Test failures on MacOS

[danielhollas]

Two tests do not pass on my macbook: SH_BUTCHER_PHASE and SH_RK4_PHASE. The problem appears to be in empty restart_sh.bin.diff. The binaries are different based on simple diff -q restart_sh.bin restart_sh.bin.ref but the subsequent diff -y -W 500 restart_sh.bin restart_sh.bin.ref | grep -e '|' -e '<' -e '>' > $test_file.diff produces empty file, which then breaks numdif.py.

=======================
Entering directory SH_BUTCHER_PHASE
Comparing numerical differences in file hopgeom.3.2.6.xyz.diff
Comparing numerical differences in file restart.xyz.5.diff
Comparing numerical differences in file restart.xyz.diff
Comparing numerical differences in file restart_sh.bin.diff
File 'restart_sh.bin.diff' is empty!
Comparing numerical differences in file stateall_grad.dat.diff
Comparing numerical differences in file wfcoef.dat.diff
SH_BUTCHER_PHASE FAILED
=======================

and

=======================
Entering directory SH_RK4_PHASE
Comparing numerical differences in file bkl.dat.diff
Comparing numerical differences in file restart.xyz.10.diff
Comparing numerical differences in file restart.xyz.diff
Comparing numerical differences in file restart_sh.bin.10.diff
File 'restart_sh.bin.10.diff' is empty!
Comparing numerical differences in file restart_sh.bin.diff
File 'restart_sh.bin.diff' is empty!
Comparing numerical differences in file wfcoef.dat.diff
SH_RK4_PHASE FAILED
=======================

Originally posted by @JanosJiri in #219 (review)

[danielhollas]

@JanosJiri thanks a lot for testing. Can you please copy-paste your make.vars and send your compiler version? It seems that on CI even with macos-26 the tests are passing.

(obviously no rush on this, maybe we can look together once I am in Prague)

[JanosJiri]

It might be the compilers, I had to slightly customize make.vars to complie ABIN, but it was a very quick solution. I will have more time to look into it once the thisis is over.. we can have a look when you are in Prague. I guess it's just my own set up since it compiles here.

# Local build configuration, included into Makefile

# This file was built as follows.
# ./configure 

# Name of the final binary
BIN = abin

# Compilers
FC = /opt/homebrew/bin/gfortran-15
CXX = /opt/homebrew/bin/g++-15

# Compiler flags
FFLAGS = -O2 -fopenmp -fimplicit-none -Wall -Wno-integer-division -Wno-maybe-uninitialized
CXXFLAGS = -O2 -Wall -std=c++11
LDFLAGS = 
LDLIBS = 

# MPI library
MPI = FALSE
MPI_PATH = 

# Compilation with Plumed library
PLUMED = FALSE
PLUMED_INC = 

# TCBP interface to TeraChem
TCPB = FALSE
TCPB_LIB = 

# FFTW library is typically shipped with your system
# It is needed for normal mode coordinate transformation in PIMD
FFTW = FALSE

# Experimental cross-compilation with CP2K
CP2K = FALSE

# Compile with unit tests
PFUNIT_PATH = 

[danielhollas]

Thanks, interesting. In CI we also run gfortran-15 but not from Homebrew. As a quick additional test, you could try compiling with -O0 since that's what we do in CI.