[clang] [coroutines][coro_lifetimebound] Detect lifetime issues with lambda captures (PR #77066)

[Chuanqi Xu via cfe-commits]

================
@@ -7575,15 +7577,27 @@ static void visitLifetimeBoundArguments(IndirectLocalPath &Path, Expr *Call,
     Path.pop_back();
   };
 
-  if (ObjectArg && implicitObjectParamIsLifetimeBound(Callee))
-    VisitLifetimeBoundArg(Callee, ObjectArg);
-
   bool CheckCoroCall = false;
   if (const auto *RD = Callee->getReturnType()->getAsRecordDecl()) {
     CheckCoroCall = RD->hasAttr<CoroLifetimeBoundAttr>() &&
                     RD->hasAttr<CoroReturnTypeAttr>() &&
                     !Callee->hasAttr<CoroDisableLifetimeBoundAttr>();
   }
+
+  if (ObjectArg) {
+    bool CheckCoroObjArg = CheckCoroCall;
+    // Ignore `__promise.get_return_object()` as it is not lifetimebound.
+    if (CheckCoroObjArg && Callee->getDeclName().isIdentifier() &&
+        Callee->getName() == "get_return_object")
+      CheckCoroObjArg = false;
+    // Coroutine lambda objects with empty capture list are not lifetimebound.
+    if (auto *LE = dyn_cast<LambdaExpr>(ObjectArg->IgnoreImplicit());
+        LE && LE->captures().empty())
+      CheckCoroObjArg = false;
----------------
ChuanqiXu9 wrote:

> > I didn't get the logic here. Why it is not good to warn the undefined things?
> 
> Sorry, I wasn't clear here. I meant that if it is UB, we should warn for those cases too and remove this code path.
> 
> > I feel it is literally undefined. Since the spec doesn't say a lot about resumption/suspension about coroutines. Also the description of http://eel.is/c++draft/dcl.fct.def.coroutine#note-3 is vague too. It says it is likely to be an undefined behavior.
> 
> Yeah, this looks suspicious. At the same time, it's very similar to `delete this`, which (IIUC) is not UB as long as people are careful to not access `this` after doing it.
> 
> I have actually asked the CWG, but got no reply so far.

Got it. Then it might not be a UB (we didn't access `this` in the empty lambda) if we're comparing this case. While I don't object wait for response from CWG, I feel we can continue here. On the one hand, we're talking about the behaviors with an extension (`[[clang::coro_lifetime_bound]]`) which can be visited as a dialect. On the other hand, it is somewhat not user friendly to emit warnings in such cases.

https://github.com/llvm/llvm-project/pull/77066