Re: [HACKERS] Improving DISTINCT with LooseScan node

time to get back to yet:

https://www.postgresql.org/message-id/flat/CADLWmXWALK8NPZqdnRQiPnrzAnic7NxYKynrkzO_vxYr8enWww%40mail.gmail.com

That was based on the idea that a DISTINCT scan using a btree index to
skip ahead is always going to be using the leading N columns and
always going to be covered by the index, so I might as well teach
Index Only Scan how to do it directly rather than making a new node to
sit on top.  As you can see in that thread I did start thinking about
using a new node to sit on top and behave a bit like a nested loop for
the more advanced feature of an Index Skip Scan (trying every value of
(a) where you had an index on (a, b, c) but had a WHERE clause qual on
(b, c)), but the only feedback I had so far was from Robert Haas who
thought that too should probably be pushed into the index scan.

FWIW I'd vote for 'skip' rather than 'loose' as a term to use in this
family of related features (DISTINCT being the easiest to build).  It
seems more descriptive than the MySQL term.  (DB2 added this a little
while ago and went with 'index jump scan', suggesting that we should
consider 'hop'... (weak humour, 'a hop, skip and a jump' being an
English idiom meaning a short distance)).