Skip to the content
Software Made Clear
About

LANGUAGE REFERENCE · /python/deque/

deque in Python

4 min read Updated September 27, 2026 Last verified September 12, 2026 CPython 3.14
ANSWER

Use a deque when you append or pop at both ends. A Python list is O(n) on the left, because removing the first item shifts every remaining element down one place.

IN PLAIN TERMS

You can couple or uncouple a carriage at either end of a train in one move, and a deque works the same way. A list is more like a numbered row of chairs: take the front one away and everybody has to shuffle up a seat.

list.pop(0) — every element moves one slot leftevery remaining element shifts one slot leftdeque.popleft() — one block is unlinked, nothing else movesblocks of 64 slots, doubly linkedlist.pop(0) — every element moves oneslot leftevery remaining element shifts one slotleftdeque.popleft() — one block isunlinked, nothing else movesblocks of 64 slots, doubly linked
FIG 1 A deque is a doubly linked list of fixed-size blocks, not a ring buffer over one array — which is exactly why both ends are cheap and the middle is not.

Try all four ends

A deque is the same container whether it acts like a stack or a queue. Add and remove items at either end, then set maxlen=3 to see which item silently falls off.

Left endRight end

    The deque is empty. Add an item at either end.

    CPython 3.14 
    from collections import deque

q = deque([1, 2, 3], maxlen=4)
q.appendleft(0)     # deque([0, 1, 2, 3])
q.append(4)         # deque([1, 2, 3, 4]) — 0 fell off the left
q.rotate(1)         # deque([4, 1, 2, 3])

    The Python-specific detail

    Because those blocks are linked rather than contiguous, indexing into the middle is O(n) while both ends are O(1). That is the reason q[len(q)//2] inside a loop is the classic way to make a deque slower than the list it replaced — the mirror image of the cost a level-order tree walk pays when it takes from the front of a plain list instead.

    The second surprise is maxlen : it discards silently from the opposite end. Perfect for a rolling window, quietly destructive if you expected an exception — the same kind of silent wrong outcome as an API that replays a stored response and sends the confirmation email a second time, because the email was never part of the record.

    Re-checked against CPython 3.14.7's collections documentation — which states that indexed access is O(1) at both ends but slows to O(n) in the middle, and that a bounded deque discards a corresponding number of items from the opposite end — and Modules/_collectionsmodule.c on the 3.14 branch, whose own comment still describes the doubly-linked list of fixed-length blocks (BLOCKLEN 64), 12 September 2026. Unchanged since the 3.12 check of 19 August 2026: every claim on this page held verbatim across both releases.