BUG: fix for f2py string scalars

[2sn]

proposed fix for #23192

in previous version, any string scalar was converted to a string array of dimension len, i.e., a definition

character(len=N) :: X

effectively became

character(len=NNN), dimension(NNN) :: X

from the point of few of the numpy (python) interface:

X.shape == (NNN,)
X.dtype == '|SNNN'

[seberg]

Thanks! @2sn could you please also add a very basic tests to numpy/f2py/tests/test_character.py that fails without the fix? (I think the file should be pretty straightfoward to expand hopefully.)

[2sn]

OK, let me have a look.

[2sn]

The tests that are present are unlike my usage, I defined in a module.

How do I execute a test?

[seberg]

Hmmm, I would suspect you can add it to the initial

code = ''

Inside the test class the module with that code is available as self.module. Maybe the 1, 3, star loop makes sense, in which case you include it there (I suspect that is the way to go).

I.e. insert the code here: https://github.com/numpy/numpy/blob/main/numpy/f2py/tests/test_character.py#L20 and add a test using self.module.name? Or can you not include the string definition from Fortran in the same .f90 files that gets generated there?

(Sorry, I am just trying to throw breadcrumbs that might help you, I don't have deep knowledge of f2py/Fortran to be sure this makes sense.)

[2sn]

I created a module test, but do I call (test) it?

[2sn]

well, with the old code my test of

MODULE string_test

  character(len=8) :: string

END MODULE string_test

gives

[f2py-string-scalar] ~/numpy>python -m pytest numpy/f2py/tests/test_character.py 
===================================================== test session starts ======================================================
platform linux -- Python 3.11.1, pytest-7.2.1, pluggy-1.0.0
rootdir: /home/alex/numpy, configfile: pytest.ini
plugins: anyio-3.6.2, hypothesis-6.68.0
collected 44 items                                                                                                             

numpy/f2py/tests/test_character.py ...................................s.......F                                          [100%]

=========================================================== FAILURES ===========================================================
__________________________________________________ TestStringScalar.test_char __________________________________________________

self = <numpy.f2py.tests.test_character.TestStringScalar object at 0x7f6455c0f3d0>

    @pytest.mark.slow
    def test_char(self):
        out = self.module.string_test.string
        expected = ()
>       assert out.shape == expected
E       assert (8,) == ()
E         Left contains one more item: 8
E         Use -v to get more diff

expected   = ()
out        = array([b'', b' O', b'\x13', b'', b'', b'\xd5\x1a\x00\x00\x05\x00\x01\x08',
       b'\x00\x00\x00\x00&\xe9\x03', b'\x00\x1d\x00\x00\x00\x00['],
      dtype='|S8')
self       = <numpy.f2py.tests.test_character.TestStringScalar object at 0x7f6455c0f3d0>

numpy/f2py/tests/test_character.py:581: AssertionError
=================================================== short test summary info ====================================================
FAILED numpy/f2py/tests/test_character.py::TestStringScalar::test_char - assert (8,) == ()
=========================================== 1 failed, 42 passed, 1 skipped in 7.25s ============================================

but runs fine with the fix

[f2py-string-scalar*] ~/numpy>python -m pytest numpy/f2py/tests/test_character.py 
===================================================== test session starts ======================================================
platform linux -- Python 3.11.1, pytest-7.2.1, pluggy-1.0.0
rootdir: /home/alex/numpy, configfile: pytest.ini
plugins: anyio-3.6.2, hypothesis-6.68.0
collected 44 items                                                                                                             

numpy/f2py/tests/test_character.py ...................................s........                                          [100%]

================================================ 43 passed, 1 skipped in 7.34s =================================================

[2sn]

I am not sure the failed checks (32bit azure pipeline) are due to my code, at least I can't see any details.

RuntimeWarning: invalid value encountered in matmul

[seberg]

I am not sure the failed checks (32bit azure pipeline) are due to my code, at least I can't see any details.

Don't worry about it. Some heisenbug may related to openblas, maybe not, it happens occasionally (and has been for a very long time).

[HaoZeke]

Also, thanks a ton for looking into fixing this @2sn

[2sn]

@HaoZeke The code needs to remain with rank='0' as coded. See Fortran code example above (gfortran 12.2.1-4).

As you write As per the Fortran standard, rank=0 is only for scalars, this is exactly what the fix does: setting rank=0 for scalars. dims needs to be '', consistent with the section below in the source code (later, in the calling routine this is converted to -1, which enters in the .c code generated.

if you set dims=1 and rank=1, as you suggest, then this would generate numpy array of shape=(1,) for a scalar, which is not what we want. A scalar needs to have shape=().

[HaoZeke]

Looks good to me.

[seberg]

Thanks a lot @2sn and @HaoZeke!

[2sn]

@seberg @HaoZeke thanks for the very quick turnaround.