Fix clock sync problems

[krillr]

When clocks are out of sync between hosts, improper logouts can occur. This stops this by not logging out when last activity is in the future in comparison to the local clock.

[jpic]

We'll need a test case though

[yscumc]

I think it would make more sense to compare timedelta directly with timedelta instead of converting the type first, due to how python handles it internally.

See:

>>> a = timedelta(0,0,-1)
>>> b = timedelta(0,0,1)
>>> a < b
True

>>> a.days < b.days
True
>>> a.seconds < b.seconds
False
>>> a.microseconds < b.microseconds
False

>>> a
datetime.timedelta(-1, 86399, 999999)
>>> b
datetime.timedelta(0, 0, 1)

This weird behavior is also why I added the client_idle_for < 0 check in 7024081 from PR #3

In the long run, I think instead of

if delta.seconds >= EXPIRE_AFTER

this should be better

if delta >= timedelta(seconds=EXPIRE_AFTER)

[krillr]

Ah, yes you're right -- this also takes care of the case where it's been 1 day and 1 second since last activity.

[yscumc]

Wow I didn't even think of that.

That explains the weird bug which I encounter sometimes -- when I suspend my laptop, and resume it a long time later, sometimes the session would still be alive. I didn't have time to track down this problem and couldn't reproduce it reliably so I never added an issue here. The root cause of this problem is probably this comparison!

Thanks for solving the mystery of this frustrating bug!

[krillr]

Pull request #14 contains a unit test for future-time last activities and a change to using timedelta comparison, to avoid the issue we just talked about. Tests should be running shortly :)

[krillr]

For some reason travis keeps failing on Django 1.5 on what appears to be an unrelated test. No idea why o.o