[clang] [clang] WIP: Warn on mismatched RequiresCapability attributes (PR #67520)

[Aaron Puchert via cfe-commits]

Timm =?utf-8?q?Bäder?= <tbaeder at redhat.com>,
Timm =?utf-8?q?Bäder?= <tbaeder at redhat.com>,
Timm =?utf-8?q?Bäder?= <tbaeder at redhat.com>,
Timm =?utf-8?q?Bäder?= <tbaeder at redhat.com>,
Timm =?utf-8?q?Bäder?= <tbaeder at redhat.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/67520 at github.com>


================
@@ -2263,6 +2265,28 @@ static bool neverReturns(const CFGBlock *B) {
   return false;
 }
 
+void ThreadSafetyAnalyzer::checkMismatchedFunctionAttrs(
+    const FunctionDecl *FD) {
+  FD = FD->getMostRecentDecl();
+
+  auto collectCapabilities = [&](const FunctionDecl *FD) {
+    SmallVector<CapabilityExpr> Args;
+    for (const auto *A : FD->specific_attrs<RequiresCapabilityAttr>()) {
+      for (const Expr *E : A->args())
+        Args.push_back(SxBuilder.translateAttrExpr(E, nullptr));
+    }
+    return Args;
+  };
+
+  auto FDArgs = collectCapabilities(FD);
+  for (const FunctionDecl *D = FD->getPreviousDecl(); D;
+       D = D->getPreviousDecl()) {
+    auto DArgs = collectCapabilities(D);
+    if (DArgs.size() != FDArgs.size())
+      Handler.handleAttributeMismatch(FD, D);
----------------
aaronpuchert wrote:

In #110523 we had a similar problem and decided to compare the sets one-by-one (search for `zip_longest`). The attributes do not have a meaningful order, but I don't see a benefit in allowing different orders. It's probably better for readability if we enforce consistent ordering.

But we can certainly also do a pairwise comparison to compute the difference sets and then emit warnings depending on which set contains `CapabilityExpr`s.

https://github.com/llvm/llvm-project/pull/67520