Re: c++/2166

[Gabriel Dos Reis <Gabriel dot Dos-Reis at cmla dot ens-cachan dot fr>]

The following reply was made to PR c++/2166; it has been noted by GNATS.

From: Gabriel Dos Reis <Gabriel.Dos-Reis@cmla.ens-cachan.fr>
To: Mark Mitchell <mark@codesourcery.com>
Cc: Gabriel.Dos-Reis@cmla.ens-cachan.fr, nobody@gcc.gnu.org,
        gcc-gnats@gcc.gnu.org, carlo@alinoe.com
Subject: Re: c++/2166
Date: 18 Apr 2001 09:40:39 +0200

 Mark Mitchell <mark@codesourcery.com> writes:
 
 | >>>>> "Gabriel" == Gabriel Dos Reis <Gabriel.Dos-Reis@cmla.ens-cachan.fr> writes:
 | 
 |     Gabriel> The Standard requires that a member of an anonymous
 |     Gabriel> namespace be given a linker-visible unique name in a
 |     Gabriel> given translation unit so that two different translation
 |     Gabriel> units don't mess up.
 | 
 | So, do you agree with me that things are OK, or with Carlo that there
 | is a problem?
 
 Well, what I meant is that g++ has to ensure that the construct
 
 	namespace {
 		int test;
 	};
 
 can appear in different translation units (within the same program)
 and each translation unit have its own idea of `test' -- the different
 `test's should not cause any conflict at the linker level.  
 
 Carlo think that means the compiler should use the file-definition name
 as part of the mangled name of `test'.
 That is a way to achieve uniqueness. Another can be the time of
 tranlation.  Actually anything which can lead to uniqueness is good.
 
 I ran several tests and I think the current scheme is not robust
 enought -- it seems to use very few characters ( < 7) to fake
 uniqueness.
 
 I didn't run into *actual* problems (I ran only 5 tests) so I cannot
 definitely say that there IS a problem.  However,  I think there is a
 potential problem. 
 
 This issue deserve further investigation -- I would do that as soon as
 I have a chance.
 
 -- Gaby