Since the STL's deque is incorrect and will be replaced in vNext, currently this implemention should be chosen unconditionally. By default, this implementation has a block size of 4KB. Please do not modify it outside of testing scenarios, as doing so would result in inconsistent ABI. libc++ also has a 4KB block size, so any performance differences reflect actual disparities. libstdc++ has 8 elements per block, which is more memory-efficient when dealing with a small number of elements.

bizwen::deque and libstdc++ each have their own strengths and weaknesses, with an overall negligible performance difference. bizwen::deque is always better than libc++ in most test items and slightly behind libc++ in a few test items. The construction speed of bizwen::deque is tens to hundreds of times faster than STL. In most test items, it is 3 to 15 times faster than STL, only falling slightly behind STL in at.