Recently, I used the Python package mdict-utils, which is well written and organized. It works well for small and medium projects. In my case of English Wikipedia, its MDX-generating speed is not sufficient.
It turns out mdict-utils compresses and writes blocks sequentially. In my fork, the speed of generating MDX from TXTs is increased significantly by
- replacing zlib with the deflate.
- employing the paradigm compress blocks in parallel and write them sequentially in batch.
You can try my fork GitHub - leanhdung1994/mdict-utils: MDict pack/unpack/list/info tool.