>I thought that even without an order specified, the engine is supposed
>to use the natural one. As Oracle has its ROWID, I think FB has its
>RDB$DBKEY (or something like that) that is generated after the
>insertions and stablished some order. So, if in regular selects I use to
>obtain the same result set (in natural order) why not in subselects?
>Aren't the rows returned in a top-down sequence, according to the
>information in the system tables?

You are right Fernando, FB has rdb$db_key and it is constant for the
duration of your transaction
(http://www.cvalde.com/document/mysteriousDbKey.htm explains it very well).
And you have convinced me that it is indeed a bug - although I am not
certain whether the fix should be to cater for FIRST/SKIP within the
subselect or raise an error at compile time saying that FIRST/SKIP in a
subselect is not allowed (you have not convinced me that your original query
is a good way to do what you tried to do).

Set