How to efficiently support "Like"-masked range queries against a CLR UserDefinedType

Question

Hi all,
Is there any way to efficiently support "Like"-masked range queries against a CLR UserDefinedType?   I've never written a UDT before and am just beginning with one that has a Varbinary type as the underlying physical type for
persistence but has a string representation to the consumer.   Looking at the articles and samples I've found, the only required method/interface for a UDT that seems to be made for queries is the requirement to provide an implementation
of IComparable.   
But from what I've read so far, doing a "Like" masked/range search - or any search for other than an exact "key" expression match would amount to doing a table scan while the database engine passes in each candidate to the UDT's IComparable
implementation.
I need to support masked searches; including some that might wild-card the first few characters instead of always providing always providing non-wild-carded characters staring in character position one of the mask.   My UDT would
have its internal structures that could vastly speed up the searches if I could get the whole Like() mask at once vs. having my IComparable methods being called once per every existing row in the underling table.    (We may have one such table
with a column of this UDT containing over 800M rows.   You can imagine the response times if every search other than a single exact key match effectively meant a table scan).
I had heard that you could implement a "Like()" method in a UDT that might allow this type of processing.   But I can't find any discussions or documentation on this issue; so I'm not sure if that's really an option for solving this issue.
Thanks!
-Bob

Answer 1

Not sure that I understand exactly, but obviously you can implement a type method that returns 0 or 1. However, searches on this method would requiring scanning the table, so even if your internal matching is fast, the query will still be slow.
It sounds like you are looking for some custom indexing, and there is nothing built-in that you can hook into, but you would have to do all the work yourself - if it is possible at all.
Erland Sommarskog, SQL Server MVP, [email protected]