OK, because it was Michael that asked originally, XML/XQuery wishes for SQL Server 200x. In no particular order.
Full compliance with "standard" W3C XQuery. If the standards committee finishes before SQL Server 200x ships. Else it will become a "standard" no one follows. Hmm…like SQL. Reason for this is twofold.
1. Folks who get a "generic" XQuery book at the store are now totally confused. Many give up after the third or fourth "not supported" message.
2. RCG (recent college grads) who may learn "generic" XQuery in college. There's a reason why C# looks somewhat like another popular programming language. Because it makes it easier to switch environments. And code.
XQuery editor. That does schema or schemaless. Strong typing. Files or the SQL table row of your choice. They had a good start in the beta, but had to finish VS XML tools.
Client side XQuery. How can I write code that might run on server or client, when it can't run on the client? XLINQ is nice, but it's not a substitute. Should be compatible with server, modulo input and ouput.
Output in encodings other than UTF-16. A CONVERT with extra parm, like DATETIME has.
Full support for XSD schema productions. Yes, I know that its techinically outrageous. But its another source of confusion/frustration. Get an industry standard schema (even Office 2003 schemas). Can't store it without changing "lax" to "skip". But then its not the same schema if I change it.
Standard grammer for composition and decomposition. Now, I'm a big fan of SELECT…FOR XML. It's easier and more powerful than the alternative. But let's talk conversions again. Just layer on XmlElement(), XmlAttribute, etc. Kind of like CAST vs CONVERT in TSQL, they implement both.
A XML.validate method.
A more powerful modify method. But again, this could wait (but not long) for the W3C. Most people don't know that modify IS based on a standard proposal. But XDR and XML Patterns were proposals too. So maybe implement "update" (in addition to insert, replace value of, and delete) for next time. Bet SQL Server is finished before W3C. Multiple node inserts/updates/replace value ofs.
XQuery collection() semantics over rows in a table. That could be considered a collection.
XQuery 1.0 and XPath 2.0 data model (or ANSI SQL 200x data model) support. They're relatively close now.
FOR XML works with UDTs.
Ability to specify DOCUMENT on well-formed data. Like it works on schema-valid. This is a nit, I can do my own constraint. Arguably.
Finally, Node-level locking or Node-level versioning. Yes, I know. Technically outrageous. But it IS a WISH list after all.