From: Tom Lane <tgl@sss.pgh.pa.us>
To: Thomas Munro <thomas.munro@gmail.com>
cc: Tomas Vondra <tomas@vondra.me>, Peter Eisentraut <peter@eisentraut.org>,
        Heikki Linnakangas <hlinnaka@iki.fi>, Japin Li <japinli@hotmail.com>,
        Andres Freund <andres@anarazel.de>,
        PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: Cannot find a working 64-bit integer type on Illumos
In-reply-to: <2385947.1733341279@sss.pgh.pa.us>
References: <ME3P282MB3166F9D1F71F787929C0C7E7B6312@ME3P282MB3166.AUSP282.PROD.OUTLOOK.COM> <20240322165305.6zrtxcmzdrywvmsu@awork3.anarazel.de> <ME3P282MB3166D876E9CB83009727A5ECB6312@ME3P282MB3166.AUSP282.PROD.OUTLOOK.COM> <944094.1711128356@sss.pgh.pa.us> <CA+hUKG+CR05ACFujUbeAd-yM+nNJBSpnyB=q3vbriCJuoSu9Ag@mail.gmail.com> <1229869.1711160593@sss.pgh.pa.us> <CA+hUKG+ZwX-reVMHak3yo-63rrh9AJwEVGW4rCbsoLeSKO54Pg@mail.gmail.com> <7cf1dcb7-306f-4418-bf25-bdc75130db10@eisentraut.org> <CA+hUKGKXEdVSfR6UHXfR3tSQ+TfBNk-ssYmO2F0wCdw7qcx86Q@mail.gmail.com> <d8d68452-7195-41cf-a31b-e660f5ac6e75@iki.fi> <CA+hUKG+rYp-=fTP5kjRmnfPq7YgK-e=J+gJxBvKarkPVzLwQmQ@mail.gmail.com> <0e2cc7af-f0c0-418c-b640-e5eb03eaa6e6@eisentraut.org> <CA+hUKGL7YprY+mkWg-9rVkrzkeC62ppdzyb7KcPp9tcff_5F3Q@mail.gmail.com> <CA+hUKGK7WQX3nTuYbiiGZFcV70CqbmBx4VUOnDy6ZGm1v=qGqQ@mail.gmail.com> <50a4c05e-c43e-4e7e-95d2-4b16493073ed@eisentraut.org> <CA+hUKGJ=6wVNNdv9XAUwNv+hhPCW3FzFEkrr6i5axx090vEmcQ@mail.gmail.com> <CA+!
 ! hUKGKRM1vk2Bd=B7O9SKeiz6c=KUB08N0M9fHVq057u9uhJw@mail.gmail.com> <2276817.1733293172@sss.pgh.pa.us> <2385947.1733341279@sss.pgh.pa.us>
Comments: In-reply-to Tom Lane <tgl@sss.pgh.pa.us>
	message dated "Wed, 04 Dec 2024 14:41:19 -0500"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <2397642.1733347237.1@sss.pgh.pa.us>
Date: Wed, 04 Dec 2024 16:20:37 -0500
Message-ID: <2397643.1733347237@sss.pgh.pa.us>
Archived-At: <https://www.postgresql.org/message-id/2397643.1733347237%40sss.pgh.pa.us>
Precedence: bulk

I wrote:
> So apparently, "off_t" was the same as "loff_t" before 962da900a,
> but it no longer is the same on 32-bit machines.

OK, I see what is happening.  On platforms that need it, we define
_FILE_OFFSET_BITS as 64 in pg_config.h to ensure that off_t is
big enough.  However, 962da900a did this:

--- a/src/include/postgres_ext.h
+++ b/src/include/postgres_ext.h
@@ -23,7 +23,7 @@
 #ifndef POSTGRES_EXT_H
 #define POSTGRES_EXT_H
 
-#include "pg_config_ext.h"
+#include <stdint.h>

Since c.h reads postgres_ext.h first, <stdint.h> is now pulled in
before pg_config.h, and that in turn pulls in <features.h>, which
locks down the decision that off_t will be 32 bits, as well as some
other decisions we don't want.  We can NOT include any system headers
before pg_config.h; I'm surprised we're not seeing related failures on
the Solaris-en, where _LARGEFILE_SOURCE is similarly critical.

Another rather serious problem here is that we no longer provide
macro PG_INT64_TYPE, which seems rather likely to break applications
that were relying on it.  That is part of our external API, we
can't just remove it on a whim.

I think the least painful solution would be to revert the parts
of 962da900a that got rid of pg_config_ext.h and PG_INT64_TYPE.
Since PG_INT64_TYPE is a macro not a typedef, it might be okay
to #define it as int64_t even before we've read that header,
so as not to give up the principle of relying on stdint.h for the
underlying definition.

Now that I see this, I'm fairly astonished that there aren't
more problems than we've noticed.  I wonder whether it'd be
a good idea to put in a static assert somewhere about the
width of off_t, so that the next screwup of this sort will
be easier to diagnose.

			regards, tom lane