From a2276bd7414ded75a23999c8a9226c2265750c54 Mon Sep 17 00:00:00 2001
From: Swapnil <78632212+swapnilpaliwal-sd@users.noreply.github.com>
Date: Mon, 28 Sep 2026 08:59:02 -0700
Subject: [PATCH] ci: key the compiled engine cache on the CPU's feature flags,
 and drop a cached engine that dies with SIGILL

The engine is compiled with -march=native, and the CI cache key named the CPU
by its model name only. Hosted runners that report the same model name do not
all expose the same instruction set, so a binary compiled on one was restored
onto another and died with an illegal instruction on every TypeScript case.

The key now hashes the model name and the flags line of /proc/cpuinfo. The
driver also removes a locally cached binary that exits with SIGILL, so a bad
entry in a shared cache kills one run, not every run after it.
---
 .github/workflows/ci.yml      |  7 +++++--
 graph/pipeline/run-souffle.sh | 13 ++++++++++++-
 2 files changed, 17 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index e09200258..a59ce1fd6 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -397,12 +397,15 @@ jobs:
       # Two things decide the binary that ENGINE_ID does not cover, so they are in the
       # key too: the driver that compiles it (run-souffle.sh holds the compiler flags),
       # and the runner's CPU, because the driver compiles with -march=native and a
-      # binary built on one CPU model can die with an illegal instruction on another.
+      # binary built on one CPU can die with an illegal instruction on another.
+      # The CPU part hashes the FEATURE FLAGS, not only the model name: hosted runners
+      # that report the same model name do not all expose the same instruction set, so
+      # a model-name key restored a binary that died with SIGILL on every case.
       - name: this language's engine id
         id: eid
         run: |
           echo "id=$(bash graph/pipeline/run-souffle.sh --language ${{ matrix.lang }} --print-engine-id)" >> "$GITHUB_OUTPUT"
-          echo "cpu=$(grep -m1 'model name' /proc/cpuinfo | sha256sum | cut -c1-12)" >> "$GITHUB_OUTPUT"
+          echo "cpu=$(grep -E '^(model name|flags)[[:space:]]*:' /proc/cpuinfo | sort -u | sha256sum | cut -c1-12)" >> "$GITHUB_OUTPUT"
 
       - name: cache the compiled Soufflé engine
         if: ${{ !inputs.fresh }}
diff --git a/graph/pipeline/run-souffle.sh b/graph/pipeline/run-souffle.sh
index 5cf84aacf..2b0cb5424 100755
--- a/graph/pipeline/run-souffle.sh
+++ b/graph/pipeline/run-souffle.sh
@@ -699,7 +699,18 @@ while [ "$iter" -lt 50 ]; do
     echo "▶ solving with profiling -> $AXIOM_SOUFFLE_PROFILE"
     "$PBIN" -F "$FACTS" -D "$RAW" -p "$AXIOM_SOUFFLE_PROFILE"
   else
-    "$BIN" -F "$FACTS" -D "$RAW"
+    # A cached binary is compiled with -march=native. Restored onto a CPU without one of
+    # the instructions it uses (a shared cache, a CI cache keyed too coarsely), it dies
+    # with SIGILL (exit 132) before solving anything. Never leave it there to kill every
+    # later run the same way: drop the cache entry, so the next run recompiles, and say so.
+    rc=0; "$BIN" -F "$FACTS" -D "$RAW" || rc=$?
+    if [ "$rc" -ne 0 ]; then
+      if [ "$rc" -eq 132 ] && [ -z "$PACKAGED" ]; then
+        rm -f "$BIN"
+        echo "❌ the cached engine $BIN died with an illegal instruction: it was compiled for a different CPU. Removed it; the next run recompiles." >&2
+      fi
+      exit "$rc"
+    fi
   fi
   # No frontier declared for this language: the solve above is the whole answer. Break
   # BEFORE the count, because the count is what misreported it. See issue #475.