One typical use of the REPEAT block that is still valuable is when you use it with a construct called a PRESELECT. To understand this usage, you need to think a little about what happens when an iterating block like a FOR EACH block begins. The AVM evaluates the record retrieval the FOR EACH statement defines. It then goes out to the database and retrieves the first record of the set of related records that satisfies the statement and makes the record available to the block. When the block iterates, the AVM goes and gets the next record. As long as it's possible to identify what the first and the next records are by using one or more indexes, The AVM doesn't bother reading all the records in advance. It just goes out and gets them when the block needs them.

If you specify a BY clause that requires a search of the database that an index can't satisfy, the AVM has no choice but to retrieve all the records in advance and build up a list in sort order. But this is not ordinarily the case. It's much more efficient simply to get the records when you need them.

Sometimes, though, you need to force the AVM to get all the records that satisfy the FOR EACH statement in advance, even when the sort order itself doesn't require it. For example, if it's possible that you will modify an indexed field in some of the records in such a way that they would appear again later in the retrieval process, you need to make sure that the set of records you're working with is predetermined. The PRESELECT keyword tells the AVM to build up a list of pointers to all the records that satisfy the selection criteria before it starts iterating through the block. This assures you that each record is accessed only once.

In summary, a REPEAT block does everything a FOR block does, but it does not automatically advance to the next record as it iterates. You should use a REPEAT block in cases where you want to control the record navigation yourself, typically using the FIND statement described in the next section. It provides you with record, frame, and transaction scoping.

Because it provides all these services, a REPEAT block is relatively expensive compared to a DO block. Use the simpler DO block instead of a REPEAT block, unless you need the record-oriented services provided by the REPEAT block.