SQLite v4 was supposed to have modular, pluggable backends, of which LMDB was only the first (of hopefully many). They were also working on their own from-scrat

SQLite v4 was supposed to have modular, pluggable backends, of which LMDB was only the first (of hopefully many). They were also working on their own from-scratch implementation of an LSM backend, which was also abandoned. As I recall, they decided that the mmap-related performance tricks they learned from working with LMDB could just be plugged into their own SQLite v3 Btree engine. (Indeed, recent SQLitev3 is a bit faster than when the SQLitev4 project began. But still doesn't match LMDB.)