Re: [gentoo-dev] RFC: The big eclasses massacre

[Andres Loeh]

> On Sunday 07 November 2004 22:12, Patrick Lauer wrote:
> >Also, what happens if someone uses an overlay with a
> > better/newer/older eclass version? As long as there is no distinction
> > between versions, I can imagine lots of *ahem* interesting problems that
> > could be avoided.
>
> Um, you should rename (e.g. datecode) it, if you put it in your overlay.
> User's overlay is nothing we have to care about. If you want to be absolutely
> safe, then use a snapshot, otherwise live with it, that Gentoo is a moving
> target. I can't see a benefit in eclass versioning, that outweighs the added
> maintenance overhead. If a new eclass is really needed, then it's always
> possible to add one, anyways. All eclasses have to be compatible to all
> ebuilds in the Portage tree, if not, open a bug report.

You're right, they should be compatible with all ebuilds in the
current portage tree. But if I understand things right, they in fact
have to be backwards compatible with all ebuilds that ever have been
in the portage tree for eternity.

The reason is (please correct me if I'm wrong) that, as far as I
can see, when an ebuild is installed on a user's machine, the ebuild
is saved, but not the eclasses it depends on. Should the user later
decide that the package should be removed from his system (or
updated), then the pkg_prerm and pkg_postrm functions from the *old*
ebuild are executed in conjunction with the *current* versions of
the eclasses it depends on. And the older an ebuild is, the more
likely it gets that the eclass has evolved beyond compatibility in
the meantime.

While this is usually unlikely to cause major breakage, it's still
an annoyance.

Cheers,
  Andres
 


--
gentoo-dev@g.o mailing list