s2/decode_amd64.s (1)

54-569: 🏗️ Heavy lift

Add a regression/stress test to prevent this asm+async-preemption crash from resurfacing.

Because the reported failure is specifically an async preemption SIGSEGV during s2.Decode on amd64, it would be good to add a targeted stress test that:

• runs many concurrent decodes,
• encourages async preemption (e.g., frequent scheduling/yield points and long-running loops),
• and repeats enough times to catch rare stack-corruption failures.

Even a best-effort test (may be flaky unless tuned) would materially reduce risk for future asm changes.

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@s2/decode_amd64.s` around lines 54 - 569, Add a targeted stress test that
repeatedly runs concurrent s2.Decode calls to reproduce async-preemption stack
corruption; create a test (e.g., TestDecodeAsyncPreemptStress) that spawns many
goroutines (N=hundreds), sets runtime.GOMAXPROCS>1, and in each goroutine loops
many times (thousands) calling s2.Decode on specially crafted inputs that
exercise long-running asm paths (long literal and copy sequences), interleaving
runtime.Gosched() or time.Sleep(0) to encourage async preemption and verifying
every Decode returns the expected result (no panic and correct error code); run
the test with short timeouts and allow tuning of goroutine/iteration counts so
CI can run a smaller default but maintain a heavy local/regression mode.