pybind: fix unicode handling in CephFSVolumeClient::purge

[jcsp]

os.path.join is sensitive to string encoding, but
just doing a straight substitution should not be.

Signed-off-by: John Spray john.spray@redhat.com

[jcsp]

Tested by ceph/ceph-qa-suite#934

[jcsp]

Passing test run here (http://pulpito.ceph.com/jspray-2016-04-30_17:56:02-fs-wip-jcsp-testing---basic-mira/) minus the valgrind failure from #8747

[jdurgin]

Looks like that run didn't include test_purge (not using the wip-15266 suite branch).

Maybe I'm missing something, but I'm not seeing how using format() handles unicode better than os.path.join - what are the types of args that fail with the latter and succeed with the former?

[onyb]

Python 3 documentation suggests the following for os.path module:

Applications are encouraged to represent file names as (Unicode) character strings.

In this particular case, one possible fix in PY2 could be: d_full = os.path.join(root_path, unicode(d.d_name)). Note that d_full in the above expression would also be a Unicode literal.

I hope things will be simpler after we have a compatibility layer (for PY2 and PY3 dual support), during Google Summer of Code.

[jcsp]

@jdurgin yeah, this was a weird one. I think Manila is giving us share IDs (which are very much ascii) as unicode python objects.

>>> os.path.join(u"/foo", chr(0xc3))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib64/python2.7/posixpath.py", line 73, in join
    path += '/' + b
UnicodeDecodeError: 'ascii' codec can't decode byte 0xc3 in position 1: ordinal not in range(128)
>>> "{0}/{1}".format(u"/", chr(0xc3))
'//\xc3'

My reproducer PR wasn't actually using unicode properly so I've updated that too (can see it failing if you back out the join vs. format change)

[jdurgin]

Ah, ran into similar passing of unicode from glance, cinder, and nova - since we didn't have a high-level library like this we ended up forcing everything to utf-8-encoded string types before passing to librbd. Of course, nothing actually used non-ascii chars for e.g. uuids, so it was just a nuisance, but allowed it to work with existing versions of the python bindings.

[jcsp]

Passing test run here: http://pulpito.ceph.com/jspray-2016-05-12_04:34:36-fs-wip-jcsp-testing-20160511---basic-smithi/187784

Any objections to merging?

[jdurgin]

lgtm fwiw