Re: thought on Lua's hash-table invariants

[云风 Cloud Wu <cloudwu@...>]

Philippe Verdy <verdyp@gmail.com> 于2020年4月28日周二 上午6:09写道：

> The basic function used by Lua for its hash table hwoever is not very strong (it is made to be fast): one way to still use Lua tables for critical data is to first create your own strong digest for any private key that you want to secure against malicious collisions, and then append this digest to the unique key to generate the actual key to be used in table, instead of indexing the table directly by this private key (such appending will still preserve the uniqueness). I wonder however if Lua allows an application to specify its own secure hashing function for storing/reading (key=value} pairs in its tables: is there a metamethod we can use in tables? (setting this methamethod to a different hashing function should instantly cause the reashing of all keys already present in that table).

My experience is that don't use a huge table if you changes it's keys
frequently. It will cause stop-the-world problem.

I found this issue years ago. There is a huge table (nearly 1M nodes)
in our online game server. We add some keys and remove keys
frequently, the total number of the nodes not changed but `lastfree`
moves in one way. So rehasing occured periodic , it stopped the world
for a long time.

My solution is spliting this huge table into multiple smaller one.

I think an user-defined hashing function may not be helpful in small
tables, and metamethod could be much slower than native one.

-- 
http://blog.codingnow.com
_______________________________________________
lua-l mailing list -- lua-l@lists.lua.org
To unsubscribe send an email to lua-l-leave@lists.lua.org