That's not really what's happening in the article. If you read through the single file benchmark, you'll see several clever algorithmic improvements (like rares

That's not really what's happening in the article. If you read through the single file benchmark, you'll see several clever algorithmic improvements (like rarest byte guessing, building a set of variants for Unicode-aware multiple pattern matching, etc...). The author literally concedes that the .gitignore feature was not done for performance, and actually carries a significant overhead in large directory trees. For the sake of comparability, the study was controlled for the .gitignore overhead.