When a tight for loop without calls is turned into two traces that transfer directly to each other, it seems that CHECK_EVAL_BREAKER() is never called, and the loop won't be interruptible.

See faster-cpython/ideas#669 (comment), where @brandtbucher and I reasoned that this might occur in the nbody benchmark (when the max trace length is reduced to 400).

While we didn't observe the predicted speedup, we really do believe that in this case the thread is uninterruptible.

@markshannon Any thoughts?

Linked PRs

• GH-117442: Check eval-breaker at start (rather than end) of tier 2 loops #118482