Avoid conversion to int_ rhs argument of enum eq/ne

[sizmailov]

Fixes #1830

[wjakob]

This change looks reasonable, but shouldn't it also be implemented consistently for the other operations (e.g. less than operator for arithmetic types, etc.)?

[sizmailov]

I think the right way is to mock python behavior. Here is test snippet with int, string and custom types:

test snippet

class A(object): 
  def __repr__(self): return "A()"
class B(object): 
  def __repr__(self): return "B()"

def test_binop(op, f):
  try:
    result = f()
    print("`{} {} {}` = `{}`({})".format(repr(a),op,repr(b),result,type(result)))
  except Exception as e:
    print("`{} {} {}` raises {}: {}".format(repr(a),op,repr(b),type(e), e))

def test_unop(op, f):
  try:
    result = f()
    print("`{} {}` = `{}`({})".format(op,repr(a),result,type(result)))
  except Exception as e:
    print("`{} {}` raises {}: {}".format(op,repr(a),type(e), e))

for a,b in [(A(), B()),
            (1, 2), 
            (1, A()),
            (A(), 1),
            ]:
  test_binop("!=", lambda: a!=b)
  test_binop("==", lambda: a==b)
  test_binop("<", lambda: a<b)
  test_binop("<=", lambda: a<=b)
  test_binop(">", lambda: a>b)
  test_binop("&", lambda: a&b)
  test_binop("|", lambda: a|b)
  test_binop("^", lambda: a^b)

for a in [A(), 1, "string"]:
 test_unop("~", lambda: ~a)

snippet on repl.it: (python2) (python3)

A() < 1 is False in python2 and TypeError in python3.

Current behavior is in line with python3. I've updated tests to check behavior of enum arithmetic and behavior on operations with unsupported types.

Should we branch on python version or leave it as is?

While current behavior is in line with pytnon3, the exception message is a bit off-putting:

>       y ^ object()
E       TypeError: int() argument must be a string, a bytes-like object or a number, not 'object'

Additional check will duplicate checks inside of py::int_, so it's a bit of redundant work on non-exceptional path. extra try-catch feels like heavy-weight, but it should not affect non-exceptional path. Should it be added?

[sizmailov]

For some reason enumeration elements order in docstring is not determined. Replaced exact docstring comparison with every-line-exists-in-docstring check.

[wjakob]

I don't think the extra check is needed, and defaulting to Python 3 behavior sounds good to me.