[mpich2-dev] nesting level warning with MPI_File_preallocate(fh, 0)

[Dave Goodell]

Valgrind won't report this type of leak.  I can explain the exact  
reason why if it really matters to you for some reason, but the basic  
takeaway is that you won't catch this with valgrind.  Someday I'll  
improve our valgrind integration and you will get this message as well  
as a message from valgrind, but not today.

-Dave

On Dec 9, 2008, at 4:28 PM, Lisandro Dalcin wrote:

> Well, my network is working fine again, so I could do svn update.
> Running the code I've posted under valgrind gives the output below:
> Clearly, no memory leak is reported, but the handle tracing code still
> shows the warning.
>
> $ valgrind a.out
> ==27547== Memcheck, a memory error detector.
> ==27547== Copyright (C) 2002-2007, and GNU GPL'd, by Julian Seward  
> et al.
> ==27547== Using LibVEX rev 1804, a library for dynamic binary  
> translation.
> ==27547== Copyright (C) 2004-2007, and GNU GPL'd, by OpenWorks LLP.
> ==27547== Using valgrind-3.3.0, a dynamic binary instrumentation  
> framework.
> ==27547== Copyright (C) 2000-2007, and GNU GPL'd, by Julian Seward  
> et al.
> ==27547== For more details, rerun with: -v
> ==27547==
> In direct memory block for handle type ATTRIBUTE KEY, 2 handles are
> still allocated
> ==27547==
> ==27547== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 17  
> from 1)
> ==27547== malloc/free: in use at exit: 0 bytes in 0 blocks.
> ==27547== malloc/free: 155 allocs, 155 frees, 8,436,618 bytes  
> allocated.
> ==27547== For counts of detected errors, rerun with: -v
> ==27547== All heap blocks were freed -- no leaks are possible.
>
>
> On Tue, Dec 9, 2008 at 7:59 PM, Dave Goodell <goodell at mcs.anl.gov>  
> wrote:
>> On Dec 9, 2008, at 3:09 PM, Robert Latham wrote:
>>
>>> On Tue, Dec 09, 2008 at 03:02:48PM -0600, Rajeev Thakur wrote:
>>>>
>>>> I guess that's because you have built a version with memory tracing
>>>> turned
>>>> on.
>>>
>>>>> In direct memory block for handle type ATTRIBUTE KEY, 2 handles  
>>>>> are
>>>>> still allocated
>>>
>>> I wonder if this is fixed in svn?  I have an mpich with memory  
>>> tracing
>>> and neither the built-in tracing or valgrind complains about  
>>> anything
>>> leaks.
>>
>> This is actually output from the recently(-ish) added handle  
>> tracing code,
>> not the memory tracing code.  You can enable handle tracing via
>> --enable-g=handle or --enable-g=all.
>>
>> -Dave
>>
>>
>
>
>
> -- 
> Lisandro Dalcín
> ---------------
> Centro Internacional de Métodos Computacionales en Ingeniería (CIMEC)
> Instituto de Desarrollo Tecnológico para la Industria Química (INTEC)
> Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET)
> PTLC - Güemes 3450, (3000) Santa Fe, Argentina
> Tel/Fax: +54-(0)342-451.1594