Plugin Manager in Kerberos

[Luke Howard]

On 10/02/2010, at 2:25 PM, Sam Hartman wrote:

> So far I haven't seen a justification for anything as grand as what
> you're talking about.

I agree with Sam. Were one sufficiently resourced to contemplate something grand, I like the idea of servers where the core itself is implemented as a plugin; everything is replaceable. The MIT code, like anything that has evolved over a long time, encapsulates a lot of history and (arguably) brittleness -- so methinks this would be a pretty risky project.

> 1) Loading plugins involves a bit too much code and to much
> specification of directory names  near the code that is specific to the
> particular service.

I agree; I quite like the autodiscovery mechanism but sometimes it's not the easiest API to use. Also I might have preferred if struct plugin_dir_handle was a pointer rather than a structure.

-- Luke