test: add an --offline option to pytest

[Exirel]

Usage:

$ py.test --offline

This option skip all tests that can not be run offline, as they are marked as "online". Either from manual tag:

@pytest.mark.online
def test_call_google():
    request.get('https://google.com')

Or from a command example:

@example('.title http://google.com', '[ Google ] - google.com', online=True)
def title_command(bot, trigger):
    # ...

This is especially useful for people like me who are always traveling and can't have access to a reliable Internet connection.

Existing online examples have been marked as such.

[dgw]

Before I dive into this—I like the idea—we might get some utility from a set of options:

• --offline, alias --offline-only: run only offline-safe tests
• --online, alias --online-only: run only online tests (which call APIs or whatever)
• --no-offline: skip offline tests
• --no-online: skip online tests

This might be overkill, true. If it makes implementation easier we can do just the -only args and skip the no- options. My thinking here is we could later configure Travis to run only the offline tests during required-pass jobs, and have separate test runs for the online tests that are allowed to fail. Then we wouldn't have random transient changes to DuckDuckGo search results (or whatever) failing related PRs.

[dgw]

LGTM. Needs only two things:

• rebasing to avoid merge conflicts introduced on master since this PR was opened
• adding the online arg to the docstring of the module.example class

As discussed on IRC, my ideas for a more complicated setup we could use to separate online tests (that might fail) in CI will come in a future PR, possibly after getting bikeshedded first in a dedicated issue.

---

I can't resist correcting the commit message, which contains an incorrect call to request.get. (Well, I guess it could be correct if the file does import requests as request, but why would you do this…)

And since I'm already commenting on that, I can also point out that "This option skip all tests" is missing something—and you will instantly know what. 😀 (There's also "can not" -> "cannot"—using that in two-word form is very uncommon in present-day writing, but it's not wrong per se.)

[Exirel]

I got reminded of this PR when I had trouble testing locally because some random website was inaccessible and/or very slow.

@dgw I think this PR is becoming more and more important, since Travis can literally fails for the wrong reasons because of online-only tests.

[dgw]

Hmm… @Exirel I think you did something weird in your last rebase. The commit is showing me as author, which can't be right. This code is all you. 😕

[Exirel]

@dgw I've honestly no idea how or why. :|

[Exirel]

Yay git reset HEAD~1 worked as intended! Should be good now.

[dgw]

We'll talk about further enhancements to reduce false CI failures for a future PR. 😸