run_test.sh: "PASS write profile" does not actually write all profiles and should FAIL

[vasdommes]

https://github.com/davidsd/sdpb/blob/b359336/test/run_test.sh#L31
"write profile" test in test/run_test.sh does not actually check that profiles are written to disk.
Profile for rank=0 is not written, El::mpi::Abort() is called, but the test nevertheless "passes".

Repro

(checked on Expanse cluster and on a laptop with WSL+Ubuntu)

1. Replace the lines https://github.com/davidsd/sdpb/blob/b359336/test/run_test.sh#L26-L39
   with

rm -rf test/io_tests

mkdir -p test/io_tests
touch test/io_tests/profile_error.profiling.0
chmod a-w test/io_tests/profile_error.profiling.0
mpirun -n 2 --quiet ./build/sdpb --precision=1024 --noFinalCheckpoint --procsPerNode=1 -s test/test -c test/io_tests/profile_error --verbosity=2 --maxIterations=1 #2>/dev/null > /dev/null
if [ $? != 0 ]
then
    echo "PASS write profile"
else
    echo "FAIL write profile"
    result=1
fi
exit $result

(i.e. do not remove test/io_tests/ after run and do not redirect output to /dev/null)

2. run ./test/run_test.sh, see at the end of output:

Process 0 caught error message:
Error when writing to: test/io_tests/profile_error.profiling.0
PASS write profile

3. Observe that test/io_tests/profile_error.profiling.0 is empty (at the same time test/io_tests/profile_error.profiling.1 is OK).

[vasdommes]

My bad, didn't notice that the test passes when mpirun fails (i.e. if [ $? != 0 ]), so this behavior is "as designed".

However, it is very unclear when you look at the test name.
I'd suggest having a test for successful "write profile" and another test "write profile should fail".