Thanks for the great write-up, @markshannon
Would there be a chance to get option 3 in two phases ?
Phase 1: Optionally switch on nogil mode and turn off specializations using a command line switch, so that the two don’t clash. You’d only be able to run one or the other, not both at the same time. People who’d need fast single thread performance would go with the specialization mode. Others with more need to scale up Python apps to multiple cores could use the nogil version.
Phase 2: Slowly work on merging the two approaches.
Or would this not be feasible because the “slow merge” would actually require redesigning the whole specialization approach ?