Common spec/interface for file metadata

Dave Cridland dave at cridland.net
Mon Sep 5 22:20:47 EEST 2005


On Mon Sep  5 20:12:45 2005, Jamie McCracken wrote:
> Nicolai Haehnle wrote:
> 
>> Since your concern seems to be performance, why not define a 
>> *plugin* interface for dynamically loaded shared objects instead 
>> of doing IPC? This allows you to avoid some overhead while still 
>> providing interoperability.
>> 
>> 
> unfortunately it wont buy you anything at the moment. If 
> KDE4/Kat/Tenor use postgres - you lose cause postgres itself is out 
> of process (its a daemon only). If you use my tracker it has to be 
> out of process cause mysql is GPL and for Beagle its out of process 
> too as its managed code!
> 
> (not to mention the other showstopper like no async access for 
> Nautilus with plugins)

No reason not to have two standard methods for accessing the API, is 
there? One dlopen(), one DBUS (or whatever)?

Should be easy to bridge between them. Apps requiring low-level, 
synchronous, fast access would prefer the dlopen() API (which 
probably bridges to DBUS, for now). Apps requiring asynchronous, rich 
access (like, say, searching) might prefer the DBUS API (which might 
well bridge to the dlopen() API in the future).

The tricky bit, as has been pointed out, is agreeing on the semantics 
of the metadata.

Dave.



More information about the xdg mailing list