* Fix new deprecation warnings
  - Change to DeprecationWarning to reduce noise in other testers
  - Set stack level so warning appears with proper source line
  - Check for warning in test

* Explicitly verify warning content

  - Bonus:  works with all testrunners.

  -  We already pass these, but some other unparsers don't,
     so it's worth adding them to the regression.

Add support for __future__ unicode_literals

  -- This only affects a very few modules that did not roundtrip properly

Deprecate old API (issue #59)

Dual docs -- issue #63

  - Plus add more verbose changelog format as suggested by Berker

Add support for ASTs compiled with single or eval options

- Clean up code providing new functionality
- (Thanks  for the initial functionality, Ryan) 👍 
- run extended test cases via astor.rtrip
- add regression tests for new features based on rtrip failures
- fix failing regressions

Otherwise, pretty printer will think it can break them up, and no joy
will come from that.

Add support for variable annotations

Include tests in release source tarball

A test suite is a very useful thing, it would be great to include it in the pypi release tarball. That way people who download and build the release can run the tests to ensure the library is working.

Linux distributions, such as Debian, base their packages of Python modules on the pypi release. This means the test suite can be run when building a Debian package, it should catch mistakes in the packaging or errors in dependencies.

I've written some more on this topic on the Debian Python mailing list.

https://lists.debian.org/debian-python/2016/04/msg00074.html

Add test for elif

precedence for `await` and `async for`

Lib/test/test_coroutines.py in the Python 3.5 codebase was observed to
not successfully roundtrip under astor.rtrip due to an await expression
of an await expression (the offending coroutine function has been
excerpted in the `test_double_await` test method introduced in this
commit). The `await` operator has very high precedence, greater than
that of exponentiation, as documented at
https://docs.python.org/3.5/reference/expressions.html#operator-precedence.

Without this, the call to set_precedence in visit_For calls
get_op_precedence, which raises a KeyError.

Add information about other codegen derivatives