Alvaro Herrera <alvherre@dcc.uchile.cl> writes:
> Actually, I'm planning to do the freelist thing, then the btree
> compaction and then replace the current REINDEX code with the compaction
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> code, probably including some means to do REINDEX ALL.
Uh ... no. The primary purpose of REINDEX is to recover from corrupted
indexes, so it has to be based on a rebuild strategy not a compaction
strategy.
If you want to add a REINDEX ALL for completeness, go ahead, but I think
the need for it will be vanishingly small once vacuum compacts btrees
properly.
regards, tom lane