Currently, nextv() can only signal the natural end of an iterator by yielding an empty array. In the fallback _nextv() implementation below, when _next() signals end by yielding an undefined key and/or value (line 147), this means _nextv() does not signal end (line 148, unless acc is empty). Consequently, a consumer will call nextv() again, which means we call _next() again (line 160) which has already signaled end.
Whether that's a problem depends on implementation. It is in many-level (the replacement for multileveldown that I'm working on atm, and can easily have its own fix) and could be in others too. It's not a problem for implementations written so far (memory-level, classic-level, browser-level) because those implement their own optimized _nextv().
Solution: set a new private kEnded property to true on line 148, and check that it's false before calling _next(). A seek() should reset it to false. Maybe use kEnded in more places, as a general protection.
Currently,
nextv()can only signal the natural end of an iterator by yielding an empty array. In the fallback_nextv()implementation below, when_next()signals end by yielding an undefined key and/or value (line 147), this means_nextv()does not signal end (line 148, unlessaccis empty). Consequently, a consumer will callnextv()again, which means we call_next()again (line 160) which has already signaled end.Whether that's a problem depends on implementation. It is in
many-level(the replacement formultileveldownthat I'm working on atm, and can easily have its own fix) and could be in others too. It's not a problem for implementations written so far (memory-level,classic-level,browser-level) because those implement their own optimized_nextv().Solution: set a new private
kEndedproperty to true on line 148, and check that it's false before calling_next(). Aseek()should reset it to false. Maybe usekEndedin more places, as a general protection.https://github.com/Level/abstract-level/blob/601b47c36903760c3232907423645833206871f5/abstract-iterator.js#L142-L161