From ec86e94d8ea18ed43a77d5ea67c890eb787bf169 Mon Sep 17 00:00:00 2001 From: "Jeong, YunWon" Date: Wed, 5 Aug 2026 12:54:19 +0900 Subject: [PATCH] bench: drop the windows closure_per_call overlay MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The counter it recorded is not stable on that host. The overlay was written from three readings of 416 (two named in check.py's overlay comment, one on #1028's CI run); the next windows cranelift run reported 415 against it, so each value fails whenever the other one is recorded. `recursive_call_frame_relocation` moved the same way in the same run and in the opposite direction relative to its shared file — guard_failures 638 -> 637 on windows dynasm, where it had agreed at 638 one run earlier. Two counters disagreeing with themselves across runs, one toward its shared value and one away from it, is not a per-host fact an overlay can hold. The stability re-runs re-run one binary inside one job, so an oscillation between jobs is invisible to them. Assisted-by: Claude --- .../bench/synth/closure_per_call.cranelift.win32.jitstats | 8 -------- 1 file changed, 8 deletions(-) delete mode 100644 pyre/bench/synth/closure_per_call.cranelift.win32.jitstats diff --git a/pyre/bench/synth/closure_per_call.cranelift.win32.jitstats b/pyre/bench/synth/closure_per_call.cranelift.win32.jitstats deleted file mode 100644 index 659b26fd177..00000000000 --- a/pyre/bench/synth/closure_per_call.cranelift.win32.jitstats +++ /dev/null @@ -1,8 +0,0 @@ -bridges_compiled=2 -descr_set_absent=0 -descr_set_ambiguous=0 -descr_set_stale_absent=0 -guard_failures=416 -internal_compile_panics=0 -loops_aborted=0 -loops_compiled=4