h5ls cores on toss4 system @llnl compiled with IntelOne API

[markcmiller86]

I am attaching a tar file with as much info as I can provide.

There is no urgency here. I am reporting on an FYI basis.

This is happening on LLNL’s rztrona and I believe HDF5-1.14.0 was compiled with IntelOne API as the gdb-session file suggests.

On LLNL’s CZ, I think any toss4 system with IntelOneAPI compilers should be able to reproduce.

h5ls-core.tgz

[lrknox]

Core dump observed on Quartz when running h5ls -vld domain0.silo/MESH with h5ls built from HDF5 1.14.0 with OpenMPI 4.1.2 / intel/2023.2.1-magic / cmake/3.26.3 modules. The same command with h5ls built from HDF5 1.14.1 and 1.14.2 release tags and hdf5_1_14 and develop branches produces this output:

quartz1916{knox8}229: ../build/bin/h5ls -vld domain0.silo/MESH
Opened "domain0.silo" with sec2 driver.
MESH Type
Attribute: silo scalar
Type: struct {
"ndims" +0 native int
"nnodes" +4 native int
"nzones" +8 native int
"facetype" +12 native int
"coord_sys" +16 native int
"planar" +20 native int
"group_no" +24 native int
"time" +28 native float
"dtime" +32 native double
"zonelist" +40 9-byte null-terminated ASCII string
"min_extents" +49 [3] native double
"max_extents" +73 [3] native double
"coord0" +97 15-byte null-terminated ASCII string
"coord1" +112 15-byte null-terminated ASCII string
"coord2" +127 15-byte null-terminated ASCII string
"label0" +142 2-byte null-terminated ASCII string
"label1" +144 2-byte null-terminated ASCII string
"label2" +146 2-byte null-terminated ASCII string
"units0" +148 3-byte null-terminated ASCII string
"units1" +151 3-byte null-terminated ASCII string
"units2" +154 3-byte null-terminated ASCII string
} 157 bytes
Data:
{ndims=3, nnodes=1000, nzones=729, facetype=100, coord_sys=120, planar=124, group_no=-1, time=0,
dtime=0, zonelist="zonelist", min_extents=[0,0,0], max_extents=[2.66666666666667,3,3],
coord0="/.silo/#1", coord1="/.silo/#2", coord2="/.silo/#3", label0="X", label1="Y",
label2="Z", units0="cm", units1="cm", units2="cm"}
Attribute: silo_type scalar
Type: native int
Data:
510
Location: 1:14912
Links: 1
Type: shared-1:14912native int

Output from HDF5 1.14.0 with the segfault was the same except for the last line (after Links:    1).
It appears the problem was fixed before the HDF5 1.14.1 release, somewhere between December 2022 and May 2023