libstdc++/16612, empty basic_strings can't live in shared memory

[Mark Mitchell]

Nathan Myers wrote:

>On Wed, Aug 04, 2004 at 09:27:37PM +0200, Paolo Carlini wrote:
>  
>
>>However, as Nathan wrote, "let's not be  hasty": it's the first time
>>in *years* that someone notices this problem, and I still have hopes
>>there is an even better solution: operator[] is such a *basic*
>>operation and would be *so* sad adding a conditional...
>>    
>>
>
>Not to mention that length(), capacity(), and (really) everything
>else besides would need the same treatment.
>
>I like the idea of using static space in libsupc++.  It's simpler,
>and should improve performance of the plain vanilla string, too, by 
>avoiding relocation indirections.  Can anybody say whether it could 
>be arranged for static storage in libsupc++ to appear at the same 
>address in all programs compiled for a given ABI version?
>  
>
It cannot be so arranged on some systems.  For example, some systems 
allocate memory to shared objects based on link order, so unless 
libsupc++ is always the first library, one can't be sure it will end up 
at the same place.  Also, if multiple libraries want to be loaded at the 
same address, the dynamic linker will have to pick one.

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