>>The behaviour is, that VFP simply deletes the old record while safing the class and appends new records to the VCX. So VFP makes shure to get the last - valid - record. For example it does not care about SET DELETED. In older versions of VFP the deleted records in the VCX where never packed.
>
>Except when you have set deleted off and
>
>insert into (targetvcx) select * from (oldvcx) where {condition here}
>
>Then it may insert deleted records as well and you end up with a duplicate. In some cases, when VFP is sure you're alone with a vcx, it may pack it on save (sure does so with a scx), but most of the time it considers the case that two people may be editing different classes from the same vcx at a time, so it just deletes the old and inserts the new. The above scenario will insert both version, and that's probably similar to what happened to Alejandro. BTW, compile classlib also packs it as a table, so those deleted records vanish. I regularly check the size before submitting to source control, and have often delayed a checkin just to compile it.
>
>Funny how this feature managed to sneak its way through the years.

The autopacking of vcx (no idea on scx) was introduced very late. I know it was not there for VFP 5.0 and 6.0. The OP doesn't provide a version.
I have never seen anything else then append class on save, but possibly this depends on MULTILOCKS or something similar.

I even had a program to crawl through all kinds of source tables and pack them, those days.