On Wed, 06 Sep 2023 at 19:35:46 +0200, Gioele Barabucci wrote:
> If there is consensus, I volunteer to create a MR to move
> autopkgtest-build-podman and related files to a separate package that can
> then Depends: on podman, dbus-user-session and all other packages required
> to make it work out of the box.
I don't think this is a good idea: this design would mean autopkgtest
needing to go through the NEW queue and be arbitrarily delayed by ftp
team review every time it gains a new optional backend with non-trivial
dependencies. (Currently that's: docker/podman, lxc, lxd, qemu, schroot,
plus perhaps some of the autopkgtest-build* tools have requirements that
are not the same as their corresponding autopkgtest-virt- backends.)
Also, autopkgtest-virt-docker and -virt-podman are the same script,
running in a different mode, with different dependencies, and the same
is true for their autopkgtest-build-* equivalents. Should the package
containing them depend on Docker, or Podman, or both, or use an "|"-group?
None of those possible answers really seems right to me.
The ftp team also seems unlikely to be impressed by a new binary
package for around 20K of scripts (plus man pages).
If you want to help with this, I think a better MR would be one to make
a-b-podman and a-v-podman log a user-comprehensible warning mentioning
dbus-user-session, if they are started in podman mode while the socket
$XDG_RUNTIME_DIR/bus doesn't exist.
(Or if you can figure out exactly what circumstances require
dbus-user-session, maybe even exit early with an error under those
circumstances - but I wouldn't want to get that wrong and prevent people
from using podman in a situation where actually it would have worked fine,
so please be conservative, or at least have some sort of --force option.)
smcv
