[spicysharkco]

On <item a="1" b="2" c="3"/>, /root/item/@*[1] selects all three attributes, while @*[2], @*[3], attribute::*[2] and @*[position()>1] select nothing. (/root/item/@*)[2] already works.

XPath 1.0 section 2.4 evaluates a predicate with "the proximity position of the node in the node-set with respect to the axis as the context position".

getNodePosition returns 1 for a query without a position() method (query.go), and attributeQuery has none, unlike the child, ancestor, descendant, following and preceding queries. So [1] matches every attribute and [2] none. attributeQuery now counts the attributes that pass its node test, from 1 again for each element, as childQuery does.

                             before    after
/root/item/@*[1]             1, 2, 3   1
/root/item/@*[2]             (empty)   2
/root/item/@*[3]             (empty)   3
/root/item/attribute::*[2]   (empty)   2
/root/item/@*[position()=2]  (empty)   2
/root/item/@*[position()>1]  (empty)   2, 3

Across item elements with the attributes a b c, a b and c (values 1 to 6), //@*[1] now selects 1, 4 and 6 instead of all six, /root/item/@*[2] selects 2 and 5, and /root/item[@*[2]]/@a selects 1 and 4.

last() is still wrong on this axis: @*[last()] now selects the first attribute of each element, where it used to select all of them. That is a separate change.

Test_attribute_position adds the attributes in a fixed order, since employee_example adds those of an element in map order. The commented-out //attribute::*[1] case in Test_attribute is enabled as a count (7), since its tag list depends on that order. Both tests fail on master.