Ann W. Harrison wrote:

>Could we find a fix for that problem by identifying statements that
>might be susceptible - including obscurities like:
>
> insert into t select * from a
> where a.t_val in (select t_val from t)
>
>
>and checking somewhere that the input values aren't dependent on a
>record that's listed in the verb undo log?
>
>

I don't undestand the ODS structure, but isn't possible something like this?

Insert the new records in "unlinked" pages without storing in the
indexes (if they exist). These pages are not visible to selects.
After the records are inserted, the pages is linked to be visible by
sequential scan, and the indexes entries are added.


Adriano