From owner-freebsd-bugs  Tue Apr 15 00:41:03 1997
Return-Path: <owner-bugs>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id AAA22874
          for bugs-outgoing; Tue, 15 Apr 1997 00:41:03 -0700 (PDT)
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.2.228.19])
          by freefall.freebsd.org (8.8.5/8.8.5) with ESMTP id AAA22865
          for <freebsd-bugs@freebsd.org>; Tue, 15 Apr 1997 00:41:00 -0700 (PDT)
Received: (from bde@localhost) by godzilla.zeta.org.au (8.8.5/8.6.9) id RAA17735; Tue, 15 Apr 1997 17:34:16 +1000
Date: Tue, 15 Apr 1997 17:34:16 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199704150734.RAA17735@godzilla.zeta.org.au>
To: peter@spinner.DIALix.COM, phk@dk.tfs.com
Subject: Re: misc/3291: md2.h, md4.h, and md5.h headers useless for c++ programms
Cc: freebsd-bugs@freebsd.org, tejblum@arc.hq.cti.ru
Sender: owner-bugs@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

>> > > I can agree to the __BEGIN_DECLS, but the #includes do >NOT< belong here.
>> > 
>> > Why? Take a look at <stdlib.h>, for example.
>> 
>> stdlib.h is a ANSI (POSIX?) header, we have to follow the crowd on that.
>
>Headers should be self sufficient.  Converting a header that used to be 

This goes without saying :-).

>self sufficent (eg: on the 2.2 series) to one that isn't (on 3.0+) 
>is a bad move. It'll just make people more reluctant to use libmd and roll 
>their own instead.  IMHO, yes, fine, make it use __BEGIN_DECLS etc, but 
>don't break add "you must now #include <sys/cdefs.h> first even though you 
>didn't before" to it.

It is a bug for an application to #include <sys/cdefs.h>.  No section 2 or
section 3 man pages refer to it (I recently fixed a couple of references).
style.9 doesn't quite say that it is an implementation detail.

>In /usr/include, the only files that use __BEGIN_DECLS without explicitly 
>including cdefs.h first are:
>...
>mpool.h is outright broken, pthread_np.h is an extention of pthreads.h, so 
>that isn't so clear.

pthread_np.h is also missing lots of typedefs.  Clearly it is not meant to
be used by itself.

md4.h now depends on <sys/types.h> for the definition of u32_t.
This was broken when 64-bit support was fixed (md4.h previously used
`unsigned long').  This shouldn't be fixed by including <sys/types.h>,
since <sys/types.h> pollutes the namespace with select and ntoh macros
and POSIX typedefs.

md5.h includes <sys/md5.h> which has the same problem as md4.h and uglier
ifdefs.

Bruce