely safe flag that is free
> from surprises, though obviously disabling it will mean lots of
> console-only applications.  Packages should not IUSE this if they
> exclusively do or don't support a GUI (ie installing USE=-gui xeyes
> shouldn't just install a bunch of doc files).
>
> An unversioned gtk/qt/etc flag may be used if appropriate if users may
> wish to enable specific GUIs.  This should be about picking a specific
> implementation, and not about turning the GUI itself on or off.  In
> general this should not be put in profiles, but it might be
> appropriate in a very specific profile such as a gnome or kde profile.
> Users should use care in setting this globally, but it might be
> appropriate for enviornments where you want more control over the
> toolkits that are in use, such as embedded.
>
> Versioned gtk/qt/etc flags may be used to control specific toolkit
> versions, when a package supports more than one.  The default should
> be whatever is most appropriate.  Users and profiles should almost
> never be setting these.  They shouldn't be used if only one version is
> supported.  If users set these globally they should expect things to
> break.  In fact, I'm not entirely opposed to sticking package names in
> the flags to make them truly local (like chromium-gtk3 to use an
> example from another thread), but I think it is still better to just
> have the name be uniform and expect users to not shoot themselves in
> the feet (and we can document flag description appropriately).

+1 from me. For applications, a generic gui flag would resolve my main
concerns about versioned toolkit flags.

However, for libraries, I believe that slotting (as long as the
particular library's build system and .so naming convention allows it)
is often better than flags due to lower build time: to change the set
of enabled toolkits, you don't need to rebuild all of the library's
binaries N times, once for each of N enabled toolkits. Which is
especially important for libraries with long compile times.

Basically, I want to make sure that the council won't force the gnome
team to combine webkit-gtk:2 and :3 into one slot :)