Proposed patch changes the complexity of deque indexing from linear to constant. All other operations preserves its amortized cost.

New deque implementation use two-level array instead of linked list. Since free space is reserved at both side, new blocks can be added at both sides for constant time. Memory consumption is almost the same.

Sorry Serhiy, I really don't want to do this.  Indexing is the least important part of the API and should not drive the overall structure.  This optimization is at odds with the design goal of a structure that is efficient at near end-points and is not intended to compete with lists which are supposed to be efficient for random access indexing.  The use cases for deques rarely use access by position anywhere are from the end-points, so in a way this is a false optimization that doesn't account for the intended use cases.  The core design has been stable for a very long time and I'm very happy with it as is.