Performance issue with COW strings?

Paolo Carlini pcarlini@suse.de
Mon Jun 6 16:56:00 GMT 2005


Benjamin Kosnik wrote:

>>Excellent! We badly need this kind of hard data. Well, I could prepare a
>>simple patch just disabling refererence-counting, would be simple.
>>However, since you don't care about compatibility, I would ask you to
>>wait just a couple of weeks: after some overdue branch maintainance, I
>>will add to v7-branch a simple alternate base class for basic_string,
>>avoiding ref-counting, selectable at configure time. Carrying out
>>comparisons will be very easy and meaningful, because the full set of
>>"utility" functions will remain the same, only the low-level bits will
>>be different in the two configurations.
>>    
>>
>Paolo, I think you need to reconsider adding the policy-based string
>class to mainline as an extension class. I know we've discussed this
>before, but I still think this is a smart move. 
>
>Any chance you can reconsider your position?
>  
>
... and, if you remember the *last* exchanges, I'm all for it! Just let
me add the alternate base-class to the v7-branch code, test it a bit and
then we can get in contact privately about the best way to do that.
There are quite a few nasty issues, as you remember ;)

Paolo.

P.S. With Dan we are in contact about his tests.



More information about the Libstdc++ mailing list