Mailing-List: contact cygwin-apps-help AT sourceware DOT cygnus DOT com; run by ezmlm
Sender: cygwin-apps-owner AT sourceware DOT cygnus DOT com
List-Subscribe: <mailto:cygwin-apps-subscribe AT sources DOT redhat DOT com>
List-Archive: <http://sources.redhat.com/ml/cygwin-apps/>
List-Post: <mailto:cygwin-apps AT sources DOT redhat DOT com>
List-Help: <mailto:cygwin-apps-help AT sources DOT redhat DOT com>, <http://sources.redhat.com/lists.html#faqs>
Delivered-To: mailing list cygwin-apps AT sources DOT redhat DOT com
Message-ID: <3BFBE824.6000007@ece.gatech.edu>
Date: Wed, 21 Nov 2001 12:45:08 -0500
From: Charles Wilson <cwilson AT ece DOT gatech DOT edu>
User-Agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:0.9.4) Gecko/20011019 Netscape6/6.2
X-Accept-Language: en-us
MIME-Version: 1.0
To: cygwin-apps AT cygwin DOT com
Subject: Re: ld  ---export--all --exclude-libs,?
References: <20011121084158 DOT 22945 DOT qmail AT web14501 DOT mail DOT yahoo DOT com> <20011121110829 DOT B24327 AT cygbert DOT vinschen DOT de>
Content-Type: text/plain; charset=us-ascii; format=flowed
Content-Transfer-Encoding: 7bit

Corinna Vinschen wrote:

> On Wed, Nov 21, 2001 at 07:41:58PM +1100, Danny Smith wrote:
> 
>>Hello, I'm after some criticism, before submitting this to binutils.
>>
>>IMO, the auto-export feature of ld needs (at least) an option to exclude
>>whole libs from export. The following patch does that. It also adds a few
>>more default lib excludes and scans for backward compat implib symbols
>>(__imp_) 
>>
>>I think this patch doesn't do enough.
>>
>>My preference would be to exclude *all* libs by default, with a command
>>line option to explicity *include* named libs.  Why? It is too easy for
>>licensed library code to sneak into dll's using --export-all. This affects
>>distributability (sic) of the dll and apps dependent on the dll.
>>The user may be unaware of this infection because its automatic.
>>At least make the user think about what she/he is exporting.
>>
>>I will gladly throw away the following to work on a patch to exclude all
>>libs by default if given encouragement.
>>
>>Any comments.
>>
> 
> Yes, a question.  First note that I'm a binutils (l)user, not a
> developer so I apologize for being thick.  Now the question:
> 
> Isn't the functionality you're searching for already in ld given by
> the --no-whole-archive options?
> 
>   gcc -shared -o cygfoo.dll $(OBJ)
>       -Wl,--out-implib,libfoo.dll.a
>       -Wl,--no-whole-archive -lbar -lbaz
> 
> Or am I missing the point?


the problem is, --no-whole-archive applies to EVERY library, even 
convenience libraries.  For instance, a common autotools idiom is to 
create various convenience libraries libsubunit1.a and libsubunit2.a. 
These archives are intended to be included in the eventual .so (.dll) 
and the internal symbols of libsubunit*.a are INTENDED to be exported 
from that .so.

--no-whole-archive makes this idiom impossible; it is not granular 
enough.  Ralf's solution removes the whole purpose of the idiom -- 
what's the use of a "convenience" library if you use --no-whole-archive 
and must then explicitly export every .o contained within the 
"convenience" library?  You're better off not creating the "convenience" 
library at all, and just explicitly adding the various .o files to your 
link command line.

I don't really care whether "export symbols from static libraries" 
defaults to yes or no -- but whatever choice is made, there must be a 
way to override it on a library-by-library basis.  So, Danny's patch in 
its current form is fine.  The inverse form (where .a's are excluded by 
default, and you can explicitly add specific static libraries whose 
symbols ARE exported) is also good.

--Chuck