[clang] [NFC][analyzer] Refactor processCFGBlockEntrance (PR #215284)

[Donát Nagy via cfe-commits]

================
@@ -325,18 +325,12 @@ void CoreEngine::HandleBlockEdge(const BlockEdge &L, ExplodedNode *Pred) {
 
   // Call into the ExprEngine to process entering the CFGBlock.
   BlockEntrance BE(L.getSrc(), L.getDst(), Pred->getStackFrame());
-  ExplodedNodeSet DstNodes;
-  NodeBuilder Builder(Pred, DstNodes, ExprEng.getBuilderContext());
-  ExprEng.processCFGBlockEntrance(BE, Builder, Pred);
-
-  // Auto-generate a node.
-  if (!Builder.hasGeneratedNodes()) {
-    Builder.generateNode(BE, Pred->State, Pred);
-  }
+  ExplodedNode *Processed = ExprEng.processCFGBlockEntrance(BE, Pred);
 
   ExplodedNodeSet CheckerNodes;
-  for (auto *N : DstNodes) {
-    ExprEng.runCheckersForBlockEntrance(BE, N, CheckerNodes);
+
+  if (Processed && !Processed->isSink()) {
----------------
NagyDonat wrote:

In fact I realized that when a `makeNode`-like method makes a node that becomes a sink due to its `PosteriorlyOverconstrained` state, then the method should return `nullptr` instead of the sink node (because these sinks – with a state that does not represent a real possibility – are never useful).

This would simplify lots of engine code by getting rid of all the `&& !Node->isSink()` checks from situations where the node can be sink only due to its `PosteriorlyOverconstrained` state.

I also wrote a discourse topic about this and a related (but more complex) issue: https://discourse.llvm.org/t/simplifying-failure-modes-in-explodednode-creation/91542

https://github.com/llvm/llvm-project/pull/215284