Andres Freund <and...@anarazel.de> writes: > On 2018-07-11 12:56:49 +0530, Amit Kapila wrote: >> Yeah, it is good practice to drop the objects at the end. It is >> strange that original commit adfb81d9e1 has this at the end of the >> test, but a later commit 367b99bbb1 by Tom has removed the Drop >> statement. AFAICS, this is just a silly mistake, but I might be >> missing something. Tom, do you remember any reason for doing so? If >> not, then I think we can revert back that change (aka commit Kuntal's >> patch).
> We actually sometimes intentionally want to persist objects past the end > of the test. Allows to test pg_dump / pg_upgrade. Don't know whether > that's the case here, but it's worthwhile to note. I don't think our pg_dump testbed makes any use of contrib regression tests, so that's not the reason here. I believe I took out the DROP because it made it impossible to do additional manual tests after the end of an installcheck run without laboriously re-creating the test table. In other words, I disagree with Amit's opinion that it's good practice to drop everything at the end of a test script. There are often good reasons to leave the objects available for later use. regards, tom lane