Re: FreeTDS support for table lookups

Victor.Duchovni@MorganStanley.com:
> On Sun, 9 May 2004, Wietse Venema wrote:
>
> > Victor.Duchovni@MorganStanley.com:
> > > - All of multi-parameter map types support both external ".cf" and
> > > "mapsource_" prefix notation in main.cf.
> >
> > The "mapsource_" prefix notation is deprecated. It is confusing
> > because the parameters don't show up in postconf(1) output, and
> > it encourages storage of passwords in main.cf.
>
> OK, if you insist (an alternative is to enhance "postconf" to output
> parameters that are defined in main.cf). The key thing I want to add is
> ${parameter} substitution even when the map is being defined in external
> files. If it were up to me, I would use ${variable} only for main.cf
> parameters and use %[alphabet-soup] for key/result substitutions. Likely
> that is too radical a change, so we will need to support both with
> ${special-variable} taking precedence over ${parameter}.

${maindotcfname} expansion is useful; I no longer insist that lookup
key fields ($domain, $user, etc.) be expressed in the same manner.

> > > - Parameters that hold passwords are read from a file when the parameter
> > > value starts with a "file:/" prefix.
> >
> > There are other main.cf parameters (myhostname) that also need file
> > name support. This would avoid the need for Debian's /etc/hostname hack.
> >
> > Introducing file name syntax for non-map lookup needs to be thought
> > out carefully to avoid conflicts (either in code or in the user's
> > brain) with map lookup.
>
> Yes, I think this makes sense.

main.cf parsing needs to be revised. The first revision is invisible:
set defaults properly before doing $name expansion. This would make
it possible to implement "postconf -e" (show expanded parameter
values) which is currently not possible.

The second revision is visible and introduces quoting (and thus
syntax versioning). But, that's unlikely to happen in 2.2 since
I would like to catch up with lost time in the release schedule.
Spring is an inconvenient time for me to do a Postfix release.

> I really do think that "postconf" and "postconf -n" (but clearly not "-d")
> should show ad-hoc parameter values.

postconf is a scary piece of code that needs revising too.

Wietse