Date: Fri, 4 Sep 2020 16:41:47 -0400
From: Alvaro Herrera <alvherre@2ndquadrant.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-hackers@lists.postgresql.org,
	Peter Eisentraut <peter.eisentraut@2ndquadrant.com>,
	Petr Jelinek <petr@2ndquadrant.com>
Subject: Re: Questionable ping logic in LogicalRepApplyLoop
Message-ID: <20200904204147.GA18660@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <959627.1599248476@sss.pgh.pa.us>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On 2020-Sep-04, Tom Lane wrote:

> While playing around with clang's scan-build I noticed this warning:
> 
> worker.c:2281:7: warning: Value stored to 'ping_sent' is never read
>                                                 ping_sent = true;
>                                                 ^           ~~~~
> 
> At first I thought it was a harmless unnecessary update, but looking
> closer I wonder whether it isn't telling us there is a logic bug here.
> Specifically, I wonder why the "ping_sent" variable is local to the
> loop starting at line 2084, rather than having the same lifespan as
> "last_recv_timestamp".  Do we really want to forget that we sent a
> ping anytime we have to wait for more data?

Ah ... maybe this bug is the reason why the bug fixed by 470687b4a5bb
did not affect logical replication.

> In short, we could remove the ping_sent variable entirely without
> changing this code's behavior.  I'm not 100% sure what semantics
> it's trying to achieve, but I don't think it's achieving them.

I imagine that moving the variable one block-scope outwards (together
with last_recv_timestamp) is what was intended.

... oh look!  commit 3f60f690fac1 moved last_recv_timestamp without
realizing that ping_sent had to get the same treatment.

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services