mock_open(read_data=b'...') gives an error:

"""Traceback (most recent call last):
  File "z.py", line 6, in <module>
    print(f.read())
  File "/usr/local/lib/python3.4/unittest/mock.py", line 896, in __call__
    return _mock_self._mock_call(*args, **kwargs)
  File "/usr/local/lib/python3.4/unittest/mock.py", line 962, in _mock_call
    ret_val = effect(*args, **kwargs)
  File "/usr/local/lib/python3.4/unittest/mock.py", line 2279, in _read_side_effect
    return ''.join(_data)
  File "/usr/local/lib/python3.4/unittest/mock.py", line 2244, in _iterate_read_data
    data_as_list = ['{}\n'.format(l) for l in read_data.split('\n')]
"""

Easy to reproduce:

"""
from unittest.mock import mock_open, patch

m = mock_open(read_data= b'abc')
with patch('__main__.open', m, create=True) :
    with open('abc', 'rb') as f :
        print(f.read())
"""

Looks like this bug was introduced as result of issue #17467. I add those people to the nosy list.

I've created a patch that fixes this, and added an accompanying unit test (which fails without the change).

Thanks for the patch Aaron. Unfortunately this doesn't quite fix the issue. There are two problems with the patch:

If a bytes object is passed into mock_open, I'd expect a bytes object in the output. In your patch, not only is this not the case (the output is a string object), but the bytes object is being coerced into its string representation in the resulting list. For example (simplified from your patch):

>>> data = b'foo\nbar'
>>> newline = b'\n'
>>>
>>> ['{}\n'.format(l) for l in data.split(newline)]
["b'foo'\n", "b'bar'\n"]

What I would expect to see in this case is:

[b'foo\n', b'bar\n']

> There are two problems with the patch

That was intended to be removed after I changed the wording :P

I've created a new patch, which addresses the problem. Your example now currently returns [b'foo\n', b'bar\n']

Thanks for the update, but this doesn't quite work either as you're assuming utf-8 (which is what .encode() and .decode() default to). For example, when using latin-1:

>>> m = mock_open(read_data= b'\xc6')
>>> with patch('__main__.open', m, create=True) :
...     with open('abc', 'rb') as f :
...         print(f.read())
...
Traceback (most recent call last):
  [snip]
UnicodeDecodeError: 'utf-8' codec can't decode byte 0xc6 in position 0: unexpected end of data

Additionally, a bytes object may simply be binary data that doesn't adhere to any specific encoding.

My suggestion is to remove the use of format() altogether as it's really not doing anything complex and simply append either '\n' or b'\n' depending on the type of object passed in. That way, you can deal with the type of object passed in directly without coercion.

Thanks, I've fixed that. Not sure why I thought decoding and re-encoding would work with any binary data.

I've also updated one of the tests to use non-utf8-decodeable binary data, to prevent a future regression.

Thanks again for the update Aaron, I've left a couple small comments in Rietveld. Other than those, the patch looks good to me. Thanks for the contribution!

I've fixed the issues you pointed out.

Is there a better way than uploading a new patch file to make changes?

Thanks for the updated patch, looks good to me. If you haven't already read it, the patch workflow is here: https://docs.python.org/devguide/patch.html and is the only workflow currently available.

I've fixed the formatting issues.

A few more comments were left in Rietveld for you, likely hidden by spam filters.

LGTM.

New changeset 3d7adf5b3fb3 by Berker Peksag in branch '3.4':
Issue #23004: mock_open() now reads binary data correctly when the type of read_data is bytes.
https://hg.python.org/cpython/rev/3d7adf5b3fb3

New changeset 526a186de32d by Berker Peksag in branch '3.5':
Issue #23004: mock_open() now reads binary data correctly when the type of read_data is bytes.
https://hg.python.org/cpython/rev/526a186de32d

New changeset ed15f399a292 by Berker Peksag in branch 'default':
Issue #23004: mock_open() now reads binary data correctly when the type of read_data is bytes.
https://hg.python.org/cpython/rev/ed15f399a292

Thanks for the patch, Aaron(also thanks to Demian for reviews). I've fixed the merge conflict and added more tests.

Post merge review:

looks like

data_as_list = read_data.splitlines(True)

would be a little cleaner.

data_as_list = read_data.splitlines(True)

is not actually the equivalent of

    data_as_list = [l + sep for l in read_data.split(sep)]

It will change the behavior of the _iterate_read_data helper. See the comment at https://github.com/python/cpython/blob/78d05eb847c6b8fede08ca74bb59210c00e4c599/Lib/unittest/mock.py#L2278

However, in default branch, we can simplify it to

    yield from read_data.splitlines(keepends=True)

splitlines(keepends=True) is not ever equivalent to splitting by just '\n'.  I don't know the details here, but switching to that would certainly be a behavior change.  (Especially if the code path also applies to non-binary data!):

>>> b'abc\nde\r\nf\x1dg'.splitlines(True)
[b'abc\n', b'de\r\n', b'f\x1dg']
>>> 'abc\nde\r\nf\x1dg'.splitlines(True)
['abc\n', 'de\r\n', 'f\x1d', 'g']