[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#150514: Uer maodifications _must_ bre preserved, even is a co-admin said otherwise a few releases ago

On Mon, Jun 24, 2002 at 01:57:36PM -0500, srivasta@acm.org wrote:
> >>"Gordon" == Gordon Russell <gor@dcs.napier.ac.uk> writes:
>  Gordon> I do ask. Thus I do not violate policy. The MUST clause you
>  Gordon> mention is clarified by the text which follows it (e.g the
>  Gordon> quoted text shown).
> 	But you don't. You retrieve the information from debconf,
> 	where it had been set at some point in the potentially remote
>  past by someone else. Had you been asking everytime, this would not
>  be an issue.

Yes I agree that this seems in breach of Policy. The Policy also goes
on to say something about not asking questions on an upgrade. I disagree
with your philosophy that debian=multi-user and thus system administration
is an uncontrolled multi-user activity. This is not sensible as a procedure,
and I will not go out of my way to support uncontrolled system administration
where the administrators do not know what they are doing and do not talk to
each other.

> 	I also note that you are overwriting /etc/serial.conf without
>  asking, too; I am not usre that this is kosher either.

Again I ask in preconf and when the package is reconfigured. You can set
this info directly in serial.conf, and thus I would argue that user
configuration of this file must be in the appropriate way to make it valid.
I can ask it every time but this is not sensible. I actually like your ucf
system and can see it as being useful for setserial controlling serial.conf.
I may indeed upgrade to using it (providing you get the priority set higher)
but I do not recognise a Policy breach and want to do it only because it
is potentially better.

>  	However, since there is an obvious difference of opinion
>  between us, perhaps this needs be brought up to the general developer
>  body for clarification; is asking a question once, and
>  potentially never asking it again, enough to satisfy the policy
>  requirement of preserving user changes? My reading of the section is
>  that the behaviour has to be similar to how dpkg handles conffiles;
>  and dpkg does ask every time.

To be honest there is a lot of truth is what you say, but I am sure that the
users will not want a debconf question every time (can be done by setting
priority high in the debconf question).

> 	The risk of doing otherwise, especially on a multiuser system,
>  is that one may just copy configuration files between all the
>  machines in ones environment, and, if not all are Debian machines,
>  one may lose changes made after copying the common file and tweaking
>  it for the local box.

/etc is local for each machine by definition.

> 	Note policy stating that common practice is to call the
>  update/configure scripts iff  the config file does not already
>  exist. 

Yes, and quite honestly I do not understand this. Things like serial.conf
and modules.conf may be changed many times in a single uptime to reflect the
system, which differs from a single-creation config file. Perhaps there is
an argument for another style of file in /etc which relates to usually changing,
persistent data, where sometimes they are reconfigured to be user-controlled
files, like the ones we are discussing.

To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org

Reply to: