Marc Haber wrote... > My idea would be to have two distinct systemd units, ser2net and > ser2net-oldconfig, with appropriate ConditionPathExists directives.
From my experience, that is a bad idea. While being easier right now,
this results in having to maintain two flavours for ... possibly
forever. In my opinion it is better to do a huge but one-time effort
than having to deal with repeated little work for a long time. Also, I
found the latter in similar situations fairly annoying every time I had
to think about it, but perhaps that's just me.
So I advise to provide an upgrade path for all users even if this means
work for you as the package maintainer. Still in my experience it's
better if the maintainer spends time on a sound automated solution
than leaving figuring out the gory details to the users. For me, this
is also an implication of the Social Contract ("guided by the needs
of our users").
To ease the job:
Upstream should implement a function every decent daemon ought to have:
Dump the configuration as it was parsed, then exit. This is very helpful
when debugging some weird configuration issues, escpecially typos in
file names. Also, unless there's already some "dry-run" option, this
allows using validate: in an Ansible playbook so configuration errors
can be detected *before* a daemon is restarted. And in that particular
case of configuration file format change, it allows users to verify
they've upgraded their configuration to the new format sucessfully.
As a bonus, that dump could be a valid configuration file. Aaaand ...
there is the converter tool that should exist anyway.
As a last resort: The old format is extremely simple, the new one is not
that complicated. It seems feasible to do the conversion with a few
lines of awk, so it could be done in preinst. Also, that method could
even preserve any comments.
All this is more work than some alert boxes or a paragraph in
README.Debian. But I think it's worth it.
Christoph, ser2net user since 2004
signature.asc
Description: PGP signature
