Uses zero-copy bytes on CPython 2.7 as well as 3.x.
Further reduces the overhead for caching objects by up to 50% (down to 96 bytes on some platforms from 152 bytes)
Most operations actually got faster. The one exception is the simple straightforward read test, which got a bit slower, but is still much faster than it used to be. I think the improvements are likely to due to better cache locality, plus the opportunity to improve the algorithms a bit. I think the added read cost is due to the use of an ordered, as opposed to hashed, container.
Improves on the work done in #358 and fixes #186.
readtest, which got a bit slower, but is still much faster than it used to be. I think the improvements are likely to due to better cache locality, plus the opportunity to improve the algorithms a bit. I think the added read cost is due to the use of an ordered, as opposed to hashed, container.