Building pyarpack fails on 32-bit arch: virtual memory exhausted

[badshah400]

Expected behavior

Running make -jN after cmake -DPYTHON3=ON should work and compile arpack + pyarpack, no matter the machine architecture.

Actual behavior

[  164s] virtual memory exhausted: Operation not permitted
[  164s] make[2]: *** [CMakeFiles/pyarpack.dir/build.make:1255: CMakeFiles/pyarpack.dir/EXAMPLES/PYARPACK/pyarpack.cpp.o] Error 1
[  164s] make[2]: Leaving directory '/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/build'
[  164s] make[1]: *** [CMakeFiles/Makefile2:388: CMakeFiles/pyarpack.dir/all] Error 2
[  164s] make[1]: Leaving directory '/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/build'
[  164s] make: *** [Makefile:185: all] Error 2

even though I am building this on a VM with 28 GB memory allocated to it, and have no problems building pyarpack for 64-bit architectures. Full log here: https://paste.opensuse.org/32848229.

Where/how to reproduce the problem

• arpack-ng: 3.8.0
• OS: openSUSE Linux
• compiler: GCC 10.2.1
• environment:

[    8s] -- Configuration summary for arpack-ng-3.8.0:
[    8s]    -- prefix: /usr
[    8s]    -- MPI: OFF
[    8s]    -- ICB: ON
[    8s]    -- INTERFACE64: 0
[    8s]    -- FC:      /usr/bin/gfortran
[    8s]    -- FCFLAGS: -O2 -g -DNDEBUG -fomit-frame-pointer -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clas
h-protection -Werror=return-type -flto=auto -g -fPIC -DNDEBUG
[    8s]    -- CC:      /var/lib/build/ccache/bin/cc
[    8s]    -- CCFLAGS: -O2 -g -DNDEBUG -fomit-frame-pointer -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clas
h-protection -Werror=return-type -flto=auto -g -fPIC -DNDEBUG
[    8s]    -- CXX:      /var/lib/build/ccache/bin/c++
[    8s]    -- CXXFLAGS: -O2 -g -DNDEBUG -fomit-frame-pointer -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-cla
sh-protection -Werror=return-type -flto=auto -g -fPIC -DNDEBUG
[    8s]    -- BLAS:
[    8s]       -- link:    /usr/lib/libblas.so
[    8s]    -- LAPACK:
[    8s]       -- link:    /usr/lib/liblapack.so
[    8s]    -- EIGEN3:
[    8s]       -- compile: /usr/include/eigen3
[    8s]    -- PYTHON:
[    8s]       -- compile: /usr/include/python3.8
[    8s]       -- link:    /usr/lib/libpython3.8.so
[    8s]       -- exe:     /usr/bin/python3
[    8s]    -- BOOST:
[    8s]       -- compile: /usr/include
[    8s]       -- link:    Boost::python
[    8s]       -- link:    Boost::numpy

Steps to reproduce the problem

• Configure with -DPYTHON3=ON on i586
• Run make.

Notes, remarks

When pyarpack is disabled, the build proceeds fine.

[sylvestre]

"VM with 28 GB" => the issue is probably triggered by limitation of i386. Doesn't matter how much memory you have on the system

I guess this is this command:

[   97s] /var/lib/build/ccache/bin/c++ -DBOOST_ALL_NO_LIB -DBOOST_NUMPY_DYN_LINK -DBOOST_PYTHON_DYN_LINK -DPY_MAJOR_VERSION=\"3\" -Dpyarpack_EXPORTS -I/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/build -I/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0 -I/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/ICB -I/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/EXAMPLES/MATRIX_MARKET -I/home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/EXAMPLES/PYARPACK -I/usr/include/eigen3 -I/usr/include/python3.8 -fomit-frame-pointer -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection -Werror=return-type -flto=auto -g -fPIC -DNDEBUG -O2 -g -DNDEBUG -fPIC -std=gnu++11 -o CMakeFiles/pyarpack.dir/EXAMPLES/PYARPACK/pyarpack.cpp.o -c /home/abuild/rpmbuild/BUILD/arpack-ng-3.8.0/EXAMPLES/PYARPACK/pyarpack.cpp

A few ideas:

• Try to use a different option of "-g"
• Try split-dwarf
• Don't build this example

[fghoussen]

Run make, not make -jN. pyarpack is based on BOOST templates which is known to consume memory and slow down compilation: this is one of the dark sides you need to pay to get "template genericity". Note: I tried to compile this with an --enable-python option in autotools configure : more than 1h of compilation on my laptop (a few minutes with cmake).

