Document extra dependencies

[aaronayres35]

fixes #126

This PR adds extras_require to setup.py and also updates the readme to mention the optional dependencies.

The issue originally mentioned tornado but I am not sure this is still really needed? tornado is listed as a required dependency of ipykernel and the only occurrences of "tornado" in envisage are:

envisage/envisage/plugins/ipython_kernel/kernelapp.py

Lines 76 to 94 in 8eddfc4

	def configure_tornado_logger(self):
	"""
	Configure tornado logging.
	Adds a NullHandler to the tornado root logger, if there are
	no handlers already present on that logger. If no tornado
	handler is present at the time the IO loop is started, tornado
	will call logging.basicConfig, which isn't what we want to happen.
	Overridden from the base class, which unconditionally adds a new
	StreamHandler every time.
	See also:
	- https://github.com/tornadoweb/tornado/blob/v6.0.3/tornado/ioloop.py#L427-L445 # noqa: E501
	- https://github.com/tornadoweb/tornado/pull/741
	"""
	logger = logging.getLogger("tornado")
	if not logger.handlers:
	logger.addHandler(logging.NullHandler())

envisage/envisage/plugins/ipython_kernel/tests/test_internal_ipkernel.py

Lines 29 to 35 in 1eccd80

	if ipykernel_available:
	import ipykernel.iostream
	import ipykernel.ipkernel
	import ipykernel.kernelapp
	import ipykernel.zmqshell
	import IPython.utils.io
	import tornado.ioloop

and

envisage/envisage/plugins/ipython_kernel/tests/test_internal_ipkernel.py

Lines 242 to 244 in 1eccd80

	def test_init_ipkernel_with_explicit_gui_backend(self):
	loop = tornado.ioloop.IOLoop.current()

It also may be worth adding more details to the readme about what the optional dependencies are useful for / that the required dependencies are needed for the core envisage package. I was not sure how much to say here, and other enthought package readmes seem to be pretty brief.

[aaronayres35]

I do not know for sure if version >=1.8 is specifically required. Admittedly, I blindly took this from traits.

[mdickinson]

Traits needs Sphinx >= 2.1; looks like the README for Traits is out of date there. I think the restriction on the Sphinx version for Traits resulted from the needs of the trait-documenter and the trait-documenter tests. It's probably safe to use an earlier version here, but 2.1 is a good minimum version.

[mdickinson]

Created enthought/traits#1320 to fix the Traits README

[aaronayres35]

ref: Note this PR is also related to #154 as it updates part of the readme.

[aaronayres35]

I maybe should add "docs": ["enthought-sphinx-theme", "Sphinx!=3.2.0"] here as well. Additionally the keys here could be more specific.

[rahulporuri]

Tthat sounds like a good idea to me. The keys don't need to be anymore specific that what they are in traits.

[rahulporuri]

Tthat sounds like a good idea to me. The keys don't need to be anymore specific that what they are in traits.

[mdickinson]

The issue originally mentioned tornado but I am not sure this is still really needed?

For what it's worth, my usual rule with dependencies is that if something is explicitly imported from, it should be listed under the dependencies (even if it's a dependency of something else). So e.g. if a package imports from both traits and traitsui, the dependencies should list both, even though traits is a dependency of traitsui.

[aaronayres35]

Should I link the github repo or the pypi page? Traits refers to pypi, I think pyface refers to the traits github. In gerenal maybe pypi links would make more sense? At the very least for non Enthought related dependencies like ipykernel

[mdickinson]

Agreed that PyPI links make most sense. People reading the README on GitHub probably don't really care; people reading the project description on PyPI are more likely to want PyPI links.

[aaronayres35]

I went with Pypi links for non enthought stuff, and github for enthought

[aaronayres35]

ah I hadn't refreshed and just saw this comment, I will use pypi links for pyface and traitsui as well

[rahulporuri]

LGTM