[OT] registry implementation (was Re: Debian, too easy?)
Russell sez:
} Gregory Seidman wrote:
} > Russell sez:
[...]
} > } and no applications are forced or required to know about it.
} >
} > No! For it to be useful, all applications must use it. That's why there
} > needs to be a good implementation; good means that it is so solid and
} > convenient that it is harder not to use it than to use it.
}
} Well, i kind of meant that old apps. could still run on the system
} in the conventional way, before being adapted for a registry.
Oh, sure. Ideally, though, there would be a translation mechanism.
That's how MacOS X is handling things, and I like it. There are standard
system calls that used to consult a file but now consult the NetInfo
lookupd.
} > } A set of utilities for examining and manipulating it would be useful
} > } too.
} >
} > Absolutely. Therein lies the appeal of using XML.
}
} Linux kind of has a scattering of registries already, eg,
} termcap/terminfo, xf86config, /etc/X11/app-defaults, /etc/X11/xkb,
} various other files in /etc... I'd think a standardized unix registry
} system would be quite doable. It's just like the fhs, but for app.
} settings. However, if the registry is system-wide, then i guess a root
} process needs to run it, which means there might be security problems
} if users wan't to save personal preferences. Maybe users should have
} local copies of registry parts that they need.
Now that I think about it, I think something procfs-like might be the
answer. The user has prefs in .registry or some such, the system prefs
are in /var/registry or something, and /etc is mounted as regfs (which
merges preferences on a user-by-user basis) such that it contains all
sorts of files, e.g. X11/app-defaults, kderc, and others. Everything
continues to look like a normal system, but the configuration is
centralized. The regfs module would have to take plugins to understand
various file formats (printcap, exports, etc.) and any file it didn't
understand would simply remain a file. This means that legacy apps (and
legacy sysadmins) aren't left out in the cold, but new apps can interact
directly with the registry APIs and recent sysadmins could use the
registry tools (or even hand edit the registry files) rather than
dealing with the legacy file formats.
...and all this starts looking more and more like LDAP. Certain parts of
the registry (e.g. the passwords) would have to be system-only,
regardless of what the user's registry had in it; probably there would
be group registries as well.
Doing it right requires astounding complexity; I think it's feasible,
but only just barely and only by a dedicated team of full-time systems
engineers. As I said, I expect to see something similar from Apple
before I see it anywhere else. Microsoft could do it more easily, since
they aren't tied to the legacy Unix file formats, but they have the
albatross of the original Win95 (or was it in earlier versions of
Windows?) system registry and will never be rid of it.
--Greg
Reply to: