From: Andrew Gierth <andrew@tao11.riddles.org.uk>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: Michael Paquier <michael@paquier.xyz>,
  Amit Kapila <akapila@postgresql.org>,  pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Make the order of the header file includes consistent in
 non-bac
In-Reply-To: 
 <CAA4eK1+Rvft5SgGnBfGdNCVMspDMiXEo=i7Ue2e9Pm0bbOC3jA@mail.gmail.com>
	(Amit Kapila's message of "Fri, 25 Oct 2019 09:32:29 +0530")
Message-ID: <87h83xmg4m.fsf@news-spur.riddles.org.uk>
References: <E1iNpHW-000855-1u@gemulon.postgresql.org>
	<20191025031702.GD8671@paquier.xyz>
	<CAA4eK1JfDfKvjSYOoQFHu=sSzaqdDTR+JZkZqJkvVnueFJPG5g@mail.gmail.com>
	<CAA4eK1+Rvft5SgGnBfGdNCVMspDMiXEo=i7Ue2e9Pm0bbOC3jA@mail.gmail.com>
User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.3 (berkeley-unix)
Date: Fri, 25 Oct 2019 09:00:03 +0100
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

>>>>> "Amit" == Amit Kapila <amit.kapila16@gmail.com> writes:

 Amit> BTW, prairiedog is also show similar failure and both seems to
 Amit> have similar OS except for versions.

Be aware that prairiedog and locust (unlike the rest of the buildfarm)
have compilers whose "bool" type is not 1 byte long; PG can't use that,
so we don't include <stdbool.h> on those platforms.

Looking at c.h, the problem seems to be that when we define our own bool
in the absence of a usable stdbool.h, we do so as a typedef and not a
macro. So pgtypeslib_extern.h ends up redefining it in that case, and
(this bit may be my fault: see d26a810eb) uses a different type to c.h
(char vs. unsigned char).

(stdbool.h is required by spec to do #define bool _Bool rather than a
typedef, hence the difference in behavior)

I'm no expert on the ECPG internals, but this doesn't look like an
exposed interface, so maybe it just shouldn't be trying to #define bool
at all and just rely on c.h (from postgres_fe.h) to declare the type?

-- 
Andrew (irc:RhodiumToad)