FIX: to_sql takes the boolean column as text column

[yelite]

In the original code, com.is_bool(arr_or_dtype) checks whether arr_or_dtype is a boolean value instead of a boolean dtype.

A new function is_bool_dtype is added to pandas.core.common to fix this bug.

[jorisvandenbossche]

@yelite Thanks for the contribution!

Can you add a test for it (look in io/tests/test_sql.py)? You can for example take the example you were working with where you experienced the issue, and slim this down to a small test case. Just ask if something is not clear for this.

@jreback There isn't already a function like this somewhere?

[jreback]

this is already in master (recently added)
core/common iirc

[jorisvandenbossche]

@yelite indeed, it is already here: https://github.com/pydata/pandas/blob/master/pandas/core/common.py#L2255

[yelite]

@jorisvandenbossche
Great, I removed my function from the commit and added a simple test.

[jorisvandenbossche]

@yelite I wanted to say your test would fail because of the index, but you already adapated it!

Can you add a line in doc/source/v0.14.1.txt in the bug fixes section for this?

[jreback]

this should be tm.assert_frame_equal(self.test_frame2,result)

[yelite]

@jreback

AssertionError: attr is not equal [inferred_type]: 'string' != 'unicode'

https://travis-ci.org/pydata/pandas/jobs/29317980

Does this difference between string and unicode really matter? Should we omit it?

[jreback]

so the returned data is unicode. so 2 solns:

• on the test calls tests with (and w/out) the server returning unicode (IIRC they usually and by default do, but can be turned on/off

e.g. in sqlalchemy (sort of manuall), I usually do

engine.dialect.description_encoding=None (with postgresql),

• add an option to assert_frame_equal to ignore this different (better soln I think)

@jorisvandenbossche ?

[yelite]

@jreback
A mistake. I thought check_column_type means checking the type of each column. Actually it checks the type of column block. So it's not important in this test. I set it to False now.

@jorisvandenbossche
I added a line at the end of doc/source/v0.14.1.txt. Is that ok or it should be organized by section?

[jorisvandenbossche]

For the whatsnew note, that is OK.

And the column names of the original dataframe and the result from read_sql are indeed different in type ... (but indeed, not important to test that here).

[jorisvandenbossche]

Travis is green, so merging.
Thanks a lot! (and if you want to tackle other issues, don't hesitate :-))

[jreback]

https://travis-ci.org/pydata/pandas/jobs/29341707

I have seen this happen if something weird is passed to an is_***_dtype (np.dtype craps out).

let's see if it repeats

[jorisvandenbossche]

hmm, the next travis builds have the same (but strange that for the PR it had not this issue)