Hugo Kornelis (10/19/2012)
L' Eomot Inversé (10/19/2012)
Actually, there's a misprint in the question (my stupid error) - it should say @ROWGUIDCOL, not ROWGUID. You can have as many rowguid columns as you like in any kind of table (just as you can have as many int columns as you like in any kind of table - after all, rowguid is a type like any other type), but only one of them can have the ROWGUIDCOL property (just as only one column can have the IDENTITY property). I'll ask Steve to correct that wording (and 2011) if possible.I disagree with this. Yes, the question was technically not completely accurate. But your explanation here is worse. The term "ROWGUID" does not exist in SQL Server. It is not a type like any other type. uniqueidentifier is a type like any other type, and the commonly accepted short form for that data type name is guid. Not ROWGUID.
To me, it was obvious that "ROWGUID columns" was intended as "columns with the ROWGUIDCOL attribute. Especially since "ROWGUIDCOL columns" would have sounded redundant.
Well, since ROWGUID (as opposed to GUID or ROWGUIDCOL) is not a legitimate term, it's possible for people to mistake it as meaning GUID (ie uniqueidentifier) type rather than the column property ROWGUIDCOL, and it seems clear that this is what Arthur.Teter did; so I wanted to eliminate the ambiguity, to make sure that more people don't fall into that trap. I agree about the sound of redundancy, though.
I think I picked up using ROWGUID instead of ROWGUIDCOL from the appalling habit displayed in certain third party databases I had which I had the misfortune to look into where the vendors had the awful habit of using uniqueidentifier surrogate keys and calling every surrogate just rowguid in the table where it was primary key (instead of something useful like tablename_surrogate_key); I think their developers must have been used to working in an old dialect of Fortran where the maximum allowed identifier length was 8 characters and believed the same length restriction applied in SQL.
Tom