Bug#935304: libpam-systemd: Please relax Depends: systemd-sysv

[Sam Hartman]

>>>>> "Michael" == Michael Biebl <biebl at debian.org> writes:

    Michael> Tbh, I'm not sure what kind of answer you expect from me.

    Michael> I guess I already provided my feedback here and mentioned
    Michael> what kind of solution I prefer. I can repeat this in this
    Michael> bug report, but I'm not sure if this is helpful.

Are you referring to the idea of   using libsystemd0 and having elogind
use the same dbus interface so be able to reuse libsystemd0?

If so, Mark explained  why that didn't work in #940034.
I think when you originally raised the concern Mark may not have
entirely  understood what you were thinking about.  But at least if I
characterized things correctly above, Mark did fully explore that option
in #940034.

A brief summary is that libelogind0 does basically use the same dbus
interface as libsystemd0.  However, libsystemd0's interface requirements
extends beyond dbus; there are a number of functions that for example
are implemented purely in terms of cgroup membership tests.  Elogind's
interface diverges among other reasons because elogind has a different
cgroup hierarchy.


I guess there's one way that we could use lybsystemd0 while running
elogind.  If libsystemd0 were patched to understand both the elogind and
systemd cgroup interfaces, I suspect that would be enough that you could
use libsystemd0 at least for applications that do not want elogind
specific functionality.  I suspect such patches are small, but my
presumption is that the systemd maintainers would not want to carry such
patches.  Exploring those patches hasn't really been done I think
because everyone involved assumed that would not be the direction the
systemd maintainers would want to go.

So, what answer would I expect?  A number of possibilities spring to
mind:

1)  "I haven't read Mark's analysis of why libsystemd can't be used and
am not likely to have time right now.  I regret that I cannot help."

2) "Ah, I understand, but I don't have any ideas of how to fix this;
perhaps something will come out of the debian-devel discussion."


3) "Mark missed something important; you could get lybsystemd to work if
you just did X."

I guess you could also say that the Debian maintainers should rewrite
elogind to use systemd's interfaces, although honestly it seems like
that's a huge burden to try and stick on package maintainers.
Upstream, elogind wanted the integration point to be the elogind library
API.  They made their library ABI compatible for us, but my
understanding is that they designed their product so that  you'd build a
given app against either libsystemd0 or libelogind0...  I understand
that our dependency management makes it tricky to do multiple builds of
the related apps to make that work well.

It's totally fine if you don't have answers.  My hope is that you will
take the time to realize that Mark seriously considered your advice and
tried to make it work.  And also, if you do have bright ideas, please
share.:-)

Thanks for listening.