From: Tom Lane <tgl@sss.pgh.pa.us>
To: Peter Eisentraut <peter.eisentraut@enterprisedb.com>
cc: pgsql-committers@lists.postgresql.org
Subject: Re: pgsql: Remove check for accept() argument types
In-reply-to: <1683239.1636726865@sss.pgh.pa.us>
References: <E1mkSdp-0008FH-7Y@gemulon.postgresql.org>
 <1440792.1636558888@sss.pgh.pa.us>
 <3863895c-d40b-4154-de8f-4961386b2515@enterprisedb.com>
 <1683239.1636726865@sss.pgh.pa.us>
Comments: In-reply-to Tom Lane <tgl@sss.pgh.pa.us>
	message dated "Fri, 12 Nov 2021 09:21:05 -0500"
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <1731091.1636739453.1@sss.pgh.pa.us>
Date: Fri, 12 Nov 2021 12:50:53 -0500
Message-ID: <1731092.1636739453@sss.pgh.pa.us>
Archived-At: 
 <https://www.postgresql.org/message-id/1731092.1636739453%40sss.pgh.pa.us>
Precedence: bulk

I wrote:
> Peter Eisentraut <peter.eisentraut@enterprisedb.com> writes:
>> What does the man page say the correct type 
>> would be?  size_t?

> The machine's not booted up right now :-(.  But I'm pretty sure we
> shouldn't consider using size_t here, as it's not real clear that that
> couldn't be 64 bits on any affected platforms.  Your previous research
> said that the desired type is 32 bits on all such platforms, so I think
> that "int" is correct; we need only debate signedness.

Hmm ... now that I'm in my office, I checked this and HPUX 10.20's
accept(2) man page saith

SYNOPSIS
      #include <sys/socket.h>

    AF_CCITT only
      #include <x25/x25addrstr.h>

      int accept(int s, void *addr, int *addrlen);

    _XOPEN_SOURCE_EXTENDED only
      int accept(int s, struct sockaddr *addr, size_t *addrlen);

gaur is using -D_XOPEN_SOURCE_EXTENDED and hence building against the
latter definition.  (I found out quite a long time ago that without
_XOPEN_SOURCE_EXTENDED, this platform has a LOT of discrepancies from
SUS v2.)  Further down, there's

FUTURE DIRECTION
      The default behavior in this release is still the classic HP-UX BSD
      Sockets, however it will be changed to X/Open Sockets in some future
      release.  At that time, any HP-UX BSD Sockets behavior which is
      incompatible with X/Open Sockets may be obsoleted.  HP customers are
      advised to migrate their applications to conform to X/Open
      specification (see xopen_networking(7)).

So what it looks like to me is

(1) Original BSD Sockets defined the argument as "int *addrlen".
(I confirmed this by looking in an ancient copy of Stevens'
Unix Network Programming.)

(2) X/Open thought it'd be better to use size_t.

(3) POSIX decided to resolve the conflict by inventing socklen_t.
However, SUS v2 says specifically that socklen_t "is an unsigned opaque
integral type of length of at least 32 bits".  The "unsigned" part
was removed in later POSIX editions, which surprises me --- they don't
usually change the standard in the direction of less definiteness.

On the whole, I still think "unsigned int" is our best compromise.
But maybe we should use size_t and cite X/Open as authority.

			regards, tom lane