Question

Consider the XML and SQL:

declare @xml xml = '
<root>
    <person id="11272">
        <notes for="107">Some notes!</notes>
        <item id="107" selected="1" />
    </person>
    <person id="77812">
        <notes for="107"></notes>
        <notes for="119">Hello</notes>
        <item id="107" selected="0" />
        <item id="119" selected="1" />
    </person>
</root>'

select  Row.Person.value('data(../@id)', 'int') as person_id,
        Row.Person.value('data(@id)', 'int') as item_id,
        Row.Person.value('data(../notes[@for=data(@id)][1])', 'varchar(max)') as notes
from    @xml.nodes('/root/person/item') as Row(Person)

I end up with:

person_id   item_id     notes
----------- ----------- -------
77812       107         NULL
77812       119         NULL
11272       107         NULL

What I want is the ‘notes’ column to be pulled based on the @id attribute of the current item. If I replace [@for=data(@id)] in the selector with [@for=107] of course I get the value Some notes! in the last record. Is it possible to do this with XPath/XQuery, or am I barking up the wrong tree here? I think the problem is that

The XML is a bit awkward, yes, but I can’t really change it I’m afraid.

I found one solution that works, but it feels awfully heavy for something like this.

select  Item.Person.value('data(../@id)', 'int') as person_id,
        Item.Person.value('data(@id)', 'int') as item_id,
        Notes.Person.value('text()[1]', 'varchar(max)') as notes
from    @xml.nodes('/root/person/item') as Item(Person)
        inner join @xml.nodes('/root/person/notes') as Notes(Person) on
            Notes.Person.value('data(@for)', 'int') = Item.Person.value('data(@id)', 'int')
            and
            Notes.Person.value('data(../@id)', 'int') = Item.Person.value('data(../@id)', 'int')

Update!

I figured it out! I’m new at XQuery but this works, so I’m calling it job done 🙂 I changed the query for the notes to:

Item.Person.value('
    let $id := data(@id)
    return data(../notes[@for=$id])[1]
', 'varchar(max)') as notes

Answer 1

I would suggest that you do a cross apply instead of doing ../ to find a parent node. According to query plan it is a lot faster.

select  P.X.value('data(@id)', 'int') as person_id,
        I.X.value('data(@id)', 'int') as item_id,
        I.X.value('let $id := data(@id)
                   return data(../notes[@for=$id])[1]', 'varchar(max)') as notes
from @xml.nodes('/root/person') as P(X)
  cross apply P.X.nodes('item') as I(X)

You can even remove the ../ in the flwor with one extra cross apply gaining a bit more.

select P.X.value('@id', 'int') as person_id,
       TI.id as item_id,
       P.X.value('(notes[@for = sql:column("TI.id")])[1]', 'varchar(max)') as notes
from @xml.nodes('/root/person') as P(X)
  cross apply P.X.nodes('item') as I(X)
  cross apply (select I.X.value('@id', 'int')) as TI(id) 

Comparing the queries against each other I got 67% on your query 17% on my first and 16% on the second. Note: these figures only give you a hint on what query will actually be faster in reality. Test the against your data to know for sure.