Hmmm. Somehow I missed the original e-mail (probably just did not read it fast enough before my finger landed on the delete button).
Since you asked... I agree with Rick. For any but the simplest transformations you almost always need to store a representation of at least part of the tree. >From a user perspective I would much prefer one syntax (XSLT) with all of the functionality available. The application could then process as much as possible streaming, but build a tree as necessary. I think that Xalan is very nearly there already. It already does SAX in to SAX out... even incrementally. The missing piece is pruning what has already been processed and is no longer needed (yup, I am still waiting for pruning). The main advantage to a dedicated streaming implementation that I see would be perhaps better performance due to the simpler syntax, but I am sure that the Xalan developers will improve Xalan's performance so that this may not be too great. My $0.02, Art -----Original Message----- From: Rick Bullotta [mailto:[EMAIL PROTECTED]] Sent: Monday, March 04, 2002 1:18 PM To: [EMAIL PROTECTED] Subject: RE: Serial/streaming transformations proposal On paper, it sounds good, but in practice, our experience seems to show that anything but the most trival stylesheets will often require some "in-memory tree representation" to process properly. Stylesheets that are "purely declarative" and are focused primarily on output formatting might benefit, although the moment any slightly complex nesting comes into play, the implementation complexity of a code to support a serial transformation seems to go up exponentially! Rick Bullotta CTO Lighthammer Software (www.lighthammer.com) Voice: 610-903-8000 x14 -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]] Sent: Monday, March 04, 2002 12:19 PM To: [EMAIL PROTECTED] Subject: Serial/streaming transformations proposal An interesting proposal for serial/streaming (SAX-to-SAX) transformations is starting to take shape at http://www.gingerall.cz/stx/stx-draft-0.02.html.; Curious on what people think. Probably we should try prototyping something along these lines soon... though I would prefer basing it on a pure XSLT subset. Or not? -scott
