- From: Richard Tobin <richard@inf.ed.ac.uk>
- Date: Thu, 6 Sep 2007 00:22:52 +0100 (BST)
- To: Norman Walsh <ndw@nwalsh.com>, public-xml-processing-model-wg@w3.org
> | I disagree. Not using a predictable algorithm in XSLT's generate-id() > | is just for efficiency reasons, and is very inconvenient for users > | (e.g. when comparing output from different XSLT processors, or even > | different runs of the same XSLT processor). I don't see any such > | efficiency consideration in our case. > The problem with sequential integers is that if you apply the step to > two different documents and then "p:wrap" them together, you're > basically guaranteed to have duplicates. > > My own implementation works hard to prevent this, so I'm in favor of > retaining the "implementation dependent" flavor. Um. Having one implementation where you can rely on that doesn't seem very useful: people will write pipelines that work in your implementation and then find they don't work in others. It's like relying on argument evaluation order, or how a++ + a++ comes out in a given C compiler. If there's a need for non-duplication between documents it should either be required or there should be a switch to guarantee it. -- Richard
Received on Wednesday, 5 September 2007 23:23:23 UTC