Add Type to bad args output

[jagerman]

The changes the error message for invalid arguments to display Type (= strvalue) instead of just strvalue for the given argument list.

Fixes #517.

[dean0x7d]

What about using py::repr? It would show type info in non-trivial cases, but leave simple things simple and avoid type duplication like NoneType (= None) and Pet (= <m.Pet object at 0>).

[jagerman]

@dean0x7d - you mean simply instead of the type + value pair? I thought it might be nice to have both type and value, even if it is redundant sometimes (like None, Pet).

[jagerman]

But yeah, perhaps just repr instead of the value would be better.

[dean0x7d]

Yeah, just repr where it was str originally. Seems like exactly what repr was intended for.

[jagerman]

While it's nicer than str, I'm on the fence as to whether it's enough. The only practical change it makes in the test scripts is adding quotes around the arguments of 'no', 'such', 'constructor'. On the other hand, the list of overloads directly lists the types, so including the type name as well might be more informative (e.g. including both type.__name__ and repr(value)).

[jagerman]

@pschella - thoughts?

[dean0x7d]

In most cases repr already contains the type:

>>> import datetime
>>> now = datetime.datetime.now()
>>> str(now)
2016-11-21 21:32:47.664461
>>> repr(now)
datetime.datetime(2016, 11, 21, 21, 32, 47, 664461)

repr is quite clear about the type and as a nice bonus we can use eval:

>>> then = eval(repr(now))
>>> str(then)
2016-11-21 21:32:47.664461

The examples in the tests are mostly builtin types which can be trivially reconstructed. That's why str(1) == repr(1) but for strings the quotes are added:

>>> s = "hello"
>>> str(s)
hello
>>> repr(s)
'hello'
>>> eval(str(s))
NameError: name 'hello' is not defined
>>> eval(repr(s))
hello

[jagerman]

Okay, I'm convinced :)

Updated to just switch to repr (I also ditched the static_cast because it seems unnecessary--I think not required since #450).