On Wed, 2010-07-21 at 22:13 -0400, Chuck Anderson wrote:
On Thu, Jul 22, 2010 at 03:49:21AM +0200, Lennart Poettering wrote:
> The logic behind chkconfig is exposed in many ways in the user
> interface, for example in the chkconfig command line, e.g.
> commands such as "resetpriorities", and stuff like that.
I think having some level of command-line user-interface compatiblity,
even if not 100%, is important. "resetpriorities" is probably so
rarely used that the systemd version of chkconfig could probably just
make it a no-op (unless it was dealing with old sysvinit scripts, in
which case it could probably just call through to the old chkconfig).
But for basics such as "chkconfig service on|off|--list", there should
be compatibility. Runlevels could perhaps be dealt with by mapping
the common cases like 3,5 to the multiuser-target, graphical-target,
etc. Likewise for "service foo start|stop|reload|condrestart" etc.
Mostly I agree. Just a small modification: we don't really have to care
about how often the commands are used *now*, just whether they would
make any sense at all with systemd. resetpriorities clearly wouldn't, so
with a full systemd system, resetpriorities should simply print an
informational message to the effect that it's an absurd command. With a
mixed system where systemd is still handling some old-school sysvinit
scripts, it should reset the priorities of those.
> Jeez. I guess you cannot be helped. I guess it doesn't help
in any way
> here to mention that we were two steps ahead of you here and provide
> "systemctl show" which can be used to introspect systemd units in all
> details and properties in a parsable way. Also, we added "systemctl
check"
> which prints a one line super-short status.
Well, there is some merit in the already stated argument for having
good UI design. In this example, you could have used long-standing
precedent of using -v -vv -vvv (or -q -qq -qqq, or --verbose=N)
arguments instead of "status" "show" "check". Now
you've created new
lore of needing to know when to use "status" vs. "show" vs.
"check",
what the differences are between them, and what their order of
increasing verbosity is.
> But anyway, I give up. If you keep looking long enough you'll find
> something you don't like in everything.
Please don't give up. I think most of this is valid criticism that
should be voiced, but that doesn't imply that the overall architecture
and implementation of systemd isn't great. I like most of what I see,
but I do cringe at some of the command, argument, and option naming
choices that were made. And I think it is asking too much for users
to have to know to use systemd-install for some things and
chkconfig/service for others, not to mention needing to update all the
RPM spec files for new scriptlets, or fixing anaconda to know when to
call which (I'm guessing that anaconda might call chkconfig or
service, but don't know for sure. Are you sure you know where all the
skeletons are hiding?)
Right. Please try to take the constructive bits from the criticism,
Lennart, and just leave the rest. You don't need to engage with it.
--
Adam Williamson
Fedora QA Community Monkey
IRC: adamw | Fedora Talk: adamwill AT fedoraproject DOT org
http://www.happyassassin.net