Dragan

I agree completely. There was a case where we had a tree structure implemented that would not practically go more than 8 levels deep. It was in Sybase and we didn't have quite the SQL set of commands that were available in SQLServer. To get the absolute best performance out we hard coded a SP to do 10 self joins, with the code that did the tree maintenence to burp out an error message should they ever attempt to make the tree more than 10 levels deep. It was something like 50 times faster to hardwire the self-joins than it was to do any sort of iterative loop through the tree.

>I guess the ratio between time programmer spends solving a speed issue against the accumulated time the users may be wasting with the slower version is usually more than 2:1... but for the sake of the few cases when it gets to be 1:10 or better, I'd rather stay in shape (the shape of a bottleneck waiting to be wider)
>
>Worse... Cobol or VB6 mentality :). I've seen recursion down a self-referencing table (organizational structure of an institution) using a new SQL statement into a cursor at each step, including the end nodes - if it returns zero records, it's an end node. A single rip field would make that sequential single pass (which I did eventually, at factor of 20).
>
>Even inside Cobol... there were ways and ways. I remember shortening a logical condition from eight lines to two (and I left it in two lines for readability's sake).