[badshah400]

Disabling that example + example MATRIX_MARKET also means we have to disable all the pyarpack tests. For example, builds work after the following patch:

Index: arpack-ng-3.8.0/CMakeLists.txt
===================================================================
--- arpack-ng-3.8.0.orig/CMakeLists.txt
+++ arpack-ng-3.8.0/CMakeLists.txt
@@ -586,75 +586,15 @@ if(ICB)
   target_link_libraries(icb_arpack_cpp arpack BLAS::BLAS LAPACK::LAPACK ${EXTRA_LDFLAGS})
   add_test(icb_arpack_cpp_tst ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/icb_arpack_cpp)
 
-  if (ICBEXMM)
-    set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/EXAMPLES/MATRIX_MARKET)
-
-    add_executable(arpackmm EXAMPLES/MATRIX_MARKET/arpackmm.cpp)
-    target_include_directories(arpackmm PUBLIC ${PROJECT_SOURCE_DIR}/ICB ${EIGEN3_INCLUDE_DIR}) # Get arpack.h + eigen
-    target_link_libraries(arpackmm arpack BLAS::BLAS LAPACK::LAPACK ${EXTRA_LDFLAGS})
-    configure_file(EXAMPLES/MATRIX_MARKET/As.mtx ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/As.mtx)
-    configure_file(EXAMPLES/MATRIX_MARKET/An.mtx ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/An.mtx)
-    configure_file(EXAMPLES/MATRIX_MARKET/Az.mtx ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/Az.mtx)
-    configure_file(EXAMPLES/MATRIX_MARKET/B.mtx  ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/B.mtx)
-    configure_file(EXAMPLES/MATRIX_MARKET/Bz.mtx ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/Bz.mtx)
-    configure_file(EXAMPLES/MATRIX_MARKET/arpackmm.sh ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/arpackmm.sh)
-    add_test(NAME arpackmm_tst WORKING_DIRECTORY ${CMAKE_RUNTIME_OUTPUT_DIRECTORY} COMMAND ${BASH_PROGRAM} arpackmm.sh)
-  endif()
-
   if (PYTHON3)
-    python_add_module(pyarpack ${arpackutil_STAT_SRCS} ${arpacksrc_STAT_SRCS} ${arpacksrc_ICB} ${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpack.cpp)
+    python_add_module(pyarpack ${arpackutil_STAT_SRCS} ${arpacksrc_STAT_SRCS} ${arpacksrc_ICB})
     target_compile_definitions(pyarpack PRIVATE PY_MAJOR_VERSION="3")
-    set(pyarpack_HDR ${PROJECT_SOURCE_DIR}/ICB ${PROJECT_SOURCE_DIR}/EXAMPLES/MATRIX_MARKET ${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK)
+    set(pyarpack_HDR ${PROJECT_SOURCE_DIR}/ICB)
     target_include_directories(pyarpack PUBLIC ${pyarpack_HDR} ${EIGEN3_INCLUDE_DIR} ${Boost_INCLUDE_DIRS} ${PYTHON_INCLUDE_DIRS})
     target_link_libraries(pyarpack BLAS::BLAS LAPACK::LAPACK ${Boost_LIBRARIES} ${PYTHON_LIBRARIES})
     install(TARGETS pyarpack
             ARCHIVE DESTINATION ${CMAKE_INSTALL_LIBDIR}/python${PYTHON_VERSION_MAJOR}.${PYTHON_VERSION_MINOR}/site-packages
             LIBRARY DESTINATION ${CMAKE_INSTALL_LIBDIR}/python${PYTHON_VERSION_MAJOR}.${PYTHON_VERSION_MINOR}/site-packages)
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseBiCGDiag.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseBiCGDiag.py" @ONLY)
-    add_test(NAME pyarpackSparseBiCGDiag_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseBiCGDiag.py)
-    set_tests_properties(pyarpackSparseBiCGDiag_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseBiCGILU.py.in"  "${CMAKE_BINARY_DIR}/pyarpackSparseBiCGILU.py" @ONLY)
-    add_test(NAME pyarpackSparseBiCGILU_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseBiCGILU.py)
-    set_tests_properties(pyarpackSparseBiCGILU_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseCGDiag.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseCGDiag.py" @ONLY)
-    add_test(NAME pyarpackSparseCGDiag_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseCGDiag.py)
-    set_tests_properties(pyarpackSparseCGDiag_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseCGILU.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseCGILU.py" @ONLY)
-    add_test(NAME pyarpackSparseCGILU_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseCGILU.py)
-    set_tests_properties(pyarpackSparseCGILU_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseLLT.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseLLT.py" @ONLY)
-    add_test(NAME pyarpackSparseLLT_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseLLT.py)
-    set_tests_properties(pyarpackSparseLLT_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseLDLT.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseLDLT.py" @ONLY)
-    add_test(NAME pyarpackSparseLDLT_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseLDLT.py)
-    set_tests_properties(pyarpackSparseLDLT_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseLU.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseLU.py" @ONLY)
-    add_test(NAME pyarpackSparseLU_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseLU.py)
-    set_tests_properties(pyarpackSparseLU_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackSparseQR.py.in" "${CMAKE_BINARY_DIR}/pyarpackSparseQR.py" @ONLY)
-    add_test(NAME pyarpackSparseQR_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackSparseQR.py)
-    set_tests_properties(pyarpackSparseQR_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackDenseLLT.py.in" "${CMAKE_BINARY_DIR}/pyarpackDenseLLT.py" @ONLY)
-    add_test(NAME pyarpackDenseLLT_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackDenseLLT.py)
-    set_tests_properties(pyarpackDenseLLT_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackDenseLDLT.py.in" "${CMAKE_BINARY_DIR}/pyarpackDenseLDLT.py" @ONLY)
-    add_test(NAME pyarpackDenseLDLT_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackDenseLDLT.py)
-    set_tests_properties(pyarpackDenseLDLT_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackDenseLURR.py.in" "${CMAKE_BINARY_DIR}/pyarpackDenseLURR.py" @ONLY)
-    add_test(NAME pyarpackDenseLURR_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackDenseLURR.py)
-    set_tests_properties(pyarpackDenseLURR_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackDenseQRRR.py.in" "${CMAKE_BINARY_DIR}/pyarpackDenseQRRR.py" @ONLY)
-    add_test(NAME pyarpackDenseQRRR_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackDenseQRRR.py)
-    set_tests_properties(pyarpackDenseQRRR_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackDenseLUPP.py.in" "${CMAKE_BINARY_DIR}/pyarpackDenseLUPP.py" @ONLY)
-    add_test(NAME pyarpackDenseLUPP_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackDenseLUPP.py)
-    set_tests_properties(pyarpackDenseLUPP_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackDenseQRPP.py.in" "${CMAKE_BINARY_DIR}/pyarpackDenseQRPP.py" @ONLY)
-    add_test(NAME pyarpackDenseQRPP_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackDenseQRPP.py)
-    set_tests_properties(pyarpackDenseQRPP_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
-    configure_file("${PROJECT_SOURCE_DIR}/EXAMPLES/PYARPACK/pyarpackRestart.py.in" "${CMAKE_BINARY_DIR}/pyarpackRestart.py" @ONLY)
-    add_test(NAME pyarpackRestart_tst COMMAND ${PYTHON_EXECUTABLE} pyarpackRestart.py)
-    set_tests_properties(pyarpackRestart_tst PROPERTIES ENVIRONMENT PYTHONPATH=${CMAKE_BINARY_DIR}/lib:$ENV{PYTHONPATH})
   endif()
 
