Bug#974870: libgdk-pixbuf2.0-dev: Plans for Xlib API deprecation

[Eduard Bloch]

Hallo,
* Simon McVittie [Sun, Nov 15 2020, 11:51:25PM]:

Thanks for the insights!

> However, it might be good to get the framework for this in place before
> the Debian 11 freeze, which would look like this:
>
> * source: gdk-pixbuf
>   - binary: libgdk-pixbuf-2.0-0
>   - binary: libgdk-pixbuf-2.0-dev
>   - binary: libgdk-pixbuf-xlib-2.0-0
>   - binary: libgdk-pixbuf-xlib-2.0-dev
>   - transitional binary: libgdk-pixbuf2.0-0
>     + Depends: libgdk-pixbuf-2.0-0, libgdk-pixbuf-xlib-2.0-0
>   - transitional binary: libgdk-pixbuf2.0-dev
>     + Depends: libgdk-pixbuf-2.0-dev, libgdk-pixbuf-xlib-2.0-dev
>
> (That'll require a trip through NEW, of course.)

Okay, sounds like a plan. I guess you will take the lead from here?
Feel free to use this bug report as anchor.

I cannot promise to have enough resources to support on that, though, I
only came along because an upstream project where I collaborate was
stormed by Gentoo users where deprecation process is apparently harsher.

> Then we can do a mass-bug-filing for packages that actively use the -xlib
> sub-library, and (perhaps later) a lower-severity mass-bug-filing for
> packages that build-depend on libgdk-pixbuf2.0-dev but should ideally
> only B-D on libgdk-pixbuf-2.0-dev.

XMAS time becoming a good candidate for a such rollout then?

> > maybe there should be even a warning of some kind printed
> > through pkg-config or GCC warnings (although this is bad, of course,
> > would break -Werror using packages).
>
> I don't *think* we can issue warnings from pkg-config, although there
> might be a mechanism available that I'm not aware of. Do you know of such
> a mechanism?

Not really, sorry. Looked around and couldn't find much useful. We could
invent some new hack here, like extension of FindPkgConfig.cmake (for
cmake users) but all that feels like an evil kludge.

> We can't issue compiler warnings and simultaneously not issue compiler
> warnings, we have to choose whichever is the lesser evil.

Yes. Too much warning is bad, not enough is equally bad.

> If packages in Debian are using -Werror for release builds, I personally
> think that's a packaging bug, because any new gcc release with better
> (or just different!) optimizations or diagnostics can start issuing new
> warnings that would break those packages; so I don't think it would be
> terrible to break them. It would probably be better to break them after
> the Debian 11 release so that people get an entire release cycle to fix
> them, though.

Agreed. Warnings from headers could be used as ultima ratio but only
THEN.

Best regards,
Eduard.