Hi Fabio,
>normally VFP create this temp index.

that was my take for "simple" cases as well, and sys(3054) giving hints usually helps cut down on dev time if the index can be created in the normal run of the program, as no temp indexing is needed.

>I remember to have seen cases where VFP uses a Cartesian product with an outer join;
>I cannot say because this is not true for the Naomi's select, but it can to be one of these.

I distinctly rembering thinking it weird that a join was not optimized - which made it literally take hours, when my assumption was that it should be optimized according to rushmore rules. Since table sizes were near limit for vfp I thought it might have to with "overflowing" memory needs even if the machine was well equipped.

Splitting the Select into 2 similar selects made the perf problem go away - I was not given access to try out alternative approaches to clearly identify the problem - data was in a secure area and no budget was alotted for me scratching my curiosity.

>I use integer pk only on local table or cursor.
Same as me if I can control it...

regards

thomas