Re: [boost] [chrono] Valgrind complains

Subject: Re: [boost] [chrono] Valgrind complains
From: vicente.botet (vicente.botet_at_[hidden])
Date: 2010-06-23 03:13:15

----- Original Message ----- 
From: "Roland Bock" <rbock_at_[hidden]>
To: <boost_at_[hidden]>
Sent: Wednesday, June 23, 2010 8:05 AM
Subject: Re: [boost] [chrono] Valgrind complains

> vicente.botet wrote:
>> ----- Original Message ----- 
>> From: "Roland Bock" <rbock_at_[hidden]>
>> To: <boost_at_[hidden]>
>> Sent: Tuesday, June 22, 2010 8:23 PM
>> Subject: Re: [boost] [chrono] Valgrind complains
>>
>>
>>   
>>> vicente.botet wrote:
>>>     
>>>> ----- Original Message ----- 
>>>> From: "Roland Bock" <rbock_at_[hidden]>
>>>> To: <boost_at_[hidden]>
>>>> Sent: Tuesday, June 22, 2010 5:34 PM
>>>> Subject: Re: [boost] [chrono] Valgrind complains
>>>>
>>>>
>>>>   
>>>>       
>>>>>>> From: "Roland Bock" <rbock_at_[hidden]>
>>>>>>>     
>>>>>>>         
>>>>>>>             
>>>>>>>> Valgrind complains about
>>>>>>>>
>>>>>>>> Conditional jump or move depends on uninitialised value(s)
>>>>>>>>
>>>>>>>>
>>>>>>>> several times for the following line:
>>>>>>>>
>>>>>>>> { boost::chrono::stopclock<> stopclock(std::cerr); }
>>>>>>>>       
>>>>>>>>               
>>>> Could you check if Valgrind complains with
>>>>
>>>> { boost::chrono::stopclock<> stopclock; }
>>>> or
>>>> { boost::chrono::stopclock<> stopclock(BOOST_CURRENT_FUNCTION); }
>>>> or
>>>> { boost::system::error_code ec; 
>>>> boost::chrono::stopclock<> stopclock(std::cerr, ec); }
>>>>
>>>> Thanks, 
>>>> Vicente
>>>>
>>>>
>>>>   
>>>>       
>>> Hi,
>>>
>>> attached are
>>>
>>>    * the source file containing the original test plus the three
>>>      additional tests suggested by you
>>>    * a file containing compiler warnings
>>>    * the valgrind output (the latter two tests do not produce warnings,
>>>      if I read it correctly)
>>>
>>>
>>> The program was compiled using the following compiler flags (gcc=4.2.4 
>>> on Ubuntu 8.04, 64bit):
>>>
>>> -Wall -Wreorder -Wnon-virtual-dtor -Wno-non-template-friend 
>>> -Woverloaded-virtual -Wsign-promo -Wextra -fvisibility=hidden 
>>> -D_GNU_SOURCE -O0 -g
>>>     
>>  
>>
>> I have seen some places where the parameter system::error_code & ec is not initialized when the function succeed. I have update the code so the variable is initialized. Most of the tests are done with the default system::throws parameter so these errors are not catched. I will try to complete the tests for the no-throw cases soon.  
>>
>> Roland, please could you download the new version, and check if the Valgrind issue is resolved?
>>
>> Thanks in advance,
>> Vicente
>>   
> Hi,
> 
> the compile warnings are gone, but valgrind stays unhappy.
> 
> I compiled libboost_chrono.so in debug mode, this time. This way,
> valgrind's output should be more helpful inside the library's code, see
> attachment.
> 
> Regards,
> 
> Roland

Hi,

it seems there was an old bug on duration operator=()

Could you replace the line
            if (rhs != *this) rep_= rhs.rep_;

by

            if (&rhs != this) rep_= rhs.rep_;

or just

            //~ if (rhs != &this) 
            rep_= rhs.rep_;

Let me know if this solves the issue.

Best,

Vicente