debug algorithms

François Dumont francois.cppdevs@free.fr
Tue Sep 7 20:15:00 GMT 2010


I didn't expect that in addition to guaranty binary compatibility of 
libstdc++ itself we also have to guaranty binary compatibility of 
different components built with different versions of libstdc++, 
especially in debug mode ! What I meant with 100% compatibility was that 
I do not expect abi or any other test to fail after this patch.

I also noticed that you added on the 31 of august a cached index in the 
hashtable implementation. This is going to introduce exactly the same 
kind of breaking change, isn't it ? IMO libraries already using this 
implementation and exposing it are much more common than libraries 
exposing debug mode structures.

I will start to consider forward_list.

François


On 09/06/2010 10:31 PM, Paolo Carlini wrote:
> On 09/06/2010 09:44 PM, François Dumont wrote:
>    
>>    IMHO, even if this patch has a lot of impact on the code, it is 100%
>> compatible with existing user code
>>      
> If you are changing _Safe_iterator, like this:
>
> -  template<typename _Iterator, typename _Sequence>
> +  template<typename _Sequence, bool _IsConst>
>       class _Safe_iterator;
>
> then certainly it is not: two different objects, one built with the old
> headers and one with the new headers and passing around an iterator
> cannot be linked together. I warn you these headers are tricky, we
> really want to be 100% sure we are not breaking anything for 4.6.0, in
> the various modes and without penalizing performance (eg, make sure the
> fill and copy for deque::iterator are still used in normal mode, etc).
>
> If you want my advice, let's first make sure we have a working debug
> mode for forward_list in 4.6.0, and let's give the algorithms *more*
> thought.
>
> Paolo.
>
>
>    



More information about the Libstdc++ mailing list