[Pkg-rrfw-general] Re: Unofficial Debian packages for RRFW 0.1.7 available

[Jurij Smakov]

Hi Marc,

On Sun, 15 Aug 2004, Marc Haber wrote:

> [Did you e-mail off-list deliberately? I think this discussion needs
> to be on-list]

No, I've just forgot to CC:, including it now.

[...]
> I have never seen a package done by somebody new to Debian and so
> cleanly done. Congratulations. Where did you learn so much about
> Debian so fast?

Well, I have only started doing packaging a couple of months ago, but I've
been lurking on mailing lists and IRC for well over a year now, so I would
not say it was fast learning. And, as you point out, the packages still
have quite a few quirks :-).

> A few more things:
>
> -common needs to depend on netbase, since rrfw_devdiscover fails
> without /etc/services.

I'll fix that. By the way, I am curious how you've discovered that, since
the ordinary chroot produced by debootstrap has netbase installed. Do you
have a specially crafted chroot with only Essential installed?

> rrfw_devdiscover writes routers.xml to /usr/share while the file
> should go to /etc. First, the local admin might want to change that
> file, second, /usr should probably only have files from the
> distribution, and third, /usr might be mounted r/o. Maybe the entire
> xmlconfig directory should be symlinked to /etc. But actually, I think
> that all files put to /usr/share/rrfw should better be in /etc.

I have spent quite some time thinking about what's the best way to do
that. I didn't want to place the entire /usr/share/rrfw stuff into
/etc/rrfw because most of these files are not intended to be modified
by the user. Currently only the modifiable files are placed into /etc/rrfw
(such as *-siteconfig.pl) and the symbolic links are created in
/usr/share/rrfw so that RRFW can find them. Since it expects all the files
to be in the same directory, the only place where the user-created tree
definitions can go is /usr/share/rrfw/xmlconfig. The ideal solution would
be if RRFW would look for the configuration files in two different
locations, so we can cleanly separate the read-only and modifiable
configuration files (by placing the former in /usr/share/rrfw and the
latter in /etc/rrfw). But if you think that placing it in /etc is ok, I
could change it pretty easily.

> If the daemon stop process might take more than the ten seconds it
> takes on my test box (configured for a single host being queries), you
> might want to steal some "progress showing" code either from squid,
> zebra, or quagga's init script just to show the user that there is
> still something happening.

Yes, that's a good idea, however I am not sure if any progress indication
is policy-compliant. After all, section 9.4 of the policy describes how
the messages from the init.d scripts should look like in quite detail.
Also, the long shutdown times are documented/explained in README.Debian
(question 6) and, as explained there, are user-configurable through the
variables in /etc/default/rrfw-common.

> The account name rrfw might be bad since there is potential for
> conflict. I'd prefer having something like Debian-rrfw, but be sure to
> read up on the flamewars about that topic before you change anything.

Yes, I have read the exim story :-) and don't quite understand why people
make such a fuss out of it. The only thing which I would change is using
all lowercase and keeping it at or under 8 symbols, so that there is less
chance to break stuff. How does rrfw-adm sound?

> The account rrfw should probably be have rrfw as primary group.

Yes, overlooked that. Thanks.

> The log files should probably be rrfw:adm instead of rrfw:rrfw.

I'll fix that.

> Greetings
> Marc

Thanks for all your comments,

Jurij Smakov                                        jurij@wooyd.org
Key: http://www.wooyd.org/pgpkey/                   KeyID: C99E03CC