[llvm-branch-commits] [clang] [analyzer] Only bind aggregate lifetime sources in LifetimeModeling for annotated functions (PR #214824)

[Balázs Benics via llvm-branch-commits]

================
@@ -471,3 +471,35 @@ IntPtrArr return_array_field_not_yet_detected() {
   // expected-note at -2    {{Address of stack memory associated with local variable 'z' returned to caller}}
   // expected-warning at -3 {{address of stack memory associated with local variable 'z' returned}}
 }
+
+struct Hold {
+  int *ptr;
+};
+
+Hold retPtr(int &x) {
+  return Hold{&x};
+}
+// Even though there is a lifetime error in the function
+// UseAfterLifetimeEnd should not emit a warning for this
+// case since there is no annotation present in the code.
+// The warning present in the test comes from core.StackAddressEscape
+// checker.
+Hold return_by_val_no_ann() {
+  int num = 4;
+  return retPtr(num);
+  // expected-warning at -1 {{Address of stack memory associated with local variable 'num' returned to caller}}
+  // expected-note at -2    {{Address of stack memory associated with local variable 'num' returned to caller}}
+}
+
+int *unwrap(Hold i [[clang::lifetimebound]]) { return i.ptr; }
+
+// FIXME: If an annotated argument is a by-value struct then
+// Arg.getAsRegion() returns null for CompoundVal/LazyCompoundVal
+// and the dangling pointer will not be detected.
+int *arg_aggregate_lifetimebound() {
+  int local_num = 5;
+  Hold h{&local_num};
+  return unwrap(h);
----------------
steakhal wrote:

Why does this not work in the checker? I figured this was handled in the stacked PR 1.

https://github.com/llvm/llvm-project/pull/214824