BUG: Boolean indexing on an empty series loses index names

[miketkelly]

#4235

[jreback]

can u put tests in to check the resulting dtype
this actually may be an API change

[miketkelly]

Added a check on dtype.

[hayd]

@jreback this looks reasonable to me. Think it's a bug fix rather than API change.

@mtkni Do you mind adding a Note in the release notes (docs/source/release.rst) mentioning 4235 (you'll probably have to rebase off current master first) ?

[jreback]

I think this might be fixed
as am pretty sure that in propogate names (but maybe not on empty)

[hayd]

The first test is already working in master, names propagate (however can't hurt to add test?).

This one isn't though:

        result = lib.infer_dtype(np.array([], dtype='i4'))
-        self.assertEqual(result, 'empty')
+        self.assertEqual(result, 'integer')

I get (in master):

In [10]: pd.lib.infer_dtype(np.array([], dtype='i4'))
Out[10]: 'empty'

maybe @mtkni has some thoughts.

[jreback]

actually the inference of 'empty' I don't think is actually used anywhere. I would ALWAYS return the type (and mixed) if you fall thru the loop, empty is definitly weird

[jreback]

if n == 0, then just return an inferred dtype (you have to map them, prob via the kind)

e.g.

if dtype.kind == 'i':
   return 'integer'
elif dtype.kind == 'f':
   return 'floating'
......

[hayd]

@jreback well apparently not, since moving just those 3 lines seems to do the trick. merge then?

[jreback]

@hayd ok then!

[hayd]

This passes on a rebase https://travis-ci.org/hayd/pandas/builds/10534501

@jreback Will wait on release note? Or I could pop it in to my branch.

[jreback]

@hayd whatever works for you!

[hayd]

ok, adding commit to my rebased branch and will close these.