PW you assume dates are unique within a group. Not a very safe assumption, esp. if there's no time portion, or frequent transactions.

Do you want the earliest of the records, latest, don't care? If you want first or last, how will you break ties? If you don't care which record you keep it...

Agreed. Applies to all other identifiers as well as table and column names.

I don't think it will work. Look at the syntax in BOL:

< generation_term > ::=

     FORMSOF ( { INFLECTIONAL | THESAURUS } , < simple_term > [ ,...n ] )...

What datatype is the field? Text or a character datatype? Charindex doesn't work on text.

Not in TSQL you can't.

[edit: oh i see it's a UDF. There are better ways of splitting strings though. But this request was not about doing that anyway.]

5 most recent sales to whom, of what?

The (scalar) UDF in a SQL statement acts like a correlated subquery. Just as the subquery is joined on Customer and Item, the...

declare @str varchar(8000)

set @str = 'a:bb:cc:'

select

substring(@str,len(@str)-charindex(':',reverse(@str))+2,8000)

that way you don't need to reverse three times to do it...

Here's one way of doing it in a single SQL statement, but it's fairly messy and essentially the same as the UDF method. Might give the optimiser some scope for improvement over...

Yes with small recordsets the impact of inefficient code can be negligable.