How many lost users is an acceptable loss in exchange for systemd?

[Gregory Maxwell]

In many of the recent systemd threads there is an underlying point
which I think is on many people's minds but which I haven't seen
called out. I think this is a generic issue, so it's a but unfair to
single out systemd but it makes a good example.

To say it bluntly:  Any significant infrastructural change _will_ cost
Fedora some users in the short term.

The amount lost depends on the specific feature and the effort put in
to minimize the bugs and disruption, — usually with enough effort and
enough compromises and concessions the number can be made arbitrarily
small but it will not be zero. This is especially the case for
features which make the system more or less unrecoverable inoperable
for 'normal' users when something goes wrong, but it's true more
generally as well.

And I do think that some loss is acceptable— without tolerating some
loss Fedora simply couldn't move forward— no matter how wise or
overdue a change is there will be bugs, differences in taste, and
disruption. There are some people who are continuing to use Fedora
only because learning SomethingElse™ is too much work but when Fedora
changes and they'll have to learn either way Fedora's "advantage"
vanishes to them.

And, yes, the harm won't be equally distributed— it seems to me that
Fedora has ignored quite a bit of harm because it didn't primarily
fall on what the developer's considered a "typical desktop"  (which,
as far as I can tell, really means a particularly narrow set of laptop
hardware with a particularly narrow set of users and use cases).  Why
Fedora keeps chasing a market which Ubuntu has undeniably won is
beyond me— but nevertheless it's not acceptable to pretend that harms
don't exist simply because they don't hit the one use case you care
about most, not unless Fedora is willing to say that people running
servers, developers, and other power users ought to use some other
distribution and that Fedora doesn't care if it loses all of these
users.

Consider systemd— even if far more work goes into it I think we can
admit that it will be very likely that there will be some users with
some weird configurations which won't boot up with it.  We can blame
their weird configurations, hardware, and random packages as much as
we like— but at the end of it some of these users are going to leave
Fedora because of the change.   Some administrators are going to hate
the management changes and switch off Fedora as a result.  We might
all agree that they're lazy or crazy but it is what it is.

These losses can be reduced by making systemd emulate the old stuff
more accurately (at a cost to systemd's long term purity), by more
testing, by providing fallbacks, etc. But some compromises aren't
acceptable, no testing is perfect, and many people will never learn
about the fallbacks.

The same stuff could have been said about kernel modesetting.

The sooner people admit this the sooner people can agree on what the
acceptable loss level is... Without knowing the acceptable loss level
it won't be easy to agree on a release criteria or agree on how much
mitigating compromises are required to get there. Denying it or
blaming other packages just makes the Fedora community blind to the
risks, which is sad since many of them can be reduced and managed.