Restructure: BUGGY (no RB_GC_GUARD) vs FIXED vs CASE_C (BUGGY+O0)

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: samuel-williams-shopify

.github/workflows/test-ruby-gc-bug.yaml

@@ -1,10 +1,21 @@
 name: Test Ruby GC Bug — Root Cause Investigation
 
+# Goal: understand WHY RB_GC_GUARD fixes the crash.
+#
+# The upstream fix (8351378bf3) adds RB_GC_GUARD after the last implicit use
+# of `blocking_operation` via the derived `operation` pointer. We want to:
+#
+#   BUGGY  — remove RB_GC_GUARD from fixed code → should crash
+#   FIXED  — base code (has RB_GC_GUARD already) → should pass
+#   CASE_C — BUGGY code + __attribute__((optimize("O0"))) on function → does O0 fix it?
+#
+# If BUGGY crashes and CASE_C does not, it is purely a compiler optimisation issue.
+
 on: [push, pull_request]
 
 jobs:
   investigate:
-    name: "CASE_A (O3, crashes) vs CASE_C (O0 on fn)"
+    name: "BUGGY vs FIXED vs CASE_C (O0)"
     runs-on: ubuntu-24.04
 
     steps:
@@ -17,90 +28,117 @@ jobs:
           ruby bison libyaml-dev libgdbm-dev libreadline-dev libncurses5-dev \
           liburing-dev autoconf automake gcc make binutils
 
-    - name: Clone Ruby (bug/blocking-operation-gc — CASE_B is base)
+    - name: Clone Ruby (bug/blocking-operation-gc — has fix applied)
       uses: actions/checkout@v4
       with:
         repository: samuel-williams-shopify/ruby
         ref: bug/blocking-operation-gc
         path: ruby-src
 
-    # ── CASE_A: NULL, full O3 (expected to crash) ─────────────────────────
-    - name: Patch to CASE_A — operation = NULL
+    # ── FIXED baseline: build with RB_GC_GUARD intact ─────────────────────
+    - name: Configure + build FIXED (RB_GC_GUARD present)
       working-directory: ruby-src
       run: |
         ./autogen.sh
-        ./configure --prefix=$HOME/.rubies/ruby-caseA --enable-shared --disable-install-doc cppflags="-DENABLE_PATH_CHECK=0"
-        sed -i 's/rb_fiber_scheduler_blocking_operation_t \*operation = get_blocking_operation(blocking_operation);/rb_fiber_scheduler_blocking_operation_t *operation = NULL;/' scheduler.c
-        grep -n "operation = NULL" scheduler.c
+        ./configure --prefix=$HOME/.rubies/ruby-fixed --enable-shared \
+          --disable-install-doc cppflags="-DENABLE_PATH_CHECK=0"
+        make -j$(nproc) && make install
+
+    # ── BUGGY: remove RB_GC_GUARD to reproduce original crash ─────────────
+    - name: Patch BUGGY — remove RB_GC_GUARD(blocking_operation)
+      working-directory: ruby-src
+      run: |
+        sed -i '/RB_GC_GUARD(blocking_operation)/d' scheduler.c
+        echo "--- scheduler.c around the change ---"
+        grep -n "blocking_operation\|funcall\|RB_GC" scheduler.c | tail -20
 
-    - name: Build + install CASE_A
+    - name: Configure + build BUGGY
       working-directory: ruby-src
-      run: make -j$(nproc) && make install
+      run: |
+        ./configure --prefix=$HOME/.rubies/ruby-buggy --enable-shared \
+          --disable-install-doc cppflags="-DENABLE_PATH_CHECK=0"
+        make -j$(nproc) && make install
 
-    # ── CASE_C: NULL + O0 on function ─────────────────────────────────────
-    - name: Patch to CASE_C — CASE_A + __attribute__((optimize("O0")))
+    # ── CASE_C: BUGGY + O0 on function ────────────────────────────────────
+    - name: Patch CASE_C — BUGGY + __attribute__((optimize("O0"))) on function
       working-directory: ruby-src
       run: |
-        git checkout scheduler.c
-        sed -i 's/rb_fiber_scheduler_blocking_operation_t \*operation = get_blocking_operation(blocking_operation);/rb_fiber_scheduler_blocking_operation_t *operation = NULL;/' scheduler.c
         sed -i 's/^VALUE rb_fiber_scheduler_blocking_operation_wait(/__attribute__((optimize("O0"))) VALUE rb_fiber_scheduler_blocking_operation_wait(/' scheduler.c
-        grep -n 'optimize\|operation = NULL' scheduler.c | head -5
-        ./configure --prefix=$HOME/.rubies/ruby-caseC --enable-shared --disable-install-doc cppflags="-DENABLE_PATH_CHECK=0"
+        grep -n 'optimize\|rb_fiber_scheduler_blocking_operation_wait' scheduler.c | head -5
 
-    - name: Build + install CASE_C
+    - name: Configure + build CASE_C (BUGGY + O0)
       working-directory: ruby-src
-      run: make -j$(nproc) && make install
+      run: |
+        ./configure --prefix=$HOME/.rubies/ruby-caseC --enable-shared \
+          --disable-install-doc cppflags="-DENABLE_PATH_CHECK=0"
+        make -j$(nproc) && make install
 
-    # ── Disassemble both ───────────────────────────────────────────────────
-    - name: Disassemble CASE_A and CASE_C
+    # ── Disassemble all three ──────────────────────────────────────────────
+    - name: Disassemble rb_fiber_scheduler_blocking_operation_wait in all builds
       run: |
-        echo "=== CASE_A (O3, operation=NULL) ==="
-        objdump -d --no-show-raw-insn -M intel $HOME/.rubies/ruby-caseA/lib/libruby.so.4.1 \
-          | awk '/^[0-9a-f]+ <rb_fiber_scheduler_blocking_operation_wait>:/{found=1} found{print} found && /^$/{exit}' \
-          > /tmp/case_a.asm
-        cat /tmp/case_a.asm
-        echo ""
-        echo "=== CASE_C (O0 on fn, operation=NULL) ==="
-        objdump -d --no-show-raw-insn -M intel $HOME/.rubies/ruby-caseC/lib/libruby.so.4.1 \
-          | awk '/^[0-9a-f]+ <rb_fiber_scheduler_blocking_operation_wait>:/{found=1} found{print} found && /^$/{exit}' \
-          > /tmp/case_c.asm
-        cat /tmp/case_c.asm
-
-    - name: Upload disassembly artifacts
+        for build in fixed buggy caseC; do
+          echo "=== $build ==="
+          objdump -d --no-show-raw-insn -M intel \
+            $HOME/.rubies/ruby-${build}/lib/libruby.so.4.1 \
+            | awk '/^[0-9a-f]+ <rb_fiber_scheduler_blocking_operation_wait>:/{found=1} found{print} found && /^$/{exit}' \
+            > /tmp/${build}.asm
+          wc -l /tmp/${build}.asm
+        done
+
+    - name: Upload disassembly
       uses: actions/upload-artifact@v4
       if: always()
       with:
-        name: disassembly-A-vs-C
-        path: /tmp/case_*.asm
+        name: disassembly-fixed-buggy-caseC
+        path: /tmp/*.asm
+
+    # ── Build io-event extension and run tests ─────────────────────────────
+    - name: Build io-event extension (FIXED Ruby)
+      run: |
+        export PATH=$HOME/.rubies/ruby-fixed/bin:$PATH
+        gem install sus --no-document
+        ruby ext/io/event/extconf.rb
+        make -C ext/io/event
+
+    - name: Run worker_pool test with FIXED Ruby
+      id: test_fixed
+      run: |
+        export PATH=$HOME/.rubies/ruby-fixed/bin:$PATH
+        ruby -I lib -rsus test/io/event/worker_pool.rb 2>&1 | tail -5
+        echo "exit:$?"
+      continue-on-error: true
 
-    # ── Build io-event extension for each Ruby and test ───────────────────
-    - name: Build io-event for CASE_A
+    - name: Build io-event extension (BUGGY Ruby)
       run: |
-        export PATH=$HOME/.rubies/ruby-caseA/bin:$PATH
-        gem install bundler --no-document
-        bundle install
-        cd ext/io/event && ruby extconf.rb && make
+        export PATH=$HOME/.rubies/ruby-buggy/bin:$PATH
+        gem install sus --no-document
+        ruby ext/io/event/extconf.rb
+        make -C ext/io/event
 
-    - name: Run worker_pool test CASE_A (expect crash)
-      id: test_caseA
+    - name: Run worker_pool test with BUGGY Ruby (expect crash)
+      id: test_buggy
       run: |
-        export PATH=$HOME/.rubies/ruby-caseA/bin:$PATH
-        (bundle exec ruby test/io/event/worker_pool.rb 2>&1 | tail -10; echo "exit:$?") || true
+        export PATH=$HOME/.rubies/ruby-buggy/bin:$PATH
+        (ruby -I lib -rsus test/io/event/worker_pool.rb 2>&1 | tail -5; echo "exit:$?") || true
       continue-on-error: true
 
-    - name: Build io-event for CASE_C
+    - name: Build io-event extension (CASE_C Ruby)
       run: |
         export PATH=$HOME/.rubies/ruby-caseC/bin:$PATH
-        cd ext/io/event && ruby extconf.rb && make
+        gem install sus --no-document
+        ruby ext/io/event/extconf.rb
+        make -C ext/io/event
 
-    - name: Run worker_pool test CASE_C (does O0 fix the crash?)
+    - name: Run worker_pool test with CASE_C Ruby (does O0 fix the crash?)
       id: test_caseC
       run: |
         export PATH=$HOME/.rubies/ruby-caseC/bin:$PATH
-        (bundle exec ruby test/io/event/worker_pool.rb 2>&1 | tail -10; echo "exit:$?") || true
+        (ruby -I lib -rsus test/io/event/worker_pool.rb 2>&1 | tail -5; echo "exit:$?") || true
       continue-on-error: true
 
     - name: Summary
       if: always()
       run: |
-        echo "If CASE_A crashes and CASE_C passes -> optimization level is the cause"
+        echo "FIXED (RB_GC_GUARD present): should pass"
+        echo "BUGGY (no RB_GC_GUARD): should crash"
+        echo "CASE_C (BUGGY + O0 on fn): if passes → optimization is the cause"