CODEWITHSUNDEEP
javascriptarraystime-complexity
Albert
Albert

Reputation: 68330

JavaScript runtime complexity of Array functions

Is the runtime complexity defined by the JS standard on common Array functions like push, pop, shift, slice or splice? Esp. I'm interested in removing and inserting entries at random positions. If the complexity is not defined, what can I expect, e.g. in V8?

(This question was inspired by this. Also, this benchmark, posted here, also makes me curious, but maybe that is some unrelated phenomena.)

(A very related question is here. However, one of the comments on the accepted answer says that it is wrong now. Also, the accepted answer does not have any reference that the standard really defines it that way.).

Upvotes: 97

Views: 80954

Answers (3)

Sarwar Sateer
Sarwar Sateer

Reputation: 475

in very simple words

push -> O(1)

pop -> O(1)

shift -> O(N)

slice -> O(N)

splice -> O(N)

Here is a complete explanation about time complexity of Arrays in JavaScript.

Upvotes: 18

Lu4
Lu4

Reputation: 15030

Just a side note, it is possible to implement shift / unshift, push / pop methods in O(1) using RingBuffer (i.o.w CircularQueue / CircularBuffer) data structure. It would be O(1) for worst case whenever the circular buffer is not required to grow. Did anyone actually measure performance of these operations? It's always better know rather than to guess...

Upvotes: 4

chuckj
chuckj

Reputation: 29625

The ECMA specification does not specify a bounding complexity, however, you can derive one from the specification's algorithms.

push is O(1), however, in practice it will encounter an O(N) copy costs at engine defined boundaries as the slot array needs to be reallocated. These boundaries are typically logarithmic.

pop is O(1) with a similar caveat to push but the O(N) copy is rarely encountered as it is often folded into garbage collection (e.g. a copying collector could only copy the used part of an array).

shift is at worst O(N) however it can, in specially cases, be implemented as O(1) at the cost of slowing down indexing so your mileage may vary.

slice is O(N) where N is end - start. Not a tremendous amount of optimization opportunity here without significantly slowing down writes to both arrays.

splice is, worst case, O(N). There are array storage techniques that divide N by a constant but they significantly slow down indexing. If an engine uses such techniques you might notice unusually slow operations as it switches between storage techniques triggered by access pattern changes.

One you didn't mention, is sort. It is, in the average case, O(N log N). However, depending on the algorithm chosen by the engine, you could get O(N^2) in some cases. For example, if the engine uses QuickSort (even with an late out to InsertionSort), it has well-known N^2 cases. This could be a source of DoS for your application. If this is a concern either limit the size of the arrays you sort (maybe merging the sub-arrays) or bail-out to HeapSort.

Upvotes: 185

Related Questions