[PATCH] [analyzer][Review request] Fix for PR18394.

Антон Ярцев anton.yartsev at gmail.com
Fri Feb 28 13:48:06 PST 2014


  Added the check for PreImplicitCall and the testcase.

  It looks like that only check for PreImplicitCall can be mirrored: purge nodes are handled in the first place, CallEnters are handled differently from ExprEngine::replayWithoutInlining - we check the successor for CallEnter, and the comparison of statements is specific to ExprEngine::replayWithoutInlining.

  Ok to commit?

Hi jordan_rose, zaks.anna, krememek,

http://llvm-reviews.chandlerc.com/D2899

CHANGE SINCE LAST DIFF
  http://llvm-reviews.chandlerc.com/D2899?vs=7413&id=7435#toc

Files:
  lib/StaticAnalyzer/Core/ExplodedGraph.cpp
  test/Analysis/NewDeleteLeaks-PR18394.cpp

Index: lib/StaticAnalyzer/Core/ExplodedGraph.cpp
===================================================================
--- lib/StaticAnalyzer/Core/ExplodedGraph.cpp
+++ lib/StaticAnalyzer/Core/ExplodedGraph.cpp
@@ -90,8 +90,8 @@
   // (7) The LocationContext is the same as the predecessor.
   // (8) Expressions that are *not* lvalue expressions.
   // (9) The PostStmt isn't for a non-consumed Stmt or Expr.
-  // (10) The successor is not a CallExpr StmtPoint (so that we would
-  //      be able to find it when retrying a call with no inlining).
+  // (10) The successor is neither a CallExpr StmtPoint nor a CallEnter (so that
+  //      we would be able to find it when retrying a call with no inlining).
   // FIXME: It may be safe to reclaim PreCall and PostCall nodes as well.
 
   // Conditions 1 and 2.
@@ -152,6 +152,12 @@
   if (Optional<StmtPoint> SP = SuccLoc.getAs<StmtPoint>())
     if (CallEvent::isCallStmt(SP->getStmt()))
       return false;
+
+  // Condition 10, continuation.
+  // Note: the retry can not be started from a PreImplicitCall node (see logic
+  // of ExprEngine::replayWithoutInlining()).
+  if (SuccLoc.getAs<CallEnter>() && !progPoint.getAs<PreImplicitCall>())
+    return false;
 
   return true;
 }
Index: test/Analysis/NewDeleteLeaks-PR18394.cpp
===================================================================
--- test/Analysis/NewDeleteLeaks-PR18394.cpp
+++ test/Analysis/NewDeleteLeaks-PR18394.cpp
@@ -0,0 +1,16 @@
+// RUN: %clang_cc1 -analyzer-config graph-trim-interval=1 -analyzer-max-loop 1 -analyze -analyzer-checker=alpha.cplusplus.NewDeleteLeaks -verify %s
+// expected-no-diagnostics
+
+class A {
+public:
+  void f() {};
+  ~A() {
+    for (int i=0; i<3; i++)
+      f();
+  }
+};
+
+void error() {
+  A *a = new A();
+  delete a;
+}
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D2899.2.patch
Type: text/x-patch
Size: 1806 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20140228/16a63a32/attachment.bin>


More information about the cfe-commits mailing list