[Patch] libstdc++/24244 (2nd try)

[Kurt Kohler]

I've now changed the real program that started all this back to using 
stdc++ shared_ptr and weak_ptr. It works!
This program creates and destroys thousands of shared_ptr's so it should 
be a pretty rigorous test.

Thanks to everyone who help with this!

Kurt

Paolo Carlini wrote:

>Kurt Kohler wrote:
>
>  
>
>>I'm sure everyone will be happy to hear that valgrind is now happy
>>with my test program. It no longer complains about the uninitialized
>>mutex. Yay!
>>    
>>
>
>Excellent. As soon as mainline reopens for normal bugfixing, I will
>commit it.
>
>  
>
>>Did anyone ever figure out why no one else could reproduce it?
>>    
>>
>
>Not really, sadly, but this kind of bug (uninitialized vars in general,
>I mean) is intrinsically nasty from the point of view of
>reproducibility: I'm sure that in the past even bugs in the compiler
>proper required very specific conditions to appear (e.g., bootstrap
>broken only for one architecture).
>
>But the bug itself is very real, thanks for your perseverance ;)
>
>  
>
>>It will take a bit longer to change my real program back to using the
>>stdc++ shared_ptr. I'll let you know what I find although I don't
>>expect any problems.
>>    
>>
>
>Thanks again,
>Paolo.
>  
>