On Thu, Mar 04, 2010 at 03:05:08PM -0800, Murray S. Kucherawy <msk_at_cloudmark.com> wrote:
> It's actually:
>
> dsn:pgsql://... (for example)
>
> Our code takes the data set name up to the first colon to determine which handler to use (i.e. which type it is) and passes the rest to the handler's "open" code.
Ah, okay. That's what I'll put in README, then.
> But this brings up an inconsistency in that the "open" code for OpenLDAP actually does use the whole string, keying on "ldap", "ldaps" or "ldapi" to select the opener but not dropping it from what gets handled. I did this because LDAP connections are widely specified as URIs already, so it seemed an easy integration step.
>
> To be consistent about it, I could take any OpenDBX scheme before the colon and pass that to the OpenDBX handler. Then only "file", "db", "csl" and "refile" would be special and handled internally, and the rest get routed to specific extension libraries. Doing so would require a hard-coded list though since there's no way to get the list of supported backends from their API.
That's probably the "clean" way to do it. Maybe a way to get a list of
supported schemes is something we could request from the OpenDBX guys.
It's available in odbxlib.c as (odbx_backend_syms)odbx_lib_ops, just not
exposed in the API.
> I'd say this is a 2.1.0 thing at this point, unless someone feels doing so at that time would cause a huge headache. (And we could always teach our stuff to just ignore "dsn:" for a while.)
Yeah, if there's a working scheme now it's definitely not
release-critical. Thanks for clearing up my misunderstanding.
--
Mike Markley <mike_at_markley.org>
Received on Thu Mar 04 2010 - 23:12:58 PST