[clang] [Lifetime Safety] Highlight lifetimebound calls in alias chain diagnostics (PR #206337)

via cfe-commits cfe-commits at lists.llvm.org
Sun Jul 12 23:45:48 PDT 2026


================
@@ -664,20 +664,66 @@ class LifetimeSafetySemaHelperImpl : public LifetimeSafetySemaHelper {
     return CurrExpr->getSourceRange() != LastExpr->getSourceRange();
   }
 
+  bool shouldCarryHiddenLifetimeBound(const Expr *HiddenExpr,
----------------
iitianpushkar wrote:

No, we cannot directly match them because even after `IgnoreImpCasts()` or `IgnoreUnlessSpelledInSource()`, those can still be different Expr.

For e.g. :
```cpp
template <typename T> struct [[gsl::Pointer]] Pointer {
  Pointer(const T &bar [[clang::lifetimebound]]);
};

void nested_local_pointer() {
  Pointer<Pointer<Pointer<Bar>>> ppp;
  Pointer<Pointer<Bar>> pp;
  Pointer<Bar> p;
  {
    Bar v;
    p = Pointer(v);     // expected-warning {{local variable 'v' does not live long enough}}
    pp = Pointer(p);    // expected-note {{local variable 'p' aliases the storage of local variable 'v' because parameter 'bar' is marked 'lifetimebound'}}
    ppp = Pointer(pp);  // expected-note {{local variable 'pp' aliases the storage of local variable 'v'}}
  }                     // expected-note {{local variable 'v' is destroyed here}}
  use(***ppp);          // expected-note {{later used here}}
}
```

For `pp = Pointer(p)`, the hidden lifetimebound step can correspond to the constructor expression Pointer(p) while the visible expression being explained is the `argument p`.
That is why the current check compares source ranges instead, after normalizing with `IgnoreImpCasts`, it accepts either the same source range or a hidden expression range that directly wraps the visible expression range.

https://github.com/llvm/llvm-project/pull/206337


More information about the cfe-commits mailing list