Re: REVIEW: EXPLAIN and nfiltered

Hitoshi Harada <umi(dot)tanuki(at)gmail(dot)com> writes:
> 2011/1/21 Florian Pflug <fgp(at)phlo(dot)org>:
>> "Rows Skipped: nnn", maybe?

> +1. Very straightforward to me.

I didn't really care for that one, because I think it *won't* be
straightforward when there's more than one filter condition at a node.
Imagine

Bitmap Heap Scan ...
Recheck Cond: blah blah
Rows Skipped: 42
Filter Cond: blah blah blah
Rows Skipped: 77

To me, "rows skipped" sounds like a statement about the overall behavior
of the plan node, and thus the above looks contradictory. Another point
is that even if you're okay with the above for textual output, we do not
have a choice about choosing distinct field names for the two counts for
XML/JSON output.

Reflecting on that, I'm inclined to suggest

Bitmap Heap Scan ...
Recheck Cond: blah blah
Rows Removed by Recheck: 42
Filter Cond: blah blah blah
Rows Removed by Filter: 77

or even more verbosely

Bitmap Heap Scan ...
Recheck Cond: blah blah
Rows Removed by Recheck Cond: 42
Filter Cond: blah blah blah
Rows Removed by Filter Cond: 77

ie repeat the label of the filtering condition exactly. This is looking
pretty long, but from the viewpoint of vertical or horizontal space
occupied by the printout, I doubt it matters.

regards, tom lane