A community for discussing the Python programming language.
The same exact thing I do with multiprocess Python, I'm not going to update all of my code.
Multiprocess approach is arguably more robust, you can't corrupt memory from another thread, because the only point of interaction you have with your other rpocesses is queue and semaphore.
But with multiprocessing you're paying extra cycles to serialize and deserialize with those data structures (plus the added cost of storing copies in multiple places), and processes are much heavier to spin up...
Yeah, there is overhead. It's not so critical for Python, because I am using it for scripting and prototyping, and when I need performance, I rewrite critical code parts in C++.
Just downloaded the free-threading RC version via pyenv. Looking forward to taking it for a spin.