A GopherCon 2025 talk exploring how Go handles concurrency correctness through its memory model and race detector. Covers why multi-core systems break sequential assumptions, how the Go memory model guarantees data-race-free sequential consistency (DRFSC), and how the happens-before relationship is formalized using vector clocks — drawing an analogy to Einstein's special relativity and Minkowski light cones. Includes a live benchmark comparing Go 1.18 (TSAN v2) vs Go 1.19 (TSAN v3), demonstrating ~6x performance improvement and removal of the ~8,000 goroutine limit in the race detector, with CPU profiling showing the shift from a spin-lock to a lock-free compare-exchange approach as the key driver.
•25m watch time
2 Impressions