Very interesting point. Concurrent hash maps are complex.
Quote Tweet
Replying to @jckarter and @pcwalton
I suspect it’s hard to get good performance without a GIL in an “every object is a hashtable” language with shared mutable state
4
24
Conversation
Replying to
There are fancy approaches but generic hash-based load balancing works fine.
Create a hash table equal in size to number of cores. Each bucket has a hash table with a separate lock. Inner hash tables reuse the hashing work for the outer one by shifting out bits that were used.
1
I don't think it's relevant to this use case since I don't think you really want more than one lock per object. The goal seems to be simply providing safety rather than a good way to do concurrency with fine-grained locking.
I think Python's issue is mostly CPython extensions.