Excellent discussion!

Question: Can we create best practices for XPath that are independent
of the host language?


I would like to pursue the ideas that David Carlisle has put forth.
(David, please correct any errors in my representation of your ideas)

Consider this XPath statement:

    if (//airplane[@tailnum='C3H1'] instance of
schema-element(airplane)) then 
            //airplane[@tailnum='C3H1']/altitude * .3048  
    else 
        if (//airplane[@tailnum='C3H1']/altitude[@unit='feet'] castable
as xs:double) then 
            //airplane[@tailnum='C3H1']/altitude * .3048
        else
            'Error'

There are three kinds of activities going on in the XPath statement:

1. There is data processing: the altitude value is converted from feet
to meters: 

    //airplane[@tailnum='C3H1']/altitude * .3048  

2. There is an XPath check: check that there is an airplane element
with a tailnum attribute having the value 'C3H1', and it has a child
altitude element with a unit attribute having the value 'feet', and its
value can be interpreted as an xs:double value:

    if (//airplane[@tailnum='C3H1']/altitude[@unit='feet'] castable as
xs:double) then ...

3. There is a schema-validation check: check that the host language has
schema-validated the input data:

    if (//airplane[@tailnum='C3H1'] instance of
schema-element(airplane)) then ...

The three activities are combined into one XPath statement.

David asserts that best practice is to keep the three activities
separate.

The advantage of keeping them separate is that the XPath will be easier
to read and maintain.

Thus, when you want to compose an XPath statement to convert the
altitude value from feet to meters, you simply write this XPath: 

    //airplane[@tailnum='C3H1']/altitude * .3048

The XPath to perform a schema-validation check and the XPath check
should be done elsewhere.  (This is pretty vague.  Can we better
characterize what this means?  Where would these checks be done?  If we
want "portable XPath statements" -- XPath statements that can be picked
up and dropped into other documents -- how robust would an XPath
statement be if it's separated from its checks?)

Here's a summary of David's proposal:

When designing an XPath statement, do not combine data processing,
XPath checks, and schema-validation checks.  Instead, keep the three
activities separate.

Do you agree with this?

/Roger