the last two rows are where we lose, and it’s not close - regex uses Teddy/SIMD for literals and short alternations, and we simply don’t have that yet. the case-insensitive row is the opposite story: regex falls back to an NFA when case-insensitivity with (?i) blows up the state space, dropping to 0.03 MiB/s. yes, that 16,833x number is real - this is exactly the NFA slowdown i described in the previous post. it’s not a contrived input, just case-insensitive matching on a dictionary. O(n * m) is technically “linear in n”, but when m blows up it stops feeling linear real fast.
数据来源：中国人民财产保险股份有限公司
，详情可参考新收录的资料
FT Edit: Access on iOS and web
Трамп призвал не бояться роста цен на нефть02:56，推荐阅读新收录的资料获取更多信息
silently reorder operations based on a precedence table. Instead, Mog enforces explicit。业内人士推荐新收录的资料作为进阶阅读
Цены на нефть взлетели до максимума за полгода17:55