[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Moving backends to table-driven config...



Howard Chu wrote:

2) What is the config schema numbering plan (if any)? I see that in bconfig.c you are using "OLcfgAt:<digit>", while in back-bdb you're using "OLcfgAt:1.<digit>", and in back-ldif you simply recycle the "dbDirectory" attribute of back-bdb. Should we use a different prefix for different backend types, e.g. a "OLcfgAt:2.<digit>" for back-ldap, recycling stuff in back-meta and so, to allow independent development with minimal conflict resolution? I also note that "OLcfgAt:1" right now is "olcAccess", while "OLcfgAt:1.1", i.e. "olcAccess.1" is "dbDirectory"...

Yes, I wanted to recycle the common terms for back-bdb and back-ldbm, and since back-ldif had only a single parameter I reused that one as well. But for other backends, you can define new ones or recycle as you deem fit. The schema loaders will silently ignore redundant attempts to define the same item.

The general idea was that OLcfgAt and OLcfgOc were for the core/global config items, and everything else would have a sub-arc under those. Part of the reason to use objectIdentifier macros here is that we can easily reassign entire segments of the schema later when a more logical arrangement surfaces. It may make more sense to use a separate arc under OLcfg for the backends and overlays, to keep them out of the core space. Haven't really seen any advantage one way or another.


SO would you suggest to keep the OLcfgAt.1. branch for backends, or futher specialize? I don't see big issues in coordination, because at some point we need to converge to an agreed numbering and that's going to be it, no more changes. One thing about names: the olc* prefix looks fine; other names, like dbDirectory (in general, the db* prefix) sounds a bit too generic and likely already in use by someone else.


I suggest using a new branch for back-ldap/meta, rooted at OLcfg. I'll move back-bdb as well. As for a prefix, I suppose we could just use olc* for everything. Or olb* for backends/databases. I'm open to suggestions.

I suggest:

- OLcfg -> 1.3.6.1.4.1.4203.666.11.1 (as of today)
- OLcfgAt -> OLcfg:3 (as of today)
- OLcfgOc -> OLcfg:4 (as of today)

- global stuff rooted at OLcfgAt:0 and OLcfgOc:0 (i.e., "olcAccess" would be OLcfgAt:0.1 and so on)
- backend stuff could be rooted at OLcfgAt:1 and OLcfgOc:1
- common backend stuff could be rooted at OLcfgAt:1.0 and OLcfgOc:1.0
- each backend type could have an assigned branch for specific configuration directives (e.g. back-bdb could get OLcfgAt:1.1 and OLcfgOc:1.1, back-ldap could get OLcfgAt:1.2 and OLcfgOc:1.2 and so on)
- overlay stuff could be rooted at OLcfgAt:2 and OLcfgOc:2
- common overlay stuff could be rooted at OLcfgAt:2.0 and OLcfgOc:2.0
- each overlay could have an assigned branch for specific configuration directives (e.g. glue could get OLcfgAt:2.1 and OLcfgOc:2.1 and so on).


The only rationale behind this is that there would be a clear separation between what is global, what is common to all backends and what is specific to each backend; moreover, each backend would have an OID assigned. There are grey areas represented by commonality across backend types (I'm thinking od back-bdb and back-ldbm, or back-ldap and back-meta); in these cases, the common directives could go into one of the backends and get shared by the others, or they could go into the common backend-specific branch, or so. In fact we already have "generic" backend-specific directives that only apply to a subset of backends: the "index" or even the "suffix" directive (back-monitor and back-config do not allow suffix), so moving the "dbDirectory" attributeType to the backend-generic branch sould not be an issue.

With respect to attribute/objectClass prefix, I suggest "olc*" for common stuff; "olcDb" for backend-generic and "olcOv*" for overlay-specific, so we save the "olc*" common prefix and add a level of specialization.

If this sounds reasonable, I can start renaming schema objects.

p.

Somewhat on this topic, currently the OID macros aren't preserved in the actual schema elements that use them. While we never expose them in the cn=Schema subentry, I think it would make sense to preserve them in the back-config data. Yes?


I agree; we could keep track of the objectidentifiers defined, and consistently trap all OIDs that are rooted at one objectidentifier and rewrite them using the objectidentifier when presenting back-config data; I think this should add clarity because (well designed) strings are more readable than dotted digits. Of course, the rewriting should go from specific to general; sorting them by string length should suffice.


SysNet - via Dossi,8 27100 Pavia Tel: +390382573859 Fax: +390382476497









SysNet - via Dossi,8 27100 Pavia Tel: +390382573859 Fax: +390382476497