From e10c934d5a756c1a4c7b547ebd02c403ddfde7b8 Mon Sep 17 00:00:00 2001 From: GRAB Racem Date: Sat, 10 Oct 2026 01:36:11 +0200 Subject: [PATCH 1/3] fix(ghidra): keep the cross-format branch callee multi-block The cross-format oracle requires a callee whose recovered control flow has more than one basic block, and rea_cross_branch provided it through a symmetric if/else of tail calls. Clang 21 folds that shape into a conditional negate plus one tail branch, a single basic block, so the lane rejected a correct Ghidra analysis of the fixture. Give the branch a loop whose trip count depends on a volatile global: no compiler can eliminate the backedge, and the callee stays multi-block at every optimization level on every toolchain. --- tests/conformance/ghidra/cross-format.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/tests/conformance/ghidra/cross-format.c b/tests/conformance/ghidra/cross-format.c index f718b9b54..22efa6159 100644 --- a/tests/conformance/ghidra/cross-format.c +++ b/tests/conformance/ghidra/cross-format.c @@ -8,10 +8,15 @@ __attribute__((noinline, used)) int rea_cross_leaf(int value) { } __attribute__((noinline, used)) int rea_cross_branch(int value) { - if (value > 10) { - return rea_cross_leaf(value); + /* A volatile-dependent loop keeps this callee multi-block at every + optimization level: newer clang folds a symmetric if/else of tail calls + into a conditional negate, and cross-format verification requires a + recovered multi-block callee. */ + int adjusted = value > 10 ? value + 1 : -value; + while (adjusted > 100) { + adjusted -= rea_cross_global; } - return rea_cross_leaf(-value); + return rea_cross_leaf(adjusted); } __attribute__((noinline, used)) int rea_cross_indirect( From 40f632a155bff2defdea09e467229c703bc0b0e9 Mon Sep 17 00:00:00 2001 From: GRAB Racem Date: Sat, 10 Oct 2026 02:06:15 +0200 Subject: [PATCH 2/3] ci: re-run the suite after the main-branch shard flake The failed shard is the JavaScriptSemanticGraphBuilder short-circuit mutation test, which also failed on main run 38003459598 at the same base commit (08ec2bf8) before this branch started CI, and passed on main run 38003459600 minutes later at the same commit. From e17290cd2a17931d7a2af70f74b76b94276bd788 Mon Sep 17 00:00:00 2001 From: GRAB Racem Date: Sat, 10 Oct 2026 02:34:52 +0200 Subject: [PATCH 3/3] ci: re-run the suite on the corrected main checkpoint b8a8190b restored the guarded-mutation uncertainty the failing test encodes, so the shard should pass against the new merge base.