How do I boost difftastic by 4x

Difftastic is a structural diff that understands syntax. The diff results it generates are very fancy, but its performance is poor, and it consumes a lot of memory. Recently, I boosted it by 4x while using only 23% of memory (#393, #395, #401). This post explains how I accomplished this. Hope it can bring you some inspiration. When I started to write this post, not all optimizations were reviewed and merged.