Hi,

> > Wouldn't it speed up increadibly when we don't fetch the linked record,

but

> > only do an index evaluate ?
>
> Unfortunately, engine must read the record because index node doesn't
> contain necessary information about row validity for actual transaction.
> (the transaction number is stored only in rows). The second reason for
> row retrieval is (if I understood it well) the second key comparison
> based on actual values (index keys are mangled for bitwise comparison).

So the record-pointer available by the index isn't transaction dependent ?

> So, Firebird can't optimize queries with index lookups by avoiding row
> retrieval.

That would be pitty :(

Regards,
Arno