Reactors break if a DelayedCall is scheduled with .time = NaN

New submission from bdew <None>:

I'm currently debugging [http://allmydata.org/trac/tahoe/ticket/737 allmydata-tahoe issue #737] the problem there seems to be that somehow a delayed call is scheduled with a time of NaN, this leads to 2 problems in the reactor

1) ReactorBase.runUntilCurrent() won't run that call or anything later

{{{
while self._pendingTimedCalls and (self._pendingTimedCalls[0].time <= now)
}}}

because NaN compared with anything always returns false - this loop never runs and no other calls after the bugged one in self._pendingTimedCalls will ever run.

Also this and any later call will never be removed even if canceled - this loop handles cancelations too.

2) ReactorBase.timeout() returns 0 when that call is in _pendingTimedCalls[0]

This will make the reactor do iterations with 0 timeout, which on most (all?) reactors will lead to 100% CPU usage

Minimal code to demonstrate this bug:


{{{
from twisted.internet import reactor

def foo():
    print 'THIS WILL NEVER BE CALLED'

def setup():
    reactor.callLater(1,reactor.stop)
    reactor.callLater(0,foo).reset(float('nan'))

reactor.callWhenRunning(setup)

reactor.run()
}}}

foo() is never called, the reactor never stops and the process eats 100% CPU, i've tested this on twisted 8.2.0 on windows (select reactor) and linux (poll reactor)

I know setting the timeout to NaN is not something sane to do, but i think twisted should break less horibly when some other bug causes that to happen ;)

----------
Type     : defect
Component: core
Keywords : 
Priority : normal
Nosy     : [email protected]
----------
http://twistedmatrix.com/trac/ticket/3884