ENH: Implement ndarray.__format__ for 0d arrays - #9883

Blank line here.

Above this line, I assume you mean?

Yep.

Is assert_equal the best way to compare strings? Maybe just assert_(str1 == str2) would be a bit cleaner.

Yes, because then it shows the two values if they differ.

OK.

In pytest you can just assert str1 == str2 and you will get values if there is a failure.

def test_failure():
    a = 'apple'
    b = 'banana'
    assert a == b

Output is

================================== FAILURES ===================================
________________________________ test_failure _________________________________

    def test_failure():
        a = 'apple'
        b = 'banana'
>       assert a == b
E       AssertionError: assert 'apple' == 'banana'
E         - apple
E         + banana

test_util.py:20: AssertionError
========================== 1 failed in 0.24 seconds ===========================

For consistency, we should probably use the same form we use in most of the test suite, which gives:

>>> np.testing.assert_equal('a', 'b')
AssertionError: 
Items are not equal:
 ACTUAL: 'a'
 DESIRED: 'b'

AFAIK, the switch to pytest was recent, and we were previously using nose

We are still using nose, the pytest conversion is still in progress and I don't think I will get back to it until 1.15.

What does the builtin do here?

It keeps the arbitrary and python-dependent behavior that we had before this patch.

Most importantly it handles '{}'.format in the typical way. On python 3, it throws an exception if a format string is given, while on python 2 it evaluates format(str(arr), 'theformatspec').

I suppose we should silence warnings in ret_and_exc, or return a third tuple item which is the warning

Maybe just figure out what the result should be and use that? Eventually object.__format__ is going to fail with non-empty argument.

Right, and I think that ndarray.__format__ should start failing at the same time. Maybe we should just make ndarray.__format__ fail with an empty argument right now...

Looks like it went away in Python 3.4.

* :meth:`object.__format__` no longer accepts non-empty format strings, it now
  raises a :exc:`TypeError` instead.  Using a non-empty string has been
  deprecated since Python 3.2.  This change has been made to prevent a
  situation where previously working (but incorrect) code would start failing
  if an object gained a __format__ method, which means that your code may now
  raise a :exc:`TypeError` if you are using an ``'s'`` format code with objects
  that do not have a __format__ method that handles it.  See :issue:`7994` for
  background.

Should ours work if an empty string is passed?

I fixed this by making the test Python version dependent.