Re: [gentoo-dev] useflag policies

[Andrew Savchenko]

On Mon, 3 Aug 2015 00:34:51 +0800 Ben de Groot wrote:
> Recently some team members of the Qt project have adopted these ebuild
> policies: https://wiki.gentoo.org/wiki/Project:Qt/Policies
>
> I have an issue with the policy adopted under "Requires one of two Qt
> versions". In my opinion, in the case where a package offers a choice
> between qt4 or qt5, we should express this in explicit useflags

This is what the policy does:
"Implement both qt4 and qt5 USE flags"

> and a
> REQUIRED_USE="^^ ( qt4 qt5 )". This offers the user the clearest choice.

This will create insane amount of blockers if users have both flags
in make.conf (and this is a common scenario).

> Other developers state that users are not interested in such implementation
> details, or that forced choice through REQUIRED_USE is too much of a
> hassle. This results in current ebuilds such as quassel to not make it
> clear that qt4 is an option.
>
> This goes against the principle of least surprise, as well as against QA
> recommendations. I would like to hear specifically from QA about how we
> should proceed, but comments from the wider developer community are also
> welcome.
 
As far as I understand this is done to simplify user's experiense:
usually people set both USE="qt4 qt5" in global make.conf, because
they want qt in the first place.

This policy will allow to USE both qt versions whichever is
available preferring newer one. Quite reasonable approach.
Alternatives (^^() and ??()) will require micromanagement (e.g.
pagkage.use.conf) for dozens if not hundreds of packages for no
good reason. If someone still needs to override such policy (e.g.
to use qt4 when both are available), this can be done by
per-package configuration.

My idea is that packages should be fully controllable, but choises
of default behaviour should be done so, that in most cases
micromanagement will not be necessary.

I like this qt policy and I'm not sure if it violates any current
rule. But even in such case this rule should be fixed. Moreover,
this problem is not limited for qt: we have exactly the same issue
with gtk2 vs gtk3 and probably some other technologies.

Of course in theory it is possible to build package with two sets
of binaries supporting both qt4 and qt5, but I see little practical
need for that.

So I propose to add somewhere to devmanual/policies the following
recommendation: "If package supports several versions of the same
technology (e.g. qt4 and qt5) and more than one is enabled by USE
flags, ebuild should prefer the later one (in terms of technology
generation).".

Best regards,
Andrew Savchenko