We are talking about this 2-line change in update-rc.d: https://salsa.debian.org/debian/init-system-helpers/-/commit/552e993488a403bf88aa342f73bf0b22ce62ff16
I think it's feasible to add that in buster in the next point release, and with it allow debhelper &co to safely move the .service files also in bullseye-backports. Wouldn't you consider this option? On Mon, 23 Aug 2021, 9:42 pm Niels Thykier, <ni...@thykier.net> wrote: > Michael Biebl: > > Hi Niels > > > > Am 23.08.21 um 08:19 schrieb Niels Thykier: > >> [...] > > > > systemd in buster (v241) does support reading unit files from > > /usr/lib/systemd/system (see systemd-analyze unit-paths). > > The changes to init-system-helpers (namely update-rc.d) to also consider > > unit files in /usr/lib/systemd/system was added in 1.58, i.e. is > > currently only available in bullseye. > > > > This code path in update-rc.d is only used for older compat levels > > though. Newer debhelper versions disentangled dh_installinit and > > dh_installsystemd and we don't use update-rc.d if > > --skip-systemd-native is used, see commit > > cba2a8a6ea64773e61ab41c218853ee729656650 in debhelper. > > > > Thanks for the analysis. :) > > > Also, the code in update-rc.d is only a fallback when the "real" > > systemctl is not available to create the enablement symlinks. > > > > If we hit this code path and update-rc.d does not find the .service > > file, it silently skips the enablement of the service. The package > > should still install successfully. > > > > So is it safe? I'd say reasonably so. > > > > My reptile brain reaction to this is that it smells like "fails to > install correctly and failing to declare to do so" if we do not enable a > system when we should have. > > I get that most installations that do not have systemd are unlikely to > switch to systemd later but I do want it to "just work(tm)". > > > Question is, if we should start moving unit files in > > bullseye(-backports) where everything is installed in /lib from a > > consistency PoV. > > > > Regards, > > Michael > > > > That is the crux of this request. The backport was requested to ensure > consistency between bullseye and buster builds (see the OP for details; > I omitted them in my forward to you as I thought they were irrelevant to > my question). Personally, it is easier for me if both cases use the > same path but only if works for -backports as well. If we are not > certain it is safe, then I will look at using /lib for -backports even > if it means I cannot comply with this request. > > Thanks, > ~Niels > >
