Py3 tests2

[takluyver]

Fixes for a couple of tests under Python 3.

@ellisonbg, I'm having some trouble with a test in nbformat. With these changes, I get:

======================================================================
FAIL: test_roundtrip (IPython.nbformat.v2.tests.test_json.TestJSON)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/thomas/Code/virtualenvs/ipy-2to3/lib/python3.2/site-packages/ipython-0.12.dev-py3.2.egg/IPython/nbformat/v2/tests/test_json.py", line 18, in test_roundtrip
    self.assertEquals(reads(s),nb0)

followed by a long dump of the before and after dicts - I can't even spot the != in there, but I think the difference is between:

{'latex': '$a$', 'prompt_number': 3, 'svg': '<svg>', 'javascript': 'var i=0;', 'jpeg': 'ZGF0YQ==\n', 'json': 'json data', 'html': 'The HTML rep', 'text': '<array a>', 'output_type': 'pyout', 'png': 'ZGF0YQ==\n'}

and (N.B. b'data'):

{'latex': '$a$', 'prompt_number': 3, 'svg': '<svg>', 'javascript': 'var i=0;', 'jpeg': b'data', 'json': 'json data', 'html': 'The HTML rep', 'text': '<array a>', 'output_type': 'pyout', 'png': b'data'}

Any immediate thoughts?

[minrk]

There are definitely some bugs in the string/unicode/bytes logic in nbformat. In the JSON roundtrip, the problem is that the inverse of the b64 encode is never called. This doesn't come up, because the bytes are already b64-encoded in actual practice, and the bytes-logic in the JSONEncoder is actually never called in Python2 (as far as I can tell).

I think the real bug is that the base64_encode/decode functions in rwbase encode cell.png / cell.jpeg, but as far as I can tell, those attributes never exist. png/jpegs are attached to outputs of a cell, not the cells themselves.

This doesn't come up in Python2, because that output appears to always already be b64-encoded bytes, which pass right through to/from JSON without issue.

[takluyver]

Is there anything simple that can be done to fix it? I'm not at all familiar with this part of the code.

[minrk]

Further detail: every bytes object in the notebook is already a b64-encoded bytes object. So, instead of performing b64 encoding/decoding around JSON, it should be an ascii encoding/decoding, leaving the contents unchanged.

[minrk]

Yes, I've got it working locally - the answer is to stop using b64 to/from JSON, as it has already been encoded. I'll PR to your branch shortly with fixes after I clean up all my code.

[minrk]

PR issued

[takluyver]

Thanks Min - I've added that to this PR, and the nbformat tests are passing now.

[minrk]

Shouldn't this be 'if not hasattr', rather than try/except? We don't need to do it if it already exists.

[takluyver]

Good point. Done.

[takluyver]

I've also made a couple of decorators to mark known failures, so it's easier to spot regressions. The parallel package is still failing - I'll try to look into it if I get time, but I think it's fairly low priority.

[fperez]

This is looking pretty solid, so I'll merge it now to avoid bitrot. When you get a chance to tackle the parallel ones, a new PR can come in for those. Thanks for this work! Looking better and better on py3 :)