- From: Marcos Caceres <marcosc@opera.com>
- Date: Mon, 20 Apr 2009 14:56:34 +0200
- To: Robin Berjon <robin@berjon.com>
- Cc: Andrew Welch <andrew.j.welch@gmail.com>, public-webapps WG <public-webapps@w3.org>
On Wed, Mar 18, 2009 at 2:14 PM, Robin Berjon <robin@berjon.com> wrote: > On Mar 18, 2009, at 11:33 , Andrew Welch wrote: >> >> - when multiple versions of the xml exist, you need some way to >> differentiate them other than checking for the existence of certain >> elements/attribute > > You state that as a requirement but don't explain what the use case for it > is. Personally, I don't see any while I see several pointing in the other > direction. > >> That's how XSLT 2.0 does it, and I think XHTML 2.0, but I notice XSD >> 1.1 is going down a different route... > > XSLT 2.0 is a programming language (as opposed to a configuration file for > instance), and if you're using 2.0 features you want it to break early (IIRC > they also changed the semantics of some bits). Me, given the changes, I > would've gone with a new namespace but that's water under the bridge. > > XHTML 2.0 is not trying to be backwards compatible, that's HTML5's job. It > initially had a different namespace, and its recent attempt to reverting to > using the XHTML 1.x namespace is IMHO a poor decision that I am rather > confident will be overturned (if it hasn't already). > > The reason there isn't One True Way of doing XML versioning is because the > use cases vary across languages. If you add a <pony> element to SVG it's > probably fine if it's not displayed as you can fallback, but if you add a > <transfer-me-money/> element to a SOAP message you probably don't want > people to ignore it. > > Our usage scenarios for improved configurations don't involve the sort of > stringent versioning that would require variants on @version or > @mustUnderstand. Furthermore, the complexity of using a version attribute > has to take into account the fact that the widget stack is modularised: what > happens when a separate specification (signatures, window modes, etc.) adds > an element to the configuration file (as they well can)? Do you increment > the version? Add other version attributes? Add some tokens as in > version="1.0 +dsig +wm"? > > I understand the draw to flagging versions, it somehow "feels neater", and > that's why people tend to want to throw them in (I just made the very same > comments to Bondi). But for our usage scenarios, it neither helps nor > scales. > FWIW, as the editor, I agree with Robin. Nothing is gained by adding explicit versioning to the spec. Kind regards, Marcos -- Marcos Caceres http://datadriven.com.au
Received on Monday, 20 April 2009 12:57:36 UTC