Docstring generation tweaks

[sizmailov]

This PR contains a couple of trivial tweaks which make generated docstring a bit better.

this PR	master
a() -> numpy.ndarray[numpy.int32]	a() -> numpy.ndarray[int32]
b1(arg0: Iterator) -> None b2(arg0: Iterable) -> None	b1(arg0: iterator) -> None b2(arg0: iterable) -> None
c() -> numpy.ndarray	c() -> array
d() -> int	d() -> int_

compare.cpp

#include "pybind11/pybind11.h"
#include "pybind11/stl_bind.h"
#include "pybind11/embed.h"
#include "pybind11/numpy.h"

namespace py = pybind11;

PYBIND11_EMBEDDED_MODULE(example, m) {

  std::array<int, 3> data = {1, 2, 3};
  m.def("a", [&]{ return py::array_t({3, 1}, data.data()); });
  m.def("b1", [](py::iterator&) { });
  m.def("b2", [](py::iterable&) { });
  m.def("c", []{ return py::array(); });
  m.def("d", []{ return py::int_(0); });
}

int main() {
  py::scoped_interpreter guard{};
  py::exec(R"(
      from example import *
      print(a.__doc__, end="")
      print(b1.__doc__, end="")
      print(b2.__doc__, end="")
      print(c.__doc__, end="")
      print(d.__doc__, end="")
)");
}

[rwgk]

Hi Sergei, I'm helping out reviewing PRs, to take some load off Wenzel. Could you please split this PR into 3? E.g. commits 1-4 stay in this PR, 5, 6 each get their separate PR. That will allow me and Wenzel to act quicker on 1-4, and give us separate contexts to discuss the points you're raising. I'll look again asap after the split.

[sizmailov]

Hi, Ralf! Thanks for looking into that. I've split PR in three as you suggested.

[rwgk]

Looks great to me, just a minor question about tests for Iterable & Iterator.
I think inserting the numpy. is great for clarity.

[rwgk]

Thanks Sergei!

[sizmailov]

fixed silenced flake8 spacing complaints

(travis.org seems to become quite unstable recently, probably migration to .com is required)

[wjakob]

These all look good to me, thanks!