- From: Sarah Wilkin <swilkin@apple.com>
- Date: Wed, 4 Feb 2004 10:27:30 -0800
- To: Michael Kay <mhk@mhk.me.uk>
- Cc: public-qt-comments@w3.org
I see what you mean. However, I don't want the proposed solution to get in the way of the problem. Our new proposal: If each evaluation of E2 results in a sequences of nodes, they are combined in document order, removing duplicates. If each evaluation of E2 results in a sequence of atomic values of the same type, they are returned in the order of the result from the primary expression. Otherwise a type error is raised. --Sarah On Feb 3, 2004, at 3:24 PM, Michael Kay wrote: > >> The rule seems simple: >> If the path ends with an AxisStep, proceed as usual (eliminate >> duplicate nodes and return in sequence order) >> If the path ends with a FilterStep, order the sequence in the same >> order as the result from the primary expression. > > I think it's not generally a good idea for the semantics of an operator > to depend on the syntactic form of its operands; it should depend only > on their value and type. Otherwise you get a lack of composability: you > can't safely replace an expression with a function that evaluates that > expression. > > It also means that A/B gives a different answer from A/(B). > > Michael Kay >
Received on Wednesday, 4 February 2004 13:27:31 UTC