Re: [gentoo-dev] GLEP 27 Proposal - Feedback Requested

[Brian Harring]

On Tue, May 30, 2006 at 09:14:01AM +0200, Danny van Dyk wrote:
> Hello Mike,
>
> Am Dienstag, 30. Mai 2006 05:29 schrieb Mike Kelly:
> > I'm Mike Kelly, one of the SoC-ers. I'll be working on GLEP 27 for
> > the summer. Right now I'm looking for some basic feedback on my
> > proposal.
> >
> > In particular, I know that at one point there was a push for the user
> > info files to be XML, but I think it may be easier to implement them
> > as simple shell variable files (like /etc/conf.d/*), since my plan
> > was to write the core of the implementation in shell (e.g. as an
> > eclass).
> >
> From your proposal:
>
>  - Add code to the eutils.eclass which will detect if the installed
>    version of portage supports the new system, notifies the operator
>    that the current ebuild is using depreciated code, and properly add
>    the user using the new system's code. This would check for the
>    proper EAPI version to know when to execute the new code instead.
>
> As a member of Release Engineering who encountered already a problem
> with user-management code in eutils.eclass, i beg you: _pleeeease_
> don't add it to that eclass. Instead, create a new eclass 'euser' or
> something similar and add it there.

Commented in irc re: this, but eclass doesn't really fly- if it's 
implemented strictly in an eclass, portage has no way of easily 
reverting the changes (mainly cause it's not aware of 'em) if the 
build fails, thus the user/group isn't needed.  Same for unmerging.

Further, if implemented strictly as a trick in one of the ebuild 
phases, setup is the likely place- means that setup cannot be deprived 
to non-root for running it (thorn in the side for trying to depriv all 
phases).

So... need it higher up then implemented in one of the ebuild phases. 
:)
~harring