[clang] [LifetimeSafety] Warn on implicit this lifetimebound violations (PR #196926)
Gábor Horváth via cfe-commits
cfe-commits at lists.llvm.org
Mon May 11 04:24:17 PDT 2026
================
@@ -72,21 +72,29 @@ getLifetimeBoundAttrFromFunctionType(const TypeSourceInfo &TSI) {
return nullptr;
}
-bool implicitObjectParamIsLifetimeBound(const FunctionDecl *FD) {
+const LifetimeBoundAttr *
+getImplicitObjectParamLifetimeBoundAttr(const FunctionDecl *FD) {
FD = getDeclWithMergedLifetimeBoundAttrs(FD);
// Attribute merging doesn't work well with attributes on function types (like
// 'this' param). We need to check all redeclarations.
- auto CheckRedecls = [](const FunctionDecl *F) {
- return llvm::any_of(F->redecls(), [](const FunctionDecl *Redecl) {
- const TypeSourceInfo *TSI = Redecl->getTypeSourceInfo();
- return TSI && getLifetimeBoundAttrFromFunctionType(*TSI);
- });
+ auto CheckRedecls = [](const FunctionDecl *F) -> const LifetimeBoundAttr * {
+ for (const FunctionDecl *Redecl : F->redecls()) {
+ if (const TypeSourceInfo *TSI = Redecl->getTypeSourceInfo())
+ if (const auto *Attr = getLifetimeBoundAttrFromFunctionType(*TSI))
+ return Attr;
+ }
+ return nullptr;
};
- if (CheckRedecls(FD))
- return true;
- if (const FunctionDecl *Pattern = FD->getTemplateInstantiationPattern();
- Pattern && CheckRedecls(Pattern))
+ if (const auto *Attr = CheckRedecls(FD))
+ return Attr;
+ if (const FunctionDecl *Pattern = FD->getTemplateInstantiationPattern())
+ return CheckRedecls(Pattern);
+ return nullptr;
+}
+
+bool implicitObjectParamIsLifetimeBound(const FunctionDecl *FD) {
+ if (getImplicitObjectParamLifetimeBoundAttr(FD))
return true;
return isNormalAssignmentOperator(FD);
----------------
Xazax-hun wrote:
We do not actually do the verification for assignment operators. I think this is fine, as this was not explicitly spelled by the user. But I wonder if we should have a potentially separate warning when an assignment operator returns an unexpected loan. This could be a separate PR.
https://github.com/llvm/llvm-project/pull/196926
More information about the cfe-commits
mailing list