XSLT 2.0 and XPath 2.0 Programmer's Reference, 4th Edition (447 page)

Here are some other examples of simple mapping expressions at work:

Expression	Description
sum(order/(@price*@quantity))	Returns the result of calculating @price times @quantity for each child and summing the results.
string-join(ancestor::*/name(), ‘/’)	Returns a string such as book/chapter/section/para giving the names of the ancestors of the context node, separated by / characters.
avg(product/number(substring-after(@price, ‘$’)))	Returns the result of taking the @price attributes of all the child elements, stripping off the leading $ sign, converting to a number, and then averaging.

In most cases the system will be able to distinguish a path expression from a simple mapping expression by the way it is written. In some cases, however—for example, if the expression calls a user-written function whose return type is not declared—the two cases can't be distinguished until runtime. The initial processing in both cases is the same: for each item in the sequence returned by the first expression, evaluate the second expression and concatenate the results. The rule is then:

• If all the items in the result are nodes, we have a path expression. Duplicate nodes are eliminated, and those that remain are sorted into document order.
  
• If all the items in the result are atomic values, we have a simple mapping expression, and we are done.
  
• If there's a mixture of nodes and atomic values, an error is reported.
  

A simple mapping expression always maps from nodes to atomic values. You can't have atomic values on the left-hand side of the
/
operator. For example, you can't write
avg(*/name()/string-length())
to get the average length of the names of the child elements. Only the last step in the path can return atomic values, so you have to write this as
avg(*/string-length(name()))
. Why the restriction? Possibly because some people didn't like the consequence that the value of
1/10
would be 10.