"leaked" Property list in nc4_open_file in hdf5open.c

[planetMDX]

version: v4.9.2
hdf5: 1.14.1-2
gcc version 13.1.1
os: arch linux
hdf5open

 {
      hid_t pid;
      unsigned int crt_order_flags;
      if ((pid = H5Fget_create_plist(h5->hdfid)) < 0)
          BAIL(NC_EHDFERR);
      if (H5Pget_attr_creation_order(pid, &crt_order_flags) < 0)
          BAIL(NC_EHDFERR);
      if (!(crt_order_flags & H5P_CRT_ORDER_TRACKED)) {
	  nc4_info->no_attr_create_order = NC_TRUE;
      }
    }

is missing a H5Pclose on the pid.

#include <H5public.h>
#include <netcdf.h>
#include <stdio.h>
#include <stdlib.h>
#include <hdf5.h>
#ifndef ITERATIONS_DEFAULT
#define ITERATIONS_DEFAULT 200
#endif
#ifndef NETCDF_FILEPATH_DEFAULT
#define NETCDF_FILEPATH_DEFAULT                                                \
    "/path/to/OMI-Aura_L2-example.nc"
#endif
int main(void) {
    int iterations = ITERATIONS_DEFAULT;
    H5open();
    nc_initialize();
    const char *netcdf_file_path = NETCDF_FILEPATH_DEFAULT;
    printf("iterations: %d\tfilepath: %s\n", iterations, netcdf_file_path);
    setenv("HDF5_USE_FILE_LOCKING", "FALSE", 1);
    for (int c = 0; c < iterations; ++c) {
        int nid = 0;
        nc_open(netcdf_file_path, 0, &nid);
        nc_close(nid);
    }
    _Exit(0);
}

produces this run-val2.log valgrind log when compiled under -O0

line 39379 to 39391 are of relevance here:

==2241928== 78,568 bytes in 1,403 blocks are still reachable in loss record 2,875 of 2,878
==2241928==    at 0x4841848: malloc (vg_replace_malloc.c:431)
==2241928==    by 0x4EAD1A7: H5MM_malloc (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x4D9DD97: H5FL__malloc (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x4D9E649: H5FL_reg_malloc (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x5053AA3: H5SL__new_node (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x505A789: H5SL__insert_common (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x505B59E: H5SL_insert (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x4FB01CA: H5P_copy_plist (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x51F9654: H5VL__native_file_get (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x51CB240: H5VL__file_get (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x51CB442: H5VL_file_get (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x4CF41BC: H5Fget_create_plist (in /home/plane/opt/lib/hdf5/lib/libhdf5.so.310.1.0)
==2241928==    by 0x49043E6: NC4_hdf5get_superblock (nc4hdf.c:2489)
==2241928==    by 0x4904C9B: NC4_read_provenance (nc4info.c:199)
==2241928==    by 0x490D87F: nc4_open_file (hdf5open.c:964)
==2241928==    by 0x490D9D1: NC4_open (hdf5open.c:1040)
==2241928==    by 0x48893A9: NC_open (dfile.c:2172)
==2241928==    by 0x48881ED: nc_open (dfile.c:668)
==2241928==    by 0x10922B: main (test_.c:22)

while if there is an H5Pclose earlier, valgrind produces something like this: run-val.fixed.log, which doesn't feature such a section. Also peak memory usage is down by ~0.2MB under these conditions on my machine.

[DennisHeimbigner]

Good catch, thanks.

[DennisHeimbigner]

Fixed by PR #2725