Anonymous Namespaces

[Jan Hubicka]

> On Sun, Feb 01, 2004 at 02:47:19AM +0100, Gabriel Dos Reis wrote:
> >     cxx_used_only_in_this_tu (tree_decl *decl)
> >     {
> >        return !TREE_PUBLIC (decl)
> >            || (cxx_no_export && declared_at_unnamed_scope (decl));
> 
> I would resist this strongly.  As a language-independent flag, TREE_PUBLIC
> corresponds directly to "does .global need to be emitted in the assembly".  
> 
> If C++ knows that a decl cannot be referenced outside this TU, then it
> should clear TREE_PUBLIC.  Anything less is a QOI mistake.

Hi,
I am having related problem with unit-at-a-time.  With -fwhole program I
need to make functions that are public in C/C++ static for the backend.
I am having prototype patch, but still I am holding because I am unsure
whether clearing up TREE_PUBLIC in cgraph_optimize is TRT.
Given that we want to declare TREE_PUBLIC to be in backend's semantic
and make frontends to use different flag when these don't match (that is
with -fwhole-program almost always),  I think I can update the patch and
propose it for mainline once I am done with merging the tree-SSA cgraph
code back.

Honza
> 
> 
> 
> r~