[clang] nonblocking/nonallocating attributes (was: nolock/noalloc) (PR #84983)

[via cfe-commits]

================
@@ -18324,6 +18324,47 @@ bool Sema::CheckOverridingFunctionAttributes(const CXXMethodDecl *New,
     return true;
   }
 
+  // Virtual overrides: check for matching effects.
+  const auto OldFX = Old->getFunctionEffects();
+  const auto NewFX = New->getFunctionEffects();
+
+  if (OldFX != NewFX) {
+    const auto Diffs = FunctionEffectSet::differences(OldFX, NewFX);
+    bool AnyDiags = false;
+
+    for (const auto &Item : Diffs) {
+      const FunctionEffect &Effect = Item.first;
+      const bool Adding = Item.second;
+      switch (Effect.diagnoseMethodOverride(Adding, *Old, OldFX, *New, NewFX)) {
+      case FunctionEffect::OverrideResult::Ignore:
+        break;
+      case FunctionEffect::OverrideResult::Warn:
+        Diag(New->getLocation(), diag::warn_mismatched_func_effect_override)
+            << Effect.name();
+        Diag(Old->getLocation(), diag::note_overridden_virtual_function);
+        // TODO: It would be nice to have a FIXIT here!
+        AnyDiags = true;
+        break;
+      case FunctionEffect::OverrideResult::Propagate: {
+