Using internal allocatable arrays with f2py

Question

I am using f2py to wrap some Fortran subroutines. When calling from Python, I always use explicit arrays for the input/output, but have a few working arrays that need to be allocatable. These arrays have no interface with Python. A small test subroutine that replicates the problem is shown below:

subroutine test(n, x, y)

    !f2py intent(in) n, x
    !f2py intent(out) y
    !f2py depend(n) x

    integer, intent(in) :: n, x(n)
    integer, intent(out) :: y

    integer, allocatable :: z(:)

    allocate(z(n))

    z(:) = 10

    y = z(2) + x(1)

end subroutine test

Whenever I use internal allocatable arrays I get the following output when doing "python setup.py develop" with NumPy's distutils package:

test.o : error LNK2019: unresolved external symbol _gfortran_runtime_error referenced in function test_
test.o : error LNK2019: unresolved external symbol _gfortran_runtime_error_at referenced in function test_
test.o : error LNK2019: unresolved external symbol _gfortran_os_error referenced in function test_
ASAP\lib\asap_lib.pyd : fatal error LNK1120: 3 unresolved externals

I am using NumPy's Configuration class from numpy.distutils.misc_utl and gfortran 5.2.0, mingw64, Windows 10.

The program will compile just fine if I use gfortran directly, but when it gets hooked up with f2py (or somewhere in that process) things fail. I can comment out the allocate line and it will compile, but obviously not work at runtime. There are always 3 unresolved externals no matter how many allocatable arrays I have.

While I'm here, I get similar link errors if the Fortran subroutine has a write or print statement in it. I don't need them, would just be useful for debugging if anyone has an idea.

Thanks for the help.

UPDATE 1:

Using the Configuration class from numpy.distutils.misc_util I added an extra link option

config.add_extension('test_lib', sources, extra_link_args=['/FORCE']

after looking at the info here at Microsoft Linker Options. The '/FORCE' is the key option. The print or write command still throw back exit code 255 in Python, but at least the allocatable arrays seem to be working, which is all I cared about.

Seems like a hack, but if anybody knows what is causing the original problem I probably still have a deeper rooted issue that I need to address.

A wild guess: In your example, does the problem persist if you explicitly deallocate(z) before exiting the subroutine? — ev-br
– ev-br, Commented Feb 21, 2016 at 11:29
Somehow you are not linking the gfortran runtime library. What happens when you have no allocatable arras, but a print statement or something similar in the code? Try to make a simple dll with simple f2py -c -m module_name. — Vladimir F Героям слава
– Vladimir F Героям слава, Commented Feb 21, 2016 at 11:40
@VladimirF Thanks for the suggestion. A print statement throws the same error, and using f2py -c -m test test.f90 gives the same error also. I should add I have installed the VC++ Compiler for Python 2.7 and using 64-bit Anaconda Python. One thing I had to do was manually point to vcvarsal.bat in the C:\Anaconda\disutils\mscv9compiler.py file (in case that gives you any clues). I've noticed other packages I've installed with f2py that use matmul give similar errors, so I think you've correctly pointed out it's not linking to the runtime lib. — trelau
– trelau, Commented Feb 21, 2016 at 15:47

trelau · Accepted Answer · 2016-04-19 16:03:38Z

1

I had a link problem and specifying compiler=mingw32 in the f2py options fixed all the problems.

answered Apr 19, 2016 at 16:03

trelau

1351 silver badge6 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Using internal allocatable arrays with f2py

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related