Merge pull request #2499 from ankitbuilds99/fix/lint-mem-gate-warning-suffix

fix(ci): parse clang-tidy analyzer warning suffix

scripts/lint-mem-gate.py

@@ -47,7 +47,15 @@ WHITELIST = Path(os.environ.get(
 MIN_WHY = 120
 MIN_TRIED = 40
 
-FINDING_RE = re.compile(r"^(?P<file>[^:]+):(?P<line>\d+):\d+:\s+(?:error|warning):\s+(?P<msg>.*?)\s*\[(?P<check>[\w.-]+)\]\s*$")
+# With WarningsAsErrors enabled, clang-tidy appends the escalation reason to
+# analyzer checks, for example:
+#   [clang-analyzer-core.NullDereference,-warnings-as-errors]
+# Keep the check name separate from that suffix so the gate still matches the
+# same whitelist key instead of silently dropping the finding.
+FINDING_RE = re.compile(
+    r"^(?P<file>[^:]+):(?P<line>\d+):\d+:\s+(?:error|warning):\s+"
+    r"(?P<msg>.*?)\s*\[(?P<check>[\w.-]+)(?:,-warnings-as-errors)?\]\s*$"
+)
 
 
 def strip_noise(text):

scripts/test.sh

@@ -295,6 +295,9 @@ bash "$ROOT/tests/test_activation_diagnostic_contract.sh"
 echo "=== Step 0e3: security gate fail-closed contract ==="
 bash "$ROOT/tests/test_security_gate_fail_closed.sh"
 
+echo "=== Step 0e4: memory-analyzer gate warning-suffix contract (#2442) ==="
+bash "$ROOT/tests/test_lint_mem_gate.sh"
+
 echo "=== Step 0f: tree-sitter runtime Makefile dependencies ==="
 bash "$ROOT/tests/test_makefile_ts_runtime_dependencies.sh"
 

tests/test_lint_mem_gate.sh

@@ -0,0 +1,105 @@
+#!/usr/bin/env bash
+# Regression contract for clang-tidy's `,-warnings-as-errors` diagnostic suffix
+# (#2442). With WarningsAsErrors enabled, clang-tidy prints an analyzer check as
+# [clang-analyzer-<check>,-warnings-as-errors]; a gate that only accepts
+# [<check>] drops every such line and reports a clean gate. Both directions are
+# pinned, and every assertion is reported, so a regression names what broke:
+#   1. a suffixed finding WITH an argued whitelist entry is counted under its
+#      plain check name and suppressed (exit 0, "1 argued false positive");
+#   2. a suffixed finding WITHOUT an entry fails the gate (exit 1) and is
+#      reported under its plain check name. This is the negative assertion: a
+#      parser that drops the suffixed line reports clean (exit 0) here.
+# Reverting the suffix group in FINDING_RE (scripts/lint-mem-gate.py) turns
+# both red.
+#
+# The gate runs from inside the fixture directory with RELATIVE paths. On the
+# Windows leg python3 is a native MSYS2 build: MSYS converts path ARGUMENTS for
+# it, but never the diagnostic text on stdin, so an absolute /c/... path in a
+# finding would not resolve there, and a C:/... one would not parse (the file
+# field ends at the first colon).
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+GATE="$ROOT/scripts/lint-mem-gate.py"
+FIX="$(mktemp -d "${TMPDIR:-/tmp}/lint-mem-gate.XXXXXX")"
+trap 'rm -rf "$FIX"' EXIT
+
+failures=0
+fail() {
+    echo "FAIL: $*" >&2
+    failures=$((failures + 1))
+}
+
+show() {
+    sed 's/^/    | /' "$1" >&2
+}
+
+# A subshell, so the test's own shell never sits inside $FIX (Windows cannot
+# remove a directory that is a process's working directory).
+run_gate() {
+    (cd "$FIX" && LINT_MEM_WHITELIST=whitelist.txt python3 "$GATE" "$@")
+}
+
+cat > "$FIX/sample.c" <<'EOF'
+int sample(int *value) {
+    return *value;
+}
+
+int unlisted(int *value) {
+    return *value;
+}
+EOF
+
+# A native Windows python3 writes CRLF; strip it before the hash is reused.
+HASH="$(run_gate --hash sample.c sample | tr -d '\r' | awk '/^segment-sha256:/ {print $2}')"
+if [[ ! "$HASH" =~ ^[0-9a-f]{64}$ ]]; then
+    echo "FAIL: lint-mem-gate.py --hash sample.c sample printed no segment hash (got '$HASH')" >&2
+    exit 1
+fi
+
+cat > "$FIX/whitelist.txt" <<EOF
+## sample.c :: sample :: clang-analyzer-core.NullDereference
+segment-sha256: $HASH
+why: |
+  The test deliberately models a diagnostic that is suppressed by the fixture whitelist.
+  This text is long enough to exercise the gate's normal argument validation path.
+tried: |
+  The fixture is intentionally minimal and only verifies diagnostic parsing.
+EOF
+
+SUFFIX='[clang-analyzer-core.NullDereference,-warnings-as-errors]'
+
+# 1. Whitelisted: counted under the plain check name, then suppressed.
+rc=0
+printf '%s\n' "sample.c:2:12: warning: possible null dereference $SUFFIX" \
+    | run_gate > /dev/null 2> "$FIX/whitelisted.err" || rc=$?
+if [ "$rc" -ne 0 ]; then
+    fail "a whitelisted suffixed finding left the gate at exit $rc, expected 0"
+    show "$FIX/whitelisted.err"
+fi
+if ! grep -q 'memory gate clean (1 argued false positive' "$FIX/whitelisted.err"; then
+    fail "a whitelisted suffixed finding was not counted as 1 argued false positive"
+    show "$FIX/whitelisted.err"
+fi
+
+# 2. Not whitelisted: the gate must fail and name the finding.
+rc=0
+printf '%s\n' "sample.c:6:12: warning: possible null dereference $SUFFIX" \
+    | run_gate > /dev/null 2> "$FIX/unlisted.err" || rc=$?
+if [ "$rc" -ne 1 ]; then
+    fail "an unwhitelisted suffixed finding left the gate at exit $rc, expected 1" \
+        "(exit 0 = the suffixed line was parsed as no finding, #2442)"
+    show "$FIX/unlisted.err"
+fi
+if ! grep -qF 'sample.c:6: possible null dereference [clang-analyzer-core.NullDereference]' "$FIX/unlisted.err" ||
+    ! grep -qF 'no whitelist entry for unlisted()' "$FIX/unlisted.err" ||
+    ! grep -qF '1 finding(s) unaccounted for' "$FIX/unlisted.err"; then
+    fail "the gate did not report the unwhitelisted suffixed finding under its plain check name"
+    show "$FIX/unlisted.err"
+fi
+
+if [ "$failures" -ne 0 ]; then
+    echo "lint-mem-gate warning suffix contract: $failures assertion(s) failed" >&2
+    exit 1
+fi
+echo 'lint-mem-gate warning suffix contract passed (whitelisted + unwhitelisted suffixed findings)'