I would absolutely agree that every table should have a primary key.

I would strongly disagree with your recommendation that a "damn good starting point is to just add a identity column called rowid and make it the primary key and the clustered key". I would consider this a bad practice. Rather, we should examine what attributes we are storing to find the "real primary key" and alway use that. I would only suggest identity columns as a last resort, in cases where there are no unique attributes that are a candidate key - the typical example being a Customer table, where we would add an artificial CustomerID. I wouldn't go as far as saying never use identity columns, but using them as a matter of course is a bad idea. I doubt you will get any agreement on identity columns though - it seems to be one of those issues that always causes arguments.

Cheers,

Andy Mackie