xsltproc does take the same amount of time to process the transformation, and the stats show that on a 4000 entry xml file, each of my "Row" entries takes one-tenth of a second to process. This accounts for the 6-minute transformation time I'm seeing.
Now I have to figure out why each row is taking so long... I am doing an xsl sort on a numeric value, and each row displays several columns each of which is formatted with a call-template function. At this point, I'm going to limit the amount of data returned and leave it at that. Thanks for the tips, Alan > -----Original Message----- > From: S Woodside [mailto:[EMAIL PROTECTED] > Sent: Friday, July 04, 2003 8:39 PM > To: Robin Berjon > Cc: [EMAIL PROTECTED]; Alan Edel > Subject: Re: AxKit performance problem > > > > On Friday, July 4, 2003, at 04:55 AM, Robin Berjon wrote: > > > Alan Edel wrote: > >> Some additional info: > >> - This is a dual-cpu system, and is not busy at the time of the > >> request. > >> - An xml file of 3,750,509 bytes will cause this problem to occur. > >> - An xml file that's 984,519 bytes takes a LONG time to process, but > >> does > >> eventually complete. In this case, the request starts at 13:30:59 > >> and > >> finishes at 13:38:01. There appear to be three choke points: > >> a. xs_get_styles_str() -- this step takes 15 seconds > >> b. [LibXSLT] performing transformation -- this takes OVER 6 MINUTES > >> (from 13:31:15 to 13:37:53) > >> c. [LibXSLT] outputting to $r -- this takes 8 seconds > > > > What does your stylesheet do? It is terribly easy to write non-optimal > > stylesheets. It won't be noticeable on small files, but on larger ones > > it becomes terrible. I remember just tweaking an XPath a little bit > > and bring a processing from 4 minutes down to a few seconds. > > xsltproc has a nice option, --norman, that will profile your code too, > it's great :-) :-) > > simon > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
