@network decorator can mask failed tests

See discussion #5812

The skip behaviour was motivated by travis failing tests due to
wonky net cinnectivity which can mask actual problems such as broken urls.

Perhaps make the behavior conditional on environment (travis can be detected via
environment vars)

[jtratner]

I made a decorator for that with_connectivity_check, which would handle this fine, at the cost of an extra network connection if the test fails with an IOError.

That's close but not quite it, I can't think of a case where it's useful given the
qualification in the docstring:

  In comparison to ``network``, this assumes an added contract to your test:
    you must assert that, under normal conditions, your test will ONLY fail if
    it does not have network connectivity.

Isn't that another way of saying "if your test is a connectivity check", since I can't
test+fail for anything else?

What's needed is to catch only network-related exception and qualify them by a connectivity check.
That would tie the functionality to the specific exceptions raised by urllib (vs. say requests or urllib2).

OTOH, we could do a connectivity check prior to any network test, perhaps via a singleton
so we don't launch 800 requests against google everytime travis runs.

#5812 just exposed the problem, we either need to apply a fix everywhere or let it go.

thoughts?

[jtratner]

I wrote the qualification poorly. The point is that, if the test fails, it
bubbles up the failure if it is able to connect to the URL (meaning that it
has Internet connectivity)

[jreback]

@y-p @jtratner what's status of this?

if @jtratner 's clarification is true, I say we just fold the check into
the network decorator by default.

[jreback]

@jtratner you have this one?

[jtratner]

let me make sure this actually works the way it does. Sorry I've been snowed under with work right now and so I've gone a bit incommunicado, catching up a bit tonight.

[jtratner]

btw - @y-p do you have an example where this failed inappropriately (and, if so, I'll try and fix, otherwise I'll clarify the docstring)

[jreback]

this should fail with a non existent URL

eg yahoo changes it

[jtratner]

I see what you mean - let me put together a PR that removes current @network and replaces with @with_connectivity_check instead.

lost you, doesn't #5812 count as failed inappropriately?

That would mess up the nosetests selector functionality.
Also, why rename the decorator on 200 seperate tests when you just
always want different behavior from it ? @network needs to be changed.

[jtratner]

@y-p I think we're saying the same thing: you okay if I remove @network and rename @with_connectivity_check to @network?

yep.

bumped to 0.14.0