I'm in the process of migrating from Solaris x86 to RHEL7 and am having issues reading (on RHEL) one particular data set that was written on the Solaris platform. The Solaris HDF5 release is 1.8.9, and RHEL is 1.8.12. I have not tried newer versions of HDF5 as 1.8.12 is what RHEL pre-packages and I'd prefer not to have to do manual builds.
The debugger says:
Program received signal SIGSEGV, Segmentation fault.
0x00007ffff5a1a81c in H5T_cmp (dt1=0x6cccc0, dt2=0x7cc9c0, superset=0)
at ../../src/H5T.c:3948
3948 if(dt1->shared->type < dt2->shared->type)
(gdb) print dt1
$1 = (const H5T_t *) 0x6cccc0
(gdb) print dt1->shared
$2 = (H5T_shared_t *) 0x6ccd30
(gdb) print dt1->shared->type
$3 = H5T_ENUM
(gdb) print dt2
$4 = (const H5T_t *) 0x7cc9c0
(gdb) print dt2->shared
$5 = (H5T_shared_t *) 0x100014d98
(gdb) print dt2->shared->type
Cannot access memory at address 0x100014da4
(gdb) where
#0 0x00007ffff5a1a81c in H5T_cmp (dt1=0x6cccc0, dt2=0x7cc9c0, superset=0)
at ../../src/H5T.c:3948
#1 0x00007ffff5a1b13e in H5T_cmp (dt1=0x6cc270, dt2=0x7c4070, superset=0)
at ../../src/H5T.c:4032
#2 0x00007ffff5a1c7dd in H5T_path_find (src=0x6cc270, dst=0x6c2630, name=0x0,
func=0x0, dxpl_id=167772168, is_api=0) at ../../src/H5T.c:4436
#3 0x00007ffff57dfe56 in H5D__typeinfo_init (dset=0x6f71b0,
dxpl_cache=0x7ffff5e2cd40 <H5D_def_dxpl_cache>, dxpl_id=167772168,
mem_type_id=50332135, do_write=0, type_info=0x7fffffffc570)
at ../../src/H5Dio.c:935
#4 0x00007ffff57ddd56 in H5D__read (dataset=0x6f71b0, mem_type_id=50332135,
mem_space=0x7154f0, file_space=0x715930, dxpl_id=167772168, buf=0x7c5480)
at ../../src/H5Dio.c:434
#5 0x00007ffff57dc5d7 in H5Dread (dset_id=83886080, mem_type_id=50332135,
mem_space_id=67108867, file_space_id=67108866, plist_id=167772168,
buf=0x7c5480) at ../../src/H5Dio.c:178
The same file can be processed without incident with the same application on Solaris. The dataset in question is being created using the high level tables API and a compound data type. I can provide a trace file or other debugging information to help resolve this issue. It didn't seem like a good idea to post files to the mailing list.
Any help would be appreciated.