[clang] [clang][analyzer] Split NullDereferenceChecker into a modeling and checker part (PR #122139)

[Donát Nagy via cfe-commits]

================
@@ -351,12 +356,30 @@ void DereferenceChecker::checkBind(SVal L, SVal V, const Stmt *S,
   C.addTransition(State, this);
 }
 
-void ento::registerDereferenceChecker(CheckerManager &mgr) {
-  auto *Chk = mgr.registerChecker<DereferenceChecker>();
-  Chk->SuppressAddressSpaces = mgr.getAnalyzerOptions().getCheckerBooleanOption(
-      mgr.getCurrentCheckerName(), "SuppressAddressSpaces");
+void ento::registerDereferenceModeling(CheckerManager &Mgr) {
+  Mgr.registerChecker<DereferenceChecker>();
+}
+
+bool ento::shouldRegisterDereferenceModeling(const CheckerManager &Mgr) {
+  return true;
+}
+
+void ento::registerNullDereferenceChecker(CheckerManager &Mgr) {
+  auto *Chk = Mgr.getChecker<DereferenceChecker>();
+  Chk->CheckNullDereference = true;
+  Chk->SuppressAddressSpaces = Mgr.getAnalyzerOptions().getCheckerBooleanOption(
+      Mgr.getCurrentCheckerName(), "SuppressAddressSpaces");
+  Chk->BT_Null.reset(new BugType(Mgr.getCurrentCheckerName(),
+                                 "Dereference of null pointer",
+                                 categories::LogicError));
+  Chk->BT_Undef.reset(new BugType(Mgr.getCurrentCheckerName(),
+                                  "Dereference of undefined pointer value",
+                                  categories::LogicError));
+  Chk->BT_Label.reset(new BugType(Mgr.getCurrentCheckerName(),
+                                  "Dereference of the address of a label",
----------------
NagyDonat wrote:

It's a pity that this commit reintroduces this dynamic initialization boilerplate. We should really introduce a clear framework for supporting bug types in a checker class that implements multiple sub-checkers...

(This is just a general long-term complaint, your change is probably the best possible short-term solution...)

https://github.com/llvm/llvm-project/pull/122139