[clang] [analyzer] Splitting TaintPropagation checker into reporting and mode… (PR #98157)

Donát Nagy via cfe-commits cfe-commits at lists.llvm.org
Wed Jul 10 03:17:51 PDT 2024


================
@@ -1122,10 +1131,20 @@ void GenericTaintChecker::taintUnsafeSocketProtocol(const CallEvent &Call,
 }
 
 /// Checker registration
-void ento::registerGenericTaintChecker(CheckerManager &Mgr) {
+void ento::registerTaintPropagationChecker(CheckerManager &Mgr) {
   Mgr.registerChecker<GenericTaintChecker>();
 }
 
+bool ento::shouldRegisterTaintPropagationChecker(const CheckerManager &mgr) {
+  return true;
+}
+
+void ento::registerGenericTaintChecker(CheckerManager &Mgr) {
+  GenericTaintChecker *checker = Mgr.getChecker<GenericTaintChecker>();
+  checker->isTaintReporterCheckerEnabled = true;
+  checker->reporterCheckerName = Mgr.getCurrentCheckerName();
----------------
NagyDonat wrote:

We only use that `unique_ptr` has a null / empty state in addition to representing the "real" `BugType` values. However, `std::optional` would be completely sufficient for this goal (here and AFAIK in every checker that uses `unique_ptr`s to juggle multiple `BugType`s).

If we really want, in _this_ checker we could use a "bare" `BugType` data member if we leave it uninitialized in the constructor of the checker class (or initialize it with dummy values) and then unconditionally overwrite it with the "real" `BugType` in this `register...` function. (Other checkers have lazy initialization patterns that check the nullness of the `BugType`.)

https://github.com/llvm/llvm-project/pull/98157


More information about the cfe-commits mailing list