Re: Recommends for metapackages
On 2012-07-10 20:15, Jonas Smedegaard wrote:
> On 12-07-10 at 07:35pm, Eugene V. Lyubimkin wrote:
> > On 2012-07-10 18:10, Jonas Smedegaard wrote:
> > > The very purpose of a meta-package is to _ensure_ that a certain set
> > > of packages is installed, not just recommend them: All (not only
> > > most) users of that package need all its dependencies satisfied
> >
> > My definition of meta-package is less strict than yours. I as user
> > sometimes want '[meta]package X, but without packages Y and Z', and
> > your definition absolutely rules that out.
> >
> > I saw many questions on forums like
> >
> > "I did '$packagemanager install $metapackage' and then after
> > '$packagemanager remove $singlepackage', why $packagemanager now wants
> > to remove all $metapackage?"
> >
> > , so I know I'm not alone.
>
> [...]
> You being alone does not make you right.
>
> A package manager wanting to remove all dependencies of a meta-package
> is quite sensible - when you understand the sense of it. Until then it
> is utterly confusing.
As someone who developed a high-level package manager for Debian from
scratch (including the autoremoval functionality) I'm pretty sure I
understand the sense of it. My message was: users who don't (yet)
understand the full picture, find that behavior confusing, and it takes
time to explain. Moreover, despite me understanding the picture, I still
has no clean, safe and documented way to do what I'd want in case the
package maintainer chosed Depends.
Next, I don't pretend I'm "right", I do pretend there are >= 1 person
who don't need all dependencies of the metapackage installed, and hence
your 'All [...] users of that package need all its dependencies
satisfied' clause is wrong. You can argue that it's not right for Debian
to support that use case, that's fine.
> > Using Recommends for non-core parts of
> > metapackages' dependencies would nicely solve that.
>
> ...but I disagree that making meta-packages more elastic is a "nice"
> solution: is a hack covering over misguided users. Possible solutions
> could be improved documentation and improved design of package managers.
... And I disagree with that. No solution can override policy's "all
Depends must be satisfied". If one choose to support the "exclude from
metapackage" one either has to change the policy, remove packages from
Depends or use non-stock metapackage (which I personally don't like).
--
Eugene V. Lyubimkin aka JackYF, JID: jackyf.devel(maildog)gmail.com
C++ GNU/Linux developer, Debian Developer
Reply to: