BUG: Find dtype in genfromtxt for single column of data

[millen1m]

BUG: Find dtype in genfromtxt for single column of data

When single column of data is given to genfromtxt and names=True,
StringConverter was not given the right dtype.
The bug caused version 1.19 to fail to load single column text files.
However, a fix for the failure to load has already been provided in
the master branch, so that the dtype is found.
The fix proposed here is consistent with the 2D behaviour.

[tylerjereddy]

Without commenting on the nature of the regression, this passed for me locally with this minor adjustment:

         ctrl = np.array([(5,), (6,)],
-                        dtype=[('a_header', int)])
+                        dtype=[('a_header', float)])
         assert_equal(mtest, ctrl)

[millen1m]

Yes, sorry the correction you have applied is needed. Actually the test passes without the change that I have made. But at line 1971
converters = [StringConverter(dtype, locked=True, parameter dtype has the header name in it and therefore the StringConverter does not find the right type and the converter[0].type is void, whereas it should be float. This only happens when you have a single column. While for multiple columns of data the dtype_flat parameter is passed in, which has the header names removed.
The correction I have provided makes it consistent with the multiple column approach and means the data type is found within the StringConverter object.

[rossbar]

I think this test is incorrect: note that the default value for dtype is float, so you should expect a float type here.

The dtype-inference-machinery is triggered with dtype=None, which gives the expected result:

>>> np.genfromtxt(
...     StringIO(data), skip_header=1, delimiter=",", names=True, usecols=0, dtype=None
... )
array([(5,), (6,)], dtype=[('a_header', '<i8')])

AFAICT the regression has been fixed and the current behavior is correct --- I think this can be closed (unless I'm missing something @millen1m )

[millen1m]

Yes - the regression has been fixed in another way - can close. Cheers.