Re: ARM PATCH: Implement COMDAT groups

[Mark Mitchell <mark at codesourcery dot com>]

Richard Earnshaw wrote:

> On Fri, 2004-09-10 at 20:26, Mark Mitchell wrote:
>  
>
> > This patch implements support for use of COMDAT groups as required by
> > the C++ ABI.  HJ's previous patch did not implement the requirement
> > that all virtual tables be placed in a single COMDAT group.
> >
> > Tested on arm-none-eabi, applied on the csl-arm-branch, and queued for
> > GCC 4.1.
> >    
>
> Doesn't this count as a but (ie failure to conform to the EABI) and
> hence doesn't that make this acceptable for stage 3 (bug fixing).  We
> aren't yet at the stage where we are only fixing regressions.
>  
I'm not sure.

For IA32 GNU/Linux, say, this seems more like a feature addition to me.  
As far as I know, the only misbehavior that a user can observe due to 
the failure to use COMDAT groups is incorrect interaction with non-GNU 
compilers, and I believe that the major players have already resigned 
themselves to working around that issue by using weak symbols and 
.gnu.linkonce sections.  Are there other problems with the current 
non-COMDAT situation?

I'm also particularly leery of committing of my own gray-area patches; 
that would severely undermine my ability to glare at other people who 
are in the gray area.

--
Mark Mitchell
CodeSourcery, LLC
(916) 791-8304
mark@codesourcery.com