Hi John,

>I thought an index in a cdx was a b-tree with no actual data in it.

In VFP the index contains the value of each index key in a compressed form. For instance, in Customers.CDX at the end you find this block:

00000039B0: 52 4E 53 48 45 41 53 54 │ 43 55 4D 4F 4E 44 52 41  RNSHEASTCUMONDRA
00000039C0: 43 44 4E 53 48 4F 4D 4D │ 49 48 4F 50 53 45 4E 54  CDNSHOMMIHOPSENT
00000039D0: 43 43 41 43 54 55 53 42 │ 45 56 54 54 4D 4E 41 50  CCACTUSBEVTTMNAP
00000039E0: 4F 4C 49 44 4F 4E 50 4C │ 41 55 53 42 45 52 47 53  OLIDONPLAUSBERGS
00000039F0: 52 4F 55 54 54 4F 4E 4E │ 41 54 52 41 4C 46 4B 49  ROUTTONNATRALFKI

which is the ID of every record in alphabetical inverse order: ALFKI is the last entry. NATR is ANATR. The first letter is shared with the previous entry and so forth. The number of shared characters is in the tree list in the same data block along with the record number.

VFP uses the index to quickly search data. If the tree had only pointers to the record, VFP would have to read in the table every time it performs an index search. With data in the index file, most optimization is done exclusively in the index. If you use INDEXSEEK() or have a fully optimized query that counts records, the DBF file isn't touched at all.