   if (MPI)

[badshah400]

Run make, not make -jN. pyarpack is based on BOOST templates which is known to consume memory and slow down compilation: this is one of the dark sides you need to pay to get "template genericity". Note: I tried to compile this with an --enable-python option in autotools configure : more than 1h of compilation on my laptop (a few minutes with cmake).

I tried using make -j1. Still no luck on i586 with pyarpack enabled.

[fghoussen]

Disabling that example + example MATRIX_MARKET also means we have to disable all the pyarpack tests.

You need to disable both ICBEXMM and PYTHON3 : PYTHON3 needs (and set) ICBEXMM. Run cmake .. instead of cmake -DICBEXMM=ON [and/or] -DPYTHON3=ON ..

[badshah400]

So, on i586 this means our only option is to disable the python bindings entirely, correct?

[sylvestre]

did you try https://github.com/opencollab/arpack-ng/issues/289#issuecomment-745891320 ?

You could also disable the binding only for this arch.

[badshah400]

did you try #289 (comment) ?

Haven't yet tried the split-dwarf. A different -g level didn't help and disabling the pyarpack example alone didn't either (only if I disabled also the MATRIX_MARKET did I get anywhere).

You could also disable the binding only for this arch.

Yes, might end up doing that. Thanks for the responses.

[fghoussen]

So, on i586 this means our only option is to disable the python bindings entirely, correct?

Yes, sounds unfortunatelly possible... Not even sure explicit template instanciation may help here.

[dimpase]

How do I even build/install pyarpack Python module ? For some reason, its sources are in EXAMPLES/PYARPACK/. With autotools, there is not even an option to build it. With cmake, apparently one needs to turn PYTHON3 and EXAMPLES (sic!) on.

