There is apparently a race between the time and low resolution now that can make mUntailAt timestamp be cleared just before the time is about to fire.

mUntailAt <= TimeStamp::NowLoRes() may eval to true just few milliseconds before the timer is scheduled to fire, but we don't cancel the timer. In Notify() - when the timer fired - we than compare against null mUntailAt. Though, the condition is redundant, since mUntailAt is always exactly: - non-null when we are inside the "keep tailed" interval - null otherwise (after the timer has fired) This is ensured by the code. Hence, checking only if mUntailAt.IsNull() is enough to evaluate the condition correctly.

Credits for reporting and verifying the fix locally: Paul Adenot. Thanks! :)

We can also remove the null TimeStamp() assignment (it's a no op).

Pushed by honzab.moz@firemni.cz: https://hg.mozilla.org/integration/mozilla-inbound/rev/b1f99cdeddec Remove potential timer race and assertion in RequestContext tailing feature, r=dragana