================
@@ -3713,52 +3721,41 @@ 
ExprEngine::notifyCheckersOfPointerEscape(ProgramStateRef State,
 }
 
 /// evalBind - Handle the semantics of binding a value to a specific location.
-///  This method is used by evalStore and (soon) VisitDeclStmt, and others.
+///  This method is used by evalStore, VisitDeclStmt, and others.
 void ExprEngine::evalBind(ExplodedNodeSet &Dst, const Stmt *StoreE,
-                          ExplodedNode *Pred, SVal location, SVal Val,
+                          ExplodedNode *Pred, SVal Location, SVal Val,
                           bool AtDeclInit, const ProgramPoint *PP) {
+  // It may be a Loc, UnknownVal or perhaps UndefinedVal.
+  assert(!isa<NonLoc>(Location) && "evalBind location should not be NonLoc!");
+
   const LocationContext *LC = Pred->getLocationContext();
-  PostStmt PS(StoreE, LC);
+  PostStmt DefaultPP(StoreE, LC);
   if (!PP)
-    PP = &PS;
+    PP = &DefaultPP;
 
   // Do a previsit of the bind.
   ExplodedNodeSet CheckedSet;
-  getCheckerManager().runCheckersForBind(CheckedSet, Pred, location, Val,
+  getCheckerManager().runCheckersForBind(CheckedSet, Pred, Location, Val,
                                          StoreE, AtDeclInit, *this, *PP);
 
-  NodeBuilder Bldr(CheckedSet, Dst, *currBldrCtx);
+  for (ExplodedNode *PredI : CheckedSet) {
+    ProgramStateRef State = PredI->getState();
 
-  // If the location is not a 'Loc', it will already be handled by
-  // the checkers.  There is nothing left to do.
-  if (!isa<Loc>(location)) {
-    const ProgramPoint L = PostStore(StoreE, LC, /*Loc*/nullptr,
-                                     /*tag*/nullptr);
-    ProgramStateRef state = Pred->getState();
-    state = processPointerEscapedOnBind(state, location, Val, LC);
-    Bldr.generateNode(L, state, Pred);
-    return;
-  }
-
-  for (const auto PredI : CheckedSet) {
-    ProgramStateRef state = PredI->getState();
+    State = processPointerEscapedOnBind(State, Location, Val, LC);
 
-    state = processPointerEscapedOnBind(state, location, Val, LC);
-
-    // When binding the value, pass on the hint that this is a initialization.
-    // For initializations, we do not need to inform clients of region
-    // changes.
-    state = state->bindLoc(location.castAs<Loc>(), Val, LC,
-                           /* notifyChanges = */ !AtDeclInit);
+    if (auto AsLoc = Location.getAs<Loc>()) {
+      // When binding the value, pass on the hint that this is a
+      // initialization. For initializations, we do not need to inform clients
+      // of region changes.
+      State = State->bindLoc(*AsLoc, Val, LC, /*notifyChanges=*/!AtDeclInit);
+    }
 
     const MemRegion *LocReg = nullptr;
-    if (std::optional<loc::MemRegionVal> LocRegVal =
-            location.getAs<loc::MemRegionVal>()) {
+    if (auto LocRegVal = Location.getAs<loc::MemRegionVal>())
----------------
NagyDonat wrote:

> we shouldn't have two versions of `LocAsInteger`.

I guess you mean either "we shouldn't have two versions of `ConcreteInt` 
(`loc::` and `nonloc::`)" or "we shouldn't differentiate `loc::MemRegionVal` 
and `nonloc::LocAsInteger`".

I agree with this, although I don't have a concrete opinion for "what should 
the `SVal` hierarchy look like" yet.

My general feeling is that in an ideal world a pure `SVal` would represent a 
symbolic _value_ without type information (like `loc` vs `nonloc` distinctions 
or the bit width and signedness of a number), while typed values would be 
represented by `{SVal, QualType}` pairs. In the current system `SVal` "falls to 
the ground between two chairs": it contains some type information (which 
complicates things and introduces unwanted distinctions), but doesn't have 
_full accurate_ type information so we cannot fully rely on it.

Instead of embedding type info into the `SVal` I think we should gradually 
transition towards a model where the `SVal` represents an untyped 
("mathematical") value (e.g. `(int)0`, `(char)0` and a nullpointer are all 
identical) and the type information comes from the AST node that 
produces/accesses the symbolic value. (Or it is specified by a checker when it 
"summarizes" things happening inside a library functions that are not spelled 
out in the AST.)

I don't know how feasible is this and I don't have any concrete plans yet, but 
I'll perhaps think about this in the future.

https://github.com/llvm/llvm-project/pull/196313
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to