>

---

AFAIK you can't do it in one command. OTOH if I understood it correctly
>you're using recno() as PK (very dangerous if also pack is allowed). Then it's
>dangerous to replace with negative recno() because negative recno()s are also
>recno()s used by buffered records. You could inadvertently duplicate them (ie:
>You delete record 6 and change its PK to -6, you have 7 records that aren't
>tableupdated yet, they would have recno()s as -1 ... -7). If your PK always
>get unique values then it wouldn't be necessary to change them before delete.

This matter needs some more explanation :

this is the code that adds records to a certain table :
IF SEEK(compound_keyvalue, "mytable", "myindex")
SELECT mytable
GATHER MEMVAR
ELSE
INSERT INTO TABLE mytable FROM MEVAR
ENDIF

As long as 'myindex' is in good condition, 'mytable' won't have compound_keyvalue's. Though in the past I have had several situations 'mytable' contained duplicate key's.

So that's why I put an candidate index on the compound_keyvalue. The system will prevent adding a duplicate value to the table.

*BUT* when the record is deleted, it's allowed to add a new record with a previous deleted key value, but due to the uniqueness constraint, VFP won't let me unless I force one of the fields of the compound index to a value that will never exist (I my case negative values won't occur) before deleting.
Also it's true that packing a table changes the RECNO() but as in this case they're only used in deleted records they don't appear in the packed file anymore ...

My conclusion to this issue is as suggested by Ricardo :
I don't like using filters on indexes for variuos already mentioned reason's, but I think in this example it is the most appropriate solution ....

Thanks for your input !