Re: [gentoo-project] Council meeting 2016-03-13

[Alexis Ballier]

On Thursday, March 10, 2016 5:40:18 PM CET, Ulrich Mueller wrote:
>>>>>> On Thu, 10 Mar 2016, Alexis Ballier wrote:
>
>> On Wednesday, March 9, 2016 5:37:08 PM CET, Andreas K. Hüttel wrote:
>>> 1) All pre-PMS, non-PMS-conformant behaviour should be considered
>>> deprecated immediately.
>>> 2) We encourage creation of trackers to hunt down and kill pre-PMS,
>>> non-PMS-conformant behaviour of ebuilds, eclasses, package managers
>>> 3) We introduce a hard deadline when all this should be fixed.
>
>> You seem to be generalizing to all cases from a very specific one:
>> multislot is breaking an important assumption (SLOT being constant)
>> and dropping it is not breaking anything.
>
>> Some examples that would fall under the scope of your proposal:
>
>> https://bugs.gentoo.org/show_bug.cgi?id=202631
>> -> Needed to comply with other PMS rules on some systems ('patch'
>> being GNU patch inside ebuilds, etc.)
>
> No, it isn't. The package manager is required to ensure that "sed",
> "patch", "find" etc. are the GNU versions. This is independent of any
> profile.bashrc. If Portage relies on aliases set in profiles instead,
> then this is a Portage bug which should be fixed.
>
> PMS reference:
> https://projects.gentoo.org/pms/6/pms.html#x1-12600011.3.1.1


Well, you can go into the debate whether perfectly working and needed 
behavior predating PMS which EAPI0 was supposed to normalize is a PMS bug 
or a portage bug, but your point would be much better emphasized with a 
patch providing the behavior you believe to be correct. It might even be a 
better solution since profiles aliases do not really work when e.g. 
building freebsd stuff from a gnu userland.
Mind the upgrade path though: If you remove the aliases from profiles, you 
might not be able to install the portage version providing them. Kind of 
defeats one of the goals of PMS :)


>> https://bugs.gentoo.org/show_bug.cgi?id=203891
>> -> Without this, we'd install a half-broken glibc by default. Any
>> deadline would have to take in consideration the time needed to have
>> a fixed glibc in stable.
>
>> (some ocaml stuff are also offenders here, but it is really minor in
>> comparison, and I've been trying to move away from the "feature"
>> causing the need for it as much as I could)
>
> We have discussed this at length (on the verge of derailing) in
> the bug. Someone has to write a spec, then it can go into EAPI 7.
> I actually like mgorny's proposal for a "dostrip" in comment 39.


I like it too and consider it a much better solution, but that's not the 
point.


> Until then, ebuilds should neither rely on the variable, nor should
> they call functions like prepallstrip that are internal to the package
> manager.
>
> PMS reference:
> https://projects.gentoo.org/pms/6/pms.html#x1-14500011.3.3.16


You prefer to install half-broken libc by default rather than fixing PMS, 
noted.



>> https://bugs.gentoo.org/show_bug.cgi?id=573306
>> -> Needed to get cross-compilation (or even ROOT!=/) to work
>> properly. (Independtly of PM getting cross-compilation deps
>> properly).
>
> I cannot say much about this one, or about cross-compilation in
> general. Reading the bug, I am not too optimistic, though. Seems there
> aren't even clear answers to the three simple questions that have been
> posed.


You should probably read again then: A few additive answers without 
ping-pong usually means all the parties agree.

[...]


Alexis.