I don't know anything about cmake support for building Python modules, but normally speaking, Python module source should come with e.g. setup.py file. Here it's just a bunch of C++ and "normal" Python files.

[sylvestre]

Probably legacy. don't hesitate to propose patches to improve the situation

[fghoussen]

With autotools, there is not even an option to build it.

pyarpack is not provided by autotools because of this: https://github.com/opencollab/arpack-ng/pull/254 @dimpase: if you have a solution to fix #254, feel free to reopen it (I guess there's no solution: libtool is super crazy slow)

How do I even build/install pyarpack Python module ?

https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L120

[fghoussen]

How do I even build/install pyarpack Python module ?

@dimpase: There is no setup.py to run since this python module is created with boost-python. boost-python creates libpyarpack.so AFAIR, then, make install will install this so in /path/to/prefix/lib/... Then you need to add in your PYTHONPATH the path where the so is. And now finally, you can import pyarpack from a python prompt.

[dimpase]

OK, but what does it do with .py files? Or there are no .py parts in the pyarpack package?

[fghoussen]

pyarpack is just the libpyarpack.so C++ library created by boost-python. This so is installed by make install.

EXAMPLES\PYARPACK have .py.in files that are meant to generate .py files: these .py are just regular python files used to test pyarpack running python script.

@dimpase: feel free to improve the pyarpack section in the README.md. I probably missed to explain how to install and run pyarpack. If you are testing this and checkout out how it works, would be good to improve the doc.

[dimpase]

Also, I can't see how to specify what BLAS/LAPACK to use. E.g. on macOS one might want to use OpenBLAS rather than Apple's BLAS.

[fghoussen]

Check out installation and usage sections in EXAMPLES/PYARPACK/README: I explained how to install and run pyarpack there.

The BLAS/LAPACK used are the ones eigen uses. ISO_C_BINDING (ICB) enables to call fortran from C/C++. From here arpackSolver.hpp uses eigen (ICBEXMM) to provide a ready to use C++ arpack solver. From there pyarpack (PYTHON3) is just a binding on top of arpackSolver.hpp which exposes C++ classes in python using boost-python.

[fghoussen]

on macOS one might want

Never had a mac! But seems that compiler options FFLAGS="-ff2c -fno-second-underscore" are mandatory to build: https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L186

[dimpase]

on macOS one might want

Never had a mac! But seems that compiler options FFLAGS="-ff2c -fno-second-underscore" are mandatory to build:

https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L186

Maybe - note that https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L191 actually requests the use of Apple's blas (a.k.a. `Accelerate Framework)

One reason I ask is that most macOS pyarpack tests fail for me (with Apple's blas). So one should test with OpenBlas (provided by Homebrew) as something more sane.

[fghoussen]

Try to build eigen (which is to be found by cmake) with the same blas/lapack (accelerate or openblas) that arpack founds/uses.

[dimpase]

Eigen does not use BLAS/LAPACK, so I'm lost here.

[fghoussen]

On mac (never had one) without -DMPI (keep it simple):

1. without -DPYTHON3, but with Apple's blas (Accelerate Framework) and FFLAGS="-ff2c -fno-second-underscore", does this work (test suite OK)? https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L191

2. Now replacing only Apple's blas with openblas, does this work (test suite OK)?

3. Now build boost-python from source for python3 and install it: https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L121 with python3 https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/.github/workflows/jobs.yml#L124

4. Now getting back to step 2, add -DPYTHON3 and check out in cmake logs that cmake found the boost-python compiled / installed in step 3, does this work (test suite OK)?

5. Now if brew provides some kind of precompiled boost-python3 package, you may try to use it too (check cmake logs).

I wrote pyarpack a long time ago: I have a doubt now if Eigen indeed needs blas/lapack in this case (I'll checkout this out when I can). Eigen is needed/used for solving the Reverse Communication Interface pattern: https://github.com/opencollab/arpack-ng/blob/5eafea4328f1631eab28b1a20e757d1f0e21f8a6/EXAMPLES/MATRIX_MARKET/arpackSolver.hpp#L774 I was thinking that Eigen would needs blas/lapack in back-end for that. Maybe not?

[dimpase]

what is eigen you mentioned? Is it https://eigen.tuxfamily.org/ ?

[fghoussen]

yes, it's a C++ template library meant for matrix-vector operations.

[fghoussen]

@dimpase: if you can test ``pyarpack on macos and report about it, would be great!

[fghoussen]

@dimpase: seems to work on mac #365. Did you brew install python3 and pip3 install numpy?