I have just been noticing that pg_dump outputs index definitions sorted by index name (within a given schema). Would it not be better to sort by <indexed_relation_name, index_name> so that we get some benefit from caching the table when building multiple indexes on the same table?
