Bug#1095967: luanti-server: file loss during migration from minetest-server due to /usr-move (DEP17 P1)

[Travis Wrightsman]

Hi Helmut,

This was my fault as I prepped most of the package changes for this
Luanti rename. I applied your patch [0] and don't see any diverted files
after upgrading from minetest-server 5.8 to luanti-server on a Trixie VM:

$ ls -alh /lib/systemd/system/{luanti,minetest}*
-rw-r--r-- 1 root root 604 Feb 14 12:15 /lib/systemd/system/luanti-server.service
-rw-r--r-- 1 root root 593 Feb 14 12:15 /lib/systemd/system/luanti-server at .service
lrwxrwxrwx 1 root root  21 Feb 14 12:15 /lib/systemd/system/minetest-server.service -> luanti-server.service
lrwxrwxrwx 1 root root  22 Feb 14 12:15 /lib/systemd/system/minetest-server at .service -> luanti-server at .service

However, I observe the same results (no file loss) without the patch so
I'm unsure exactly how to reproduce the conditions under which file loss
will occur. I saw case1.sh under DEP17 uses dpkg directly whereas I used
apt so maybe some magic is happening I don't yet understand.

Cheers,
Travis

[0] https://salsa.debian.org/twrightsman/minetest/-/tree/fix-1095967

On Fri, 14 Feb 2025 13:50:45 +0100 Helmut Grohne <helmut at subdivi.de> wrote:
> Package: luanti-server
> Version: 5.10.0+dfsg-2
> Severity: serious
> Tags: patch
> User: helmutg at debian.org
> Usertags: dep17p1
> Control: affects -1 + minetest-server
> 
> Hi Tobi,
> 
> Your luanti-server upload introduced a /usr-move regression. The systemd
> units were formerly part of minetest-server and are now in
> luanti-server. At the same time, they moved from /lib to /usr/lib. This
> is a classical DEP17 P1 situation. I recommend upgrading the
> Breaks+Replaces relation to Conflicts as well as introducing brief
> protective diversions. You'll find them in the attached patch.
> Unfortunately, I was not able to test them as luanti-server has an
> unsatisfiable dependency. Please verify that no diversions exist once
> luanti-server has been successfully installed. They should only exist
> from preinst to postinst and not beyond.
> 
> Helmut