I wrote:
> That patch looks reasonable to me, will push.
On closer inspection, the patch did contain a bug: it tested for
compression being active with "compression > 0", which is the wrong
thing --- everyplace else in pg_dump tests with "compression != 0".
This is important because Z_DEFAULT_COMPRESSION is -1. The net
effect is that the patch would have appeared to fix the bug when
writing "-Z1".."-Z9", but not when writing just "-Z". Possibly
this explains why Kuntal couldn't confirm success at one point
upthread.
Pushed with that correction and some wordsmithing on the comment.
regards, tom lane