On Thu, Apr 17, 2014 at 01:08:11PM -0400, Wietse Venema wrote:
> For example, "dc=%{labels:d,}" produces a comma-separated list of
> "dc=example, dc=com" given an email address of u...@example.com.
This specific example works poorly, because the construct in question
is part of a larger single string template, and there is no way to
know how back or forward to go in repeating prefixes and suffixes.
search_base = OU=People,dc=%{labels:d,},o=Acme Inc
Unlike the pipe(8) case, the output is a single string, not an argv
vector. So the "dc=" would have to go inside the construct in
something plausibly less verbose than:
%{labels:d;prefix=dc=;separator=,}
which is all fine and good, but I still have no plausible use-case
for any such generalization.
The Dovecot trick of leaving out the first prefix, is cute, it is
far from clear why one would expect a fixed strategy of a different
RDN type for just the first domain component to work generally for
multiple domains, and especially because the first domain component
is in fact a domain component, and NOT some other kind of name.
So while I understand Wietse's point, no sensible generalization
comes to mind that is not just mere code and complexity with no
plausible application.
I'd like to see a reasonable use-case for doing something else with
domain components in LDAP, and perhaps also MySQL, PgSQL, ...
tables, that generalizes "dc=example,dc=com" with input "example.com",
before writing more general code that may never get used.
I realize this leaves "%," deadlocked for now between opposite
poles of "too specialized" and "needlessly over-generalized".
Resolving this requires either new insight to find the more general
use-cases, or sufficient evidence to conclude that nothing more
general is likely to emerge.
--
Viktor.
