Re: [patch] empty base class optimization completed in stl_tree.h

[Howard Hinnant <hhinnant at apple dot com>]

On Jan 10, 2007, at 5:04 AM, Maxim Yegorushkin wrote:

> Hi,
>
> It looks like an intended empty base class optimization has not been
> finished for struct _Rb_tree_impl in stl_tree.h. If it is so indeed,
> this patch completes it.
>
> main trunk:
>  sizeof(std::set<int>) == 72
>  sizeof(std::set<int, bool(*)(int, int)>) == 72
>
> with patch applied:
>  sizeof(std::set<int>) == 64
>  sizeof(std::set<int, bool(*)(int, int)>) == 72
>
> Compile tested only.<stl_tree.patch>

I have a question about ABI stability.

Is the reason this is not considered ABI-breaking because std::set is  
never instantiated within libstdc++?  I.e. we can change std::set all  
we want and it will never impact the compiled lib.

However consider client A who creates a library which uses std::set in  
its interface, and compiles prior to the change above.  Client B  
compiles after the change above, and links to Client A's library, and  
say passes a std::set<int> back and forth.  Since the two pieces of  
code have a different layout for std::set<int>, I would guess that  
there might be some problems.

I'm not saying that the above change should necessarily be rejected.   
I merely wanting to clarify (at least for myself) our ABI stability  
policy.  Fwiw, I consider reducing the sizeof std::types very important.

Thanks,
Howard