Date: Thu, 17 Oct 2019 10:30:06 -0500
From: Justin Pryzby <pryzby@telsasoft.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Stephen Frost <sfrost@snowman.net>, imai.yoshikazu@fujitsu.com,
	Euler Taveira <euler@timbira.com.br>,
	Andrew Gierth <andrew@tao11.riddles.org.uk>,
	PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: v12 and pg_restore -f-
Message-ID: <20191017153006.GA3599@telsasoft.com>
References: <20191006190839.GE18030@telsasoft.com>
 <28770.1570394593@sss.pgh.pa.us>
 <875zl11tla.fsf@news-spur.riddles.org.uk>
 <30069.1570396524@sss.pgh.pa.us>
 <20191008180840.GQ6962@tamriel.snowman.net>
 <CAHE3wgi=VU0G9=S60=gQH+yyk8-Ke99dkt8Dp4OpZXkSu=67fQ@mail.gmail.com>
 <20191009124505.GT6962@tamriel.snowman.net>
 <OSBPR01MB4616298336E32F8EAEDA0FF394920@OSBPR01MB4616.jpnprd01.prod.outlook.com>
 <20191016172148.GH6962@tamriel.snowman.net>
 <19891.1571307850@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <19891.1571307850@sss.pgh.pa.us>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk

On Thu, Oct 17, 2019 at 12:24:10PM +0200, Tom Lane wrote:
> Alternatively, we could revoke the requirement to use "-f -" in 12,
> and wait a couple releases before enforcing it.  The fundamental
> problem here is that we tried to go from "-f - doesn't work" to
> "you must use -f -" with no grace period where "-f - is optional".
> In hindsight that was a bad idea.

I'm going to make an argument in favour of keeping the enforcement of -f- in
v12.

If there's no enforcement, I don't know if many people would naturally start to
use -f-, which means that tools which need to work across a wide range of
(minor) versions may never confront this until it's enforced in v14/v15, at
which point we probably end up revisiting the whole thing again.

Failing pg_restore forces people to confront the new/different behavior.  If we
defer failing for 2 years, it probably just means it'll be an issue again 2
years from now.

However, it's still an issue if (old) back branches (like 11.5) don't support
-f-, and I think the differing behavior should be called out in the v12 release
notes, as succinctly as possible.

Also, I'm taking the opportunity to correct myself, before someone else does:

On Wed, Oct 16, 2019 at 02:28:40PM -0500, Justin Pryzby wrote:
> And vendors (something like pgadmin) will end up "having to" write to a file
> to be portable, or else check the full version, not just the major version.

I take back that part .. before v12, they'd get stdout by not specifying -f,
and since 12.0 they'd get stdout with -f-.  No need to check the minor version,
since the "need to" specify -f- wouldn't be backpatched, of course.

Justin