Re: real LSB compliance
On Tue, Jul 03, 2001 at 06:10:50AM -0500, Scott M. Dier wrote:
>
> > * LSB defines runlevels, while we define 2-5 identically by default.
>
> Isn't this a specific redhat-ism? I'm not impressed with using init to
> handle xdm/gdm.
Actually, most other Distributions are also using this runlevel
scheme. Debian was the the only distribution which wasn't.
> Since init scripts in rc?.d are just links, why not have a directory
> called 'vendor' and have vendors create directories for their name of
> their company under it and put their init scripts under it. Have the
> vendor-dirs assigned instead to give them a namespace. Have the init
> handling scripts handle namespace collisions at the rc?.d level. When
> removing just look for the symlink that refrences the script your
> looking for.
We didn't want to do things that were completely new and different
without very good reasons. We did specify an init.d installation
script because every distribution uses different S50/K90 numbers to
control ordering. (For no good reason but an accident of history,
but no-one seemed to want to deal with the pain that would be
necessary to standardize the S??/K?? numbering scheme.)
But in the case of the /etc/init.d/* namespace, a lot of system
administrators are used enabling and disabling system services by
typing commands such as "/etc/init.d/apache stop" or
"/etc/init.d/bind9 restart".
Creating a new directory for the LSB packages would mean segregating
them off into a "ghetto", so that users would have to type something
else (say, /etc/lsb-init.d/oracle), where most other things would have
been /etc/init.d. The same reasoning is why we didn't simply require
all lsb-compliant packages to use an "lsb-" prefix. It didn't seem
fair to force LSB packages to use a new and awkward namespace,
especially when that namespace was going to be user-visible.
In places where the namespace management wasn't going to be
user-visible --- such as in /etc/cron.* directories, for example ---
it was much easier to solve those problems by simply require LSB
packages to use a registered prefix.
Finally, I want to underline again the fact that the discussions on
this subject have been happening for over a year, and most of these
issues (runlevels in particular) had been settled a long time ago.
Many other distributions (SuSE and Caldera in particular) had already
started making changes to their distributions in preparation for LSB.
Where were all of the Debian developers back then when we were
actually discussing these issues? It was an open process, and you
could have affected the course of the standard back then. (There have
been a number of very good points that were raised in this thread; I
just wish they were raised a year ago.)
Where were all of the Debian developers when we started the one month
review process before the final standardization of 1.0? We received a
lot of comments and carefully considered all of them before putting
out the 1.0 standard. Although it would have been much more
convenient to have received these sorts of comments a year ago, it
still would have been much easier for all concerned if we had received
these comments a month ago, instead of now, after LSB 1.0 written
specification has been released.
Unfortunately, human nature being what it is, most folks refuse to
actually pay attention to a standard until after it's finally released
--- at which point they start kvetching. Well, if you don't like what
happened with LSB 1.0, please help us with LSB 1.1! Volunteers are
always appreciated.
- Ted
Reply to: