Tests are broken on master about zope.testbrowser

[mgedmin]

THE BLOODY NARRATIVE TEST SETS A COOKIE WITH AN EXPIRATION DATE OF JANUARY THE FIRST TWO THOUSAND AND NINETEEN ANNO DOMINI.

Nice ⏲️ 💣 !

[mgedmin]

I've looked at it, and ugh. Dealing with long narrative doctests where the only test runner is tox -e docs is like walking through mud.

Something changed in webob's cookie parser? Or we're not using it correctly? I noticed two problems:

• using a space in a cookie's comment field causes a RuntimeWarning from webob (but the space gets encoded as %20, correctly, like the doctest expects) (this happened before and did not cause the build to fail)
• cookie expiration date gets lost somewhere (on my machine)
• the entire cookie disappears (on Travis CI)

The last successful build was https://travis-ci.org/zopefoundation/zope.testbrowser/builds/472125193.
The first failed build was https://travis-ci.org/zopefoundation/zope.testbrowser/builds/474117777.
The difference (looking at the Python 2.7 job):

• successful build had beautifulsoup4 4.6.3
• failed build had beautifulsoup4 4.7.0
• failed build also got new dependencies on backports.functools-lru-cache (1.5) and soupsieve (1.6), but I don't think it is relevant

Both builds had the same zope.testbrowser commit and had the same webob version.

[mgedmin]

cookie expiration date gets lost somewhere (on my machine)

This is because the tests are sensitive to the user's locale! 🌐 🤡 🍾

If I set LC_TIME=C, I can reproduce the KeyError problem I see on Travis.

(This might be an actual bug lurking somewhere!)