Re: [gentoo-dev] rfc: backward-incompatible changes in eclasses

[Michał Górny]

On Mon, 2020-03-23 at 14:26 -0500, William Hubbs wrote:
> On Mon, Mar 23, 2020 at 08:19:20PM +0100, Michał Górny wrote:
> > On Mon, 2020-03-23 at 13:23 -0500, William Hubbs wrote:
> > > Hey all,
> > >
> > > it has been brought to my attention that there have been several
> > > backward-incompatible changes made to the python eclasses lately.
> >
> > Does 'several' in this case mean more than one?  Please correct me if
> > I'm mistaken but the only change I can think of were the changes
> > in python-single-r1.
> >
> > > It is true that everything in ::gentoo has been fixed along with the
> > > changes to the eclasses; however, when a change like this goes into a
> > > widely used eclass it breaks overlays with little to no notice;
> > > especially since we do not require developers to be subscribed to this
> > > mailing list.
> > >
> > > I do agree that overlay authors are on their own to fix things, but we need to
> > > find a way to notify them when a breaking change is going into a widely
> > > used eclass and give them time to adjust their ebuilds.
> > >
> > > If the old way of doing things cannot be supported
> > > along side the new way the correct path forward is a new version of the
> > > eclass then a lastrites on the old version. That would give overlay
> > > authors time to switch to the new eclass.
> > >
> > > If the old and new way can be supported in the same code base, a
> > > reasonable way forward is to  allow both ways to exist while ::gentoo is
> > > migrated to the new code path then do the equivalent of a lastrites for
> > > the old code path so overlay authors can adjust their ebuilds.
> > >
> >
> > The lesson was learned.  If a similar change would be necessary
> > in the future, I will bump the eclass instead.  I don't understand why
> > you bring that today.
>
> The change that blew us up today was
>
> https://archives.gentoo.org/gentoo-dev/message/aa45f4f86f9b865eb0fe7344d83a7258

I'm not sure if you're referring to that specific patch or the whole
thread.  Because I don't see how that patch could break anything.  It's
switching between two methods of getting scriptdir that were added
simultaneously.

> and this is the reason I brought it to the ml.
>
> The previous change that blew us up was
>
> https://archives.gentoo.org/gentoo-dev/message/4bf9f0250115c57779f93817356df871

I don't understand how that could break anything either.  Before
the change, '${PYTHON_USEDEP}' wasn't valid (anymore) in dep strings,
so I don't see how anything could be more broken after the change.
In any case, there could be a little chance it could *fix* some old
ebuilds that just happened to use the right combination.

-- 
Best regards,
Michał Górny