atom feed18 messages in org.oasis-open.lists.docbook-appsRe: [docbook-apps] gentext.template e...
FromSent OnAttachments
Stefan KostMar 8, 2010 6:28 am 
Bob StaytonMar 12, 2010 12:08 pm 
Ron CatterallMar 12, 2010 3:38 pm 
Stefan KostMar 12, 2010 5:28 pm 
Sam FischmannMar 12, 2010 5:38 pm 
Stefan KostMar 12, 2010 8:04 pm 
Bob StaytonMar 13, 2010 10:00 am 
Jirka KosekJul 13, 2010 11:17 am 
deannelsonJul 13, 2010 12:05 pm 
Jirka KosekJul 13, 2010 2:49 pm 
Sam FischmannJul 13, 2010 4:05 pm 
Jirka KosekJul 14, 2010 10:01 am 
Sam FischmannJul 14, 2010 10:27 am 
deannelsonJul 14, 2010 1:42 pm 
Jirka KosekJul 20, 2010 3:45 pm 
deannelsonJul 20, 2010 4:58 pm 
Sam FischmannJul 20, 2010 5:19 pm 
Jirka KosekJul 20, 2010 5:22 pm 
Subject:Re: [docbook-apps] gentext.template eats 1/3 of CPU
From:Stefan Kost (enso@hora-obscura.de)
Date:Mar 12, 2010 8:04:21 pm
List:org.oasis-open.lists.docbook-apps

Sam Fischmann wrote:

I, for one, would be interested in seeing a customization layer that implements some of your patches. Do you think you could provide a link some time?

http://git.gnome.org/browse/gtk-doc/tree/gtk-doc.xsl

check the templates from here on http://git.gnome.org/browse/gtk-doc/tree/gtk-doc.xsl#n59

I got the major speedup from hardcoding 'en'.

Stefan

-Sam

On Fri, Mar 12, 2010 at 5:28 PM, Stefan Kost <enso@hora-obscura.de <mailto:enso@hora-obscura.de>> wrote:

thanks for the reply. Thanks to my profiler I could make several optimizations that don't make the stylesheets worse or loose on functionality. Right now I have them in my customization layer. Not sure if anyone would be interested in patches.

Main problem with gentext right now is also that it does not remember any lookups (memoization in functional languages). Dunno if such thing is doable in xslt. I could certainly see the same lookups several time.

What I am now looking into is writing a xslt pre processor, that reads the xsl + parameters, resolved includes + imports, kicks unused templates, replaces statics and so on. Then that pre processed stylesheet should be faster. Lots of work, lets see :)

Stefan