Le 02/04/2019 à 23:01, Thorsten Glaser a écrit : > Most people using Debian without systemd have APT pinning or other > measures in place that prevent the systemd package, which ships the > systemd-sysusers binary (and service?), from being installed, in > order to not sneakily being converted to systemd (it did happen).
I did some tests in a VM with a minimal install where I switched to sysvinit with: apt install sysvinit-core cp /usr/share/sysvinit/inittab /etc/inittab reboot apt remove systemd In Stretch I can install the systemd package and it won't switch the init system as advertised. In Buster unfortunately installing systemd pulls systemd-sysv through libpam-systemd and the init system is switched. The --no-install-recommends flag has to be used to avoid that. I've filed a bug for systemd (#926316). Assuming #926316 gets fixed, I think we should focus only on providing a usable sysvinit script as required by the policy. Supporting people allergic to systemd and using APT pinning to exclude it is out of topic (they should only exclude systemd-sysv anyway, not systemd). > What is the issue with using adduser, which is the standard Debian > tool doing the same job, instead? After all, depending on systemd > just to create a system user and group is very heavy-weight. There is a growing consensus around the idea that imperative maintainer scripts are a bad thing and they should be replaced with something declarative. systemd-sysusers does exactly that for the user creation, that's why I favored it over the traditional adduser. Regarding the weight, at this point you've already installed the JRE and Tomcat, the few extra MB for systemd are negligible. > OK, removed. > > > OK, reverted. Thank you > Did you have a chance to test this on a buster/systemd Debian? > I don’t currently have such a machine existing in a meaningful > way. (Granted, I could probably cobble together some test VM, > but I’m sure you have something at hand.) I haven't checked yet. Emmanuel Bourg