Fix C++ function identity collisions in the benchmark pipeline

.github/workflows/function-identity.yml

@@ -0,0 +1,124 @@
+name: Function identity validation
+
+on:
+  push:
+    branches: [fix/cpp-function-identity-clean]
+  pull_request:
+    paths:
+      - 'decbench/**'
+      - 'scripts/**'
+      - 'tests/**'
+      - '.github/workflows/function-identity.yml'
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+env:
+  DECBENCH_NO_CACHE: '1'
+  IDENTITY_MAIN: ae070b885b4e005ed37510a2c3445af6284dc2b9
+  RUST_JOERN_REV: 9fd97f50885202596b125ebd2504e64910874bcb
+
+jobs:
+  regression:
+    runs-on: ubuntu-latest
+    timeout-minutes: 40
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          ref: ${{ github.event.pull_request.head.sha || github.sha }}
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+      - name: Record exact checkout
+        run: |
+          mkdir -p evidence
+          git rev-parse HEAD HEAD^{tree} | tee evidence/checkout.txt
+          git status --porcelain | tee evidence/status.txt
+      - name: Install compilers and Python dependencies
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y clang lld g++ gcc-arm-none-eabi g++-mingw-w64-x86-64
+          python -m pip install -e '.[dev]' r2pipe
+      - name: Build pinned Rust Joern
+        run: |
+          git clone https://github.com/Noelo-Lab/rust-joern "$RUNNER_TEMP/rust-joern"
+          git -C "$RUNNER_TEMP/rust-joern" checkout "$RUST_JOERN_REV"
+          cargo build --manifest-path "$RUNNER_TEMP/rust-joern/Cargo.toml" --release --locked
+          python -m pip install -e "$RUNNER_TEMP/rust-joern"
+          test -f "$RUNNER_TEMP/rust-joern/target/release/librust_joern.so"
+          echo "RUST_JOERN_LIBRARY=$RUNNER_TEMP/rust-joern/target/release/librust_joern.so" >> "$GITHUB_ENV"
+      - name: Record environment
+        run: |
+          python -m pip freeze > evidence/python-environment.txt
+          gcc --version > evidence/gcc.txt
+          arm-none-eabi-g++ --version > evidence/arm.txt
+          x86_64-w64-mingw32-g++ --version > evidence/mingw.txt
+          ld.lld --version > evidence/lld.txt
+      - name: Identity and consumer regressions
+        run: |
+          pytest -q --no-cov tests/test_cpp_identity*.py tests/test_cpp_support.py tests/test_pr91_review_boundaries.py tests/test_evalkit.py tests/test_materialized.py tests/test_report_extras.py tests/test_reeval_ged.py tests/test_arm_thumb_extraction.py --junitxml=evidence/focused.xml
+      - name: Lint review changes
+        run: |
+          ruff check decbench/utils/function_identity.py decbench/evalkit/export.py decbench/evalkit/ingest.py decbench/evalkit/resolve.py decbench/models/decompilation.py decbench/scoring/report_extras.py scripts/run_benchmark.py scripts/rebuild_function_data.py scripts/reeval_ged.py scripts/validate_function_identity.py tests/test_pr91_review_boundaries.py tests/test_cpp_identity_pe_probe.py
+          black --check decbench/utils/function_identity.py decbench/evalkit/export.py decbench/evalkit/ingest.py decbench/evalkit/resolve.py decbench/models/decompilation.py decbench/scoring/report_extras.py scripts/run_benchmark.py scripts/rebuild_function_data.py scripts/reeval_ged.py scripts/validate_function_identity.py tests/test_pr91_review_boundaries.py tests/test_cpp_identity_pe_probe.py
+          mypy --follow-imports=silent decbench/utils/function_identity.py decbench/evalkit/export.py decbench/evalkit/ingest.py decbench/evalkit/resolve.py scripts/validate_function_identity.py
+      - name: Full suite differential against pinned main
+        run: |
+          git worktree add --detach "$RUNNER_TEMP/identity-main" "$IDENTITY_MAIN"
+          set +e
+          (cd "$RUNNER_TEMP/identity-main" && PYTHONPATH="$RUNNER_TEMP/identity-main" python -m pytest -q --no-cov --junitxml="$GITHUB_WORKSPACE/evidence/main.xml") > evidence/main.log 2>&1
+          main_status=$?
+          PYTHONPATH="$GITHUB_WORKSPACE" python -m pytest -q --no-cov --junitxml=evidence/candidate.xml > evidence/candidate.log 2>&1
+          candidate_status=$?
+          set -e
+          printf 'main=%s\ncandidate=%s\n' "$main_status" "$candidate_status" | tee evidence/pytest-status.txt
+          test "$main_status" -le 1
+          test "$candidate_status" -le 1
+          python scripts/validate_function_identity.py compare evidence/main.xml evidence/candidate.xml --output evidence/differential.json
+      - uses: actions/upload-artifact@v4
+        if: always()
+        with:
+          name: function-identity-${{ github.sha }}
+          path: evidence/
+
+  leveldb:
+    runs-on: ubuntu-latest
+    timeout-minutes: 35
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          ref: ${{ github.event.pull_request.head.sha || github.sha }}
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+      - uses: actions/setup-java@v4
+        with:
+          distribution: temurin
+          java-version: '21'
+      - name: Install dependencies and Ghidra
+        run: |
+          python -m pip install -e . pyghidra
+          curl -L --fail -o "$RUNNER_TEMP/ghidra.zip" https://github.com/NationalSecurityAgency/ghidra/releases/download/Ghidra_12.0.4_build/ghidra_12.0.4_PUBLIC_20260303.zip
+          echo "c3b458661d69e26e203d739c0c82d143cc8a4a29d9e571f099c2cf4bda62a120  $RUNNER_TEMP/ghidra.zip" | sha256sum -c -
+          unzip -q "$RUNNER_TEMP/ghidra.zip" -d "$RUNNER_TEMP"
+          echo "GHIDRA_INSTALL_DIR=$RUNNER_TEMP/ghidra_12.0.4_PUBLIC" >> "$GITHUB_ENV"
+      - name: Build pinned LevelDB
+        run: |
+          git clone https://github.com/google/leveldb "$RUNNER_TEMP/leveldb"
+          git -C "$RUNNER_TEMP/leveldb" checkout 99b3c03b3284f5886f9ef9a4ef703d57373e61be
+          cmake -S "$RUNNER_TEMP/leveldb" -B "$RUNNER_TEMP/leveldb/build" -DCMAKE_BUILD_TYPE= -DCMAKE_CXX_FLAGS='-O0 -g -fno-builtin' -DLEVELDB_BUILD_TESTS=OFF -DLEVELDB_BUILD_BENCHMARKS=OFF -DBUILD_SHARED_LIBS=ON
+          cmake --build "$RUNNER_TEMP/leveldb/build" -j2
+      - name: Stripped cross-decompiler identity
+        env:
+          _JAVA_OPTIONS: '-Xmx4g -Djava.awt.headless=true'
+        run: |
+          mkdir -p evidence
+          git rev-parse HEAD HEAD^{tree} > evidence/checkout.txt
+          python scripts/validate_function_identity.py leveldb "$RUNNER_TEMP/leveldb/build/libleveldb.so.1.23.0" --output evidence/leveldb.json
+      - uses: actions/upload-artifact@v4
+        if: always()
+        with:
+          name: leveldb-identity-${{ github.sha }}
+          path: evidence/

decbench/decompilers/raw/angr_raw.py

@@ -32,6 +32,7 @@
     LineMapping,
     VariableInfo,
 )
+from decbench.utils.function_identity import insert_function
 
 _l = logging.getLogger(__name__)
 
@@ -141,7 +142,7 @@ def _dump() -> None:
                     _l.debug("angr-raw: failed to decompile %s: %s", func_name, e)
 
                 if func_result is not None:
-                    decompiled_functions[func_name] = func_result
+                    insert_function(decompiled_functions, func_result)
                 else:
                     failed_functions.append(func_name)
                 _dump()

decbench/decompilers/raw/binja_raw.py

@@ -35,6 +35,7 @@
     LineMapping,
     VariableInfo,
 )
+from decbench.utils.function_identity import insert_function
 
 _l = logging.getLogger(__name__)
 
@@ -149,8 +150,8 @@ def _dump() -> None:
             bv = self._load(binary_path)
             enumerated = self._enumerate(bv, elf_base, text_range, addr_targets)
             if functions is not None:
-                requested = {n for (n, _a) in functions}
-                enumerated = [(n, a) for (n, a) in enumerated if n in requested]
+                requested_addrs = {a for (_n, a) in functions}
+                enumerated = [(n, a) for (n, a) in enumerated if a in requested_addrs]
             enumerated = common.narrow_to_source(
                 enumerated,
                 function_names,
@@ -170,7 +171,7 @@ def _dump() -> None:
                     except Exception as e:  # noqa: BLE001
                         _l.debug("binja-raw: failed to decompile %s: %s", func_name, e)
                 if func_result is not None:
-                    decompiled_functions[func_name] = func_result
+                    insert_function(decompiled_functions, func_result)
                 else:
                     failed_functions.append(func_name)
                 _dump()

decbench/decompilers/raw/dewolf_raw.py

@@ -61,6 +61,7 @@
     FunctionDecompilation,
     VariableInfo,
 )
+from decbench.utils.function_identity import insert_function
 
 _l = logging.getLogger(__name__)
 
@@ -291,7 +292,7 @@ def _consume(addrs_arg: str) -> None:
                             metadata=common.extract_metrics(code),
                         )
                         with lock:
-                            decompiled_functions[name] = fd
+                            insert_function(decompiled_functions, fd)
                             _dump()
                     elif kind == "fail":
                         with lock:

decbench/decompilers/raw/ghidra_raw.py

@@ -44,6 +44,7 @@
     LineMapping,
     VariableInfo,
 )
+from decbench.utils.function_identity import insert_function
 
 _l = logging.getLogger(__name__)
 
@@ -264,11 +265,11 @@ def _dump() -> None:
                 program = flat.getCurrentProgram()
                 image_base = int(program.getImageBase().getOffset())
 
-                requested = {n for (n, _a) in functions} if functions is not None else None
+                requested_addrs = {a for (_n, a) in functions} if functions is not None else None
 
                 enumerated = self._enumerate(program, elf_base, text_range, addr_targets)
-                if requested is not None:
-                    enumerated = [(n, a) for (n, a) in enumerated if n in requested]
+                if requested_addrs is not None:
+                    enumerated = [(n, a) for (n, a) in enumerated if a in requested_addrs]
 
                 enumerated = common.narrow_to_source(
                     enumerated,
@@ -281,11 +282,11 @@ def _dump() -> None:
                 ifc.openProgram(program)
                 monitor = ConsoleTaskMonitor()
                 timeout_s = int(self.config.function_timeout_seconds)
-                by_name = self._functions_by_name(program)
+                by_address = self._functions_by_address(program, elf_base)
 
                 for func_name, file_addr in enumerated:
                     func_result = None
-                    g_func = by_name.get(func_name)
+                    g_func = by_address.get(file_addr)
                     if g_func is not None:
                         try:
                             func_result = self._decompile_one(
@@ -305,7 +306,7 @@ def _dump() -> None:
                                 e,
                             )
                     if func_result is not None:
-                        decompiled_functions[func_name] = func_result
+                        insert_function(decompiled_functions, func_result)
                     else:
                         failed_functions.append(func_name)
                     _dump()
@@ -370,12 +371,15 @@ def _enumerate(
         return sorted(out, key=lambda x: x[1])
 
     @staticmethod
-    def _functions_by_name(program: Any) -> dict[str, Any]:
-        """Map (last-seen) function name -> Ghidra Function object."""
-        out: dict[str, Any] = {}
+    def _functions_by_address(program: Any, elf_base: int) -> dict[int, Any]:
+        """Map binary-local function address -> Ghidra Function object."""
+        image_base = int(program.getImageBase().getOffset())
+        out: dict[int, Any] = {}
         fm = program.getFunctionManager()
         for g_func in fm.getFunctions(True):
-            out[g_func.getName()] = g_func
+            offset = int(g_func.getEntryPoint().getOffset())
+            file_addr = (offset - image_base) + elf_base
+            out[file_addr] = g_func
         return out
 
     def _decompile_one(

decbench/decompilers/raw/ida_raw.py

@@ -34,6 +34,7 @@
     LineMapping,
     VariableInfo,
 )
+from decbench.utils.function_identity import insert_function
 
 _l = logging.getLogger(__name__)
 
@@ -158,8 +159,12 @@ def _dump() -> None:
             try:
                 enumerated = self._enumerate(elf_base, text_range, addr_targets)
                 if functions is not None:
-                    requested = {n for (n, _a) in functions}
-                    enumerated = [(n, a) for (n, a) in enumerated if n in requested]
+                    requested_addrs = {address for (_name, address) in functions}
+                    enumerated = [
+                        (name, address)
+                        for name, address in enumerated
+                        if address in requested_addrs
+                    ]
                 enumerated = common.narrow_to_source(
                     enumerated,
                     function_names,
@@ -173,7 +178,7 @@ def _dump() -> None:
                     except Exception as e:  # noqa: BLE001
                         _l.debug("ida-raw: failed to decompile %s: %s", func_name, e)
                     if func_result is not None:
-                        decompiled_functions[func_name] = func_result
+                        insert_function(decompiled_functions, func_result)
                     else:
                         failed_functions.append(func_name)
                     _dump()

decbench/decompilers/raw/kuna_raw.py

@@ -58,6 +58,7 @@
     FunctionDecompilation,
     VariableInfo,
 )
+from decbench.utils.function_identity import insert_function
 
 _l = logging.getLogger(__name__)
 
@@ -158,32 +159,36 @@ def _dump() -> None:
             _l.error("kuna-raw failed on %s: %s", binary_path, e)
             return self._error_result(binary_path, start, str(e))
 
-        records = {str(r.get("name") or ""): r for r in self._records(payload)}
+        records_by_address = self._records_by_address(payload)
         enumerated = sorted(
             (
-                (n, int(r.get("address") or 0))
-                for n, r in records.items()
+                (str(record.get("name") or ""), address)
+                for address, record in records_by_address.items()
                 if not common.should_skip_function(
-                    n, int(r.get("address") or 0), text_range, addr_targets
+                    str(record.get("name") or ""), address, text_range, addr_targets
                 )
             ),
             key=lambda x: x[1],
         )
         if functions is not None:
-            requested = {n for (n, _a) in functions}
-            enumerated = [(n, a) for (n, a) in enumerated if n in requested]
+            requested_addrs = {address for (_name, address) in functions}
+            enumerated = [
+                (name, address)
+                for name, address in enumerated
+                if address in requested_addrs
+            ]
         enumerated = common.narrow_to_source(
             enumerated, function_names, backend="kuna", binary_name=binary_path.name
         )
 
         for func_name, file_addr in enumerated:
             fd = None
             try:
-                fd = self._build_function(records[func_name], func_name, file_addr)
+                fd = self._build_function(records_by_address[file_addr], func_name, file_addr)
             except Exception as e:  # noqa: BLE001
                 _l.debug("kuna-raw: assembling %s failed: %s", func_name, e)
             if fd is not None:
-                decompiled[func_name] = fd
+                insert_function(decompiled, fd)
             else:
                 failed.append(func_name)
             _dump()
@@ -285,6 +290,21 @@ def _records(payload: Any) -> list[dict[str, Any]]:
             return list(payload.get("functions") or [])
         return payload if isinstance(payload, list) else []
 
+    @classmethod
+    def _records_by_address(cls, payload: Any) -> dict[int, dict[str, Any]]:
+        """Index kuna records by binary-local address without collapsing names."""
+        out: dict[int, dict[str, Any]] = {}
+        for record in cls._records(payload):
+            value = record.get("address")
+            if value is None:
+                continue
+            try:
+                address = int(value)
+            except (TypeError, ValueError):
+                continue
+            out[address] = record
+        return out
+
     def _build_function(
         self, rec: dict[str, Any], name: str, file_addr: int
     ) -> FunctionDecompilation | None:
@@ -399,4 +419,4 @@ def _error_result(
                 failed_functions=["all"],
                 extra={"error": err, "backend": "kuna", "via": "raw"},
             ),
-        )
+        )

decbench/evalkit/export.py

@@ -123,13 +123,15 @@ def _resolve_manifest(
             failures.append(f"{project}/{opt}/{stem}: {e}")
             continue
         stems = resolve.source_stems(tree, opt, project)
-        addr_by_name = resolve.name_to_addr(path, names=names, stems=stems or None)
-        for fn in sorted(names - set(addr_by_name)):
+        from decbench.utils.function_identity import unresolved_manifest_keys
+
+        selected = resolve.manifest_functions(path, names=names, stems=stems or None)
+        for fn in sorted(unresolved_manifest_keys(selected, names)):
             failures.append(
                 f"{project}/{opt}/{stem}::{fn}: no DWARF address "
                 f"(decl-file-filtered subprogram walk over {path.name})"
             )
-        funcs = {fn: addr_by_name[fn] for fn in sorted(names & set(addr_by_name))}
+        funcs = {key: address for address, key in selected.items()}
         if funcs:
             bins.append(_KitBinary(project, opt, stem, path, funcs))
         elif allow_unresolved: