Allow self::(a|b|c)

[michaelhkay]

After an explicit axis specifier (including the abbreviated axis specifier "@", but not including the default axis specifier), allow a composite NodeTest that consists of a |-separated list of NodeTest's in parentheses. For example

@(id|name)

ancestor::(section|chapter)

child::(comment()|processing-instruction())

The composite NodeTest matches a node if any of its constituent NodeTest's matches.

[joewiz]

This would be great! This would be a far more natural way to express NodeTests like these.

[michaelhkay]

Here's the detail.

AbbrevForwardStep ::= "@"?  NodeTest
changes to

AbbrevForwardStep  ::= "@"  NodeTest | SimpleNodeTest
NodeTest ::= SimpleNodeTest | ParenthesizedNodeTest
ParenthesizedNodeTest ::= "(" SimpleNodeTest ("|" SimpleNodeTest)* ")"
SimpleNodeTest ::= NameTest | KindTest

(Note, I've chosen not to allow child::(A | (B|C) ) - no nesting allowed or needed.

In 4.6.2.2 we need to add "A ParenthesizedNodeTest matches a node if any of its constituent SimpleNodeTests matches the node".

Apart from that it's just a case of adding examples and notes.

XSLT needs to pick up the change in the syntax for patterns. A pattern that uses a ParenthesizedNodeTest has a default priority of +0.5.

[michaelhkay]

Note that most of the time, this construct is just syntactic sugar: @(A|B) is equivalent to @A|@B.

However, with a reverse axis, it does add expressive power. ancestor::(div1|div2)[2] is not easily expressed any other way, because (ancestor::div1|ancestor::div2)[2] has a different meaning (the union operator sorts nodes into document order before applying the predicate).

[michaelhkay]

This proposal overlaps issue #23. It's wider than issue #23 in that it allows unions of NodeTests, not just NameTests. But it's narrower in that it doesn't allow multiple names in an elementTest (for example element(A|B|C)).

[michaelhkay]

This change has been implemented.