Re: bonobo-activation cache ...

[Michael Meeks <michael ximian com>]

Hi Darin,

On Tue, 18 Dec 2001, Darin Adler wrote:
> Under what circumstances does the cache need to be invalidated? For
> example, does installation of new packages need to cause that? Or is
> there some other strategy?

	If you install new system software, then you may need to restart
applications that have cached bonobo-activation queries - this is typicaly
a process that is fraught with risks anyway; if you upgrade your WM -
restarting it is a good plan etc. etc.

	There is a technicaly easy solution to this problem - it involves
creating a new client side interface, registering it with the server,
adding an 'invalidate cache' method - essentialy a little bit of coding, a
new IDL file, a new module or two, server mods etc. Whilst the technical
issues are trivial I am not confidant that I can get a change of that
magnitude into bonobo-activation in an ammount of time that makes it worth
doing myself.

	And lastly, this reduces the remote bonobo-activation traffic from
the order of 10 queries per new nautilus window creation, to 0 - resulting
in a fairly substantial improvement to my mind.

> I don't think we should add the cache until we've addressed that
> issue. I've seen this too many times at too many levels of too many
> different projects. The tough issue with a cache is always when to
> discard the cached data.

	In this case - simply quitting / re-starting the application
discards the cache, which is most likely what you want anyway.

	Regards,

		Michael.

-- 
 mmeeks gnu org  <><, Pseudo Engineer, itinerant idiot