Hi Datong, This proposal amounts to ‘incremental Hyperscan compilation’ – which would be pretty nice to have, I agree. However, it would be a very large change and difficult to QA – our testing systems would need to simulate many different add/delete sequences that yield the same regular expression set but perhaps through different paths. Can you give some examples of cases where Hyperscan is very slow to compile for you? Given a certain amount of engineering effort, we usually conclude that it’s better to make normal Hyperscan compile faster rather than add incremental compilation. It’s quite possible that you are finding inefficiencies in the Hyperscan compiler that should just be fixed. Regards, Geoff. From: Hyperscan [mailto:hyperscan-bounces@lists.01.org] On Behalf Of Datong Li Sent: Tuesday, April 25, 2017 8:04 PM To: Hyperscan regular expression matching library <hyperscan(a)lists.01.org&gt; Subject: [Hyperscan] Questions about NG object and database rebuilding time-consume Hi all, I’ve read the code in hs_compile_multi_int, I found that building hs_database by using ng object. I wonder to know if we could keep the ng object in memory and use in next time we build a new database after a little modification, when adding/deleting a expression. As we knows, adding/deleting some expression for matching is frequency happened in real product situation, and rebuild database is time-consume a lot. I want to know how could I reduce the time of rebuilding hs_database when I use hyperscan in frequency expression modification situation. -- With best regards, Datong Li