memory leak in pnetcdf?

[Wei-keng Liao]

Hi, Ed

Thanks for reporting the problem.

I can reproduce the same valgrind output when using 1.6.1, but not 1.7.0.
The source locations pointed out by valgrind does not seem to be the cause as
the files are basically the same between the two PnetCDF versions.

I suggest to use 1.7.0, because some bugs have been fixed since 1.6.1 that may
also fix the memory leak by accident.

Wei-keng

On Mar 18, 2016, at 12:30 PM, Ed Hartnett wrote:

> Howdy pnetcdf folk!
> 
> While working on the PIO library I found what appears to be a memory leak in pnetcdf.
> 
> I am using version 1.6.1 of the parallel-netcdf library. The following simple program results in a 5-byte leak, as indicated by the valgrind output which follows the code listing.
> 
> A little experimentation shows that it is the attribute name which is being copied to a new memory location and then never freed.
> 
> Thanks,
> Ed
> 
> 
> #include <pnetcdf.h>
> 
> int
> main(int argc, char **argv)
> {
>     int ncid, ret;
>     
>     /* Initialize MPI. */
>     if ((ret = MPI_Init(&argc, &argv)))
> 	return ret;
> 
>     /* Create the netCDF output file. */
>     if ((ret = ncmpi_create(MPI_COMM_WORLD, "test_pnetcdf_leak.nc", NC_CLOBBER,
> 			    MPI_INFO_NULL, &ncid)))
> 	return ret;
> 
>     /* Define a global attribute. */
>     int att_val = 42;
>     if ((ret = ncmpi_put_att_int(ncid, NC_GLOBAL, "foo1", NC_INT,
> 				 1, &att_val)))
> 	return ret;
> 
>     /* Close the netCDF file. */
>     if ((ret = ncmpi_close(ncid)))
> 	return ret;
> 
>     /* Finalize the MPI library. */
>     MPI_Finalize();
> 
>     return 0;
> }
> 
> 
> 
> 
> ==10556== Memcheck, a memory error detector
> ==10556== Copyright (C) 2002-2013, and GNU GPL'd, by Julian Seward et al.
> ==10556== Using Valgrind-3.10.1 and LibVEX; rerun with -h for copyright info
> ==10556== Command: ./test_pnetcdf_leak
> ==10556== 
> ==10556== Warning: invalid file descriptor -1 in syscall read()
> ==10556== 
> ==10556== HEAP SUMMARY:
> ==10556==     in use at exit: 5 bytes in 1 blocks
> ==10556==   total heap usage: 631 allocs, 630 frees, 25,236,667 bytes allocated
> ==10556== 
> ==10556== 5 bytes in 1 blocks are definitely lost in loss record 1 of 1
> ==10556==    at 0x4A08B9D: realloc (in /usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
> ==10556==    by 0x4C75C68: utf8proc_map (dutf8proc.c:550)
> ==10556==    by 0x4C75D04: utf8proc_NFC (dutf8proc.c:566)
> ==10556==    by 0x4D3616B: ncmpii_NC_findattr (in /usr/local/lib/libnetcdf.so.11.0.0)
> ==10556==    by 0x4D391C9: ncmpii_put_att (in /usr/local/lib/libnetcdf.so.11.0.0)
> ==10556==    by 0x4D3969D: ncmpi_put_att_int (in /usr/local/lib/libnetcdf.so.11.0.0)
> ==10556==    by 0x4009A3: main (test_pnetcdf_leak.c:23)
> ==10556== 
> ==10556== LEAK SUMMARY:
> ==10556==    definitely lost: 5 bytes in 1 blocks
> ==10556==    indirectly lost: 0 bytes in 0 blocks
> ==10556==      possibly lost: 0 bytes in 0 blocks
> ==10556==    still reachable: 0 bytes in 0 blocks
> ==10556==         suppressed: 0 bytes in 0 blocks
> ==10556== 
> ==10556== For counts of detected and suppressed errors, rerun with: -v
> ==10556== ERROR SUMMARY: 1 errors from 1 contexts (suppressed: 0 from 0)
> [ed at localhost unit]$