romio fails to build due to missing stdatomic.h

[naughtont3]

when building an ompi-5.0.x snapshot i get errors about building romio due to missing stdatomic.h.

This seems like a new dependency/build require. Should this have been caught during configure phase?

Background information

What version of Open MPI are you using? (e.g., v3.0.5, v4.0.2, git branch name and hash, etc.)

• openmpi-5.0.0 nightly tarball 12-may-2022

Describe how Open MPI was installed (e.g., from a source/distribution tarball, from a git clone, from an operating system distribution package, etc.)

• nightly tarball
• gcc/g++ 4.8.5 (i know it is old)

Please describe the system on which you are running

• Operating system/version: Linux
• Computer hardware: x86-64
• Network type:

---

Details of the problem

A build of openmpi fails during compile phase due to missing stdatomic.h while building 3rd-party/romio341.

./configure \
       --enable-debug \
       --with-devel-headers \
    --enable-mpirun-prefix-by-default \
    --prefix=${OMPI_INSTALL_DIR} \

 ...<snip>...

make[2]: Entering directory `/home/3t4/projects/ompi-x/scratch/CLOUDY-FALCON/openmpi-v5.0.x-202205120246-ef29ef7/build2/3rd-party/romio341'
Making all in mpl
make[3]: Entering directory `/home/3t4/projects/ompi-x/scratch/CLOUDY-FALCON/openmpi-v5.0.x-202205120246-ef29ef7/build2/3rd-party/romio341/mpl'
  CC       src/atomic/mpl_atomic.lo
  CC       src/bt/mpl_bt.lo
  CC       src/env/mpl_env.lo
  CC       src/dbg/mpl_dbg.lo
In file included from ../../../../3rd-party/romio341/mpl/include/mpl_atomic.h:101:0,
                 from ../../../../3rd-party/romio341/mpl/src/atomic/mpl_atomic.c:11:
../../../../3rd-party/romio341/mpl/include/mpl_atomic_c11.h:20:23: fatal error:stdatomic.h: No such file or directory
 #include <stdatomic.h>
                       ^
compilation terminated.
In file included from ../../../../3rd-party/romio341/mpl/include/mpl_atomic.h:101:0,
                 from ../../../../3rd-party/romio341/mpl/include/mpl.h:15,
                 from ../../../../3rd-party/romio341/mpl/src/env/mpl_env.c:6:
../../../../3rd-party/romio341/mpl/include/mpl_atomic_c11.h:20:23: fatal error:stdatomic.h: No such file or directory
 #include <stdatomic.h>
                       ^
compilation terminated.
In file included from ../../../../3rd-party/romio341/mpl/include/mpl_atomic.h:101:0,
                 from ../../../../3rd-party/romio341/mpl/include/mpl.h:15,
                 from ../../../../3rd-party/romio341/mpl/src/dbg/mpl_dbg.c:12:
../../../../3rd-party/romio341/mpl/include/mpl_atomic_c11.h:20:23: fatal error:stdatomic.h: No such file or directory
 #include <stdatomic.h>
                       ^
compilation terminated.
In file In file In file included from ../../../../3rd-party/romio341/mpl/include/mpl_atomic.h:101:0,
                 from ../../../../3rd-party/romio341/mpl/include/mpl.h:15,
                 from ../../../../3rd-party/romio341/mpl/src/bt/mpl_bt.c:6:
../../../../3rd-party/romio341/mpl/include/mpl_atomic_c11.h:20:23: fatal error:stdatomic.h: No such file or directory
 #include <stdatomic.h>
                       ^
compilation terminated.
make[3]: *** [src/atomic/mpl_atomic.lo] Error 1
make[3]: *** Waiting for unfinished jobs....
make[3]: *** [src/env/mpl_env.lo] Error 1
make[3]: *** [src/dbg/mpl_dbg.lo] Error 1
make[3]: *** [src/bt/mpl_bt.lo] Error 1
make[3]: Leaving directory `/home/3t4/projects/ompi-x/scratch/CLOUDY-FALCON/openmpi-v5.0.x-202205120246-ef29ef7/build2/3rd-party/romio341/mpl'
make[2]: *** [all-recursive] Error 1
make[2]: Leaving directory `/home/3t4/projects/ompi-x/scratch/CLOUDY-FALCON/openmpi-v5.0.x-202205120246-ef29ef7/build2/3rd-party/romio341'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/3t4/projects/ompi-x/scratch/CLOUDY-FALCON/openmpi-v5.0.x-202205120246-ef29ef7/build2/3rd-party'
make: *** [all-recursive] Error 1

[naughtont3]

Possibly ignore b/c i have ancient gcc/g++, but figured I'd add just in case.

[jsquyres]

@naughtont3 Perfectly valid. configure should elect not to build ROMIO in this case. Sounds like a legit bug.

[awlauria]

FWIW - it looks like it is a issue specific to gcc v4.8, not present on gcc v4.9.

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58016
https://gcc.gnu.org/gcc-4.9/changes.html

[ggouaillardet]

And FWIW, from 3rd-party/romio341/mpl/include/mpl_atomic.h

#if defined(MPL_USE_NO_ATOMIC_PRIMITIVES)
#include "mpl_atomic_none.h"
#elif defined(MPL_HAVE_C11_ATOMICS)
#include "mpl_atomic_c11.h"
#elif defined(MPL_HAVE_GCC_INTRINSIC_ATOMIC)
#include "mpl_atomic_gcc_atomic.h"

That seems to point at a configure but when detecting C11 atomics

[ggouaillardet]

The root cause is 3rd-party/romio341/mpl/include/mplconfig.h should not be in the tarball (this is an automatically generated file)

That being said, I am not sure why this file is not re-generated at configure time, and if it is and VPATH is used, why the file in the source directory is used instead of the one in the build directory.

For the time being, a workaround is to remove this file after untarring Open MPI and before running configure

[awlauria]

@naughtont3 this got overlooked, and is finally merged here into v5.0.x: #11323

if it turns out the fix come from upstream ompio like we would prefer, we'll revert it out. closing.