Re: Performance ideas (resend) ...

[Michael Meeks <michael meeks novell com>]

On Tue, 2009-03-03 at 23:12 -0500, Matthias Clasen wrote:
> >        For gconfd to be faster - we have to read and parse less data.
>
> Certainly one of the bigger wins in this area would be mpitti's
> delocalization patch, if we can fix it up to actually work ?!

	Well - it would really help accelerate the gconftool-2
--makefile-schema-install treacle-pouring, and of course shrink the size
of our packages by not duplicating strings; but in terms of size to
parse I have:

$ ls -lSh %gconf-tree.xml 
-rw-r--r-- 1 root root 2.4M 2009-02-27 16:02 %gconf-tree.xml
-rw-r--r-- 1 root root 909K 2009-02-27 16:02 %gconf-tree-el.xml
-rw-r--r-- 1 root root 847K 2009-02-27 16:02 %gconf-tree-gu.xml
-rw-r--r-- 1 root root 784K 2009-02-27 16:02 %gconf-tree-fr.xml
-rw-r--r-- 1 root root 768K 2009-02-27 16:02 %gconf-tree-ja.xml

	So - we would save up to 30% of the time (really nice to have); but
around 94% of the time of the gconf package install stuff in each
RPM ;-) which is even better.

	If we could then split out the localised defaults from
%gconf-tree.xml (somehow) that makes things another 700K smaller. I
still don't see conceptually why they all need to be in the one file,
surely we can look for a default in a lang-specific file, after all we
can only return a single default for the current lang ;-)

	HTH,

		Michael.

-- 
 michael meeks novell com  <><, Pseudo Engineer, itinerant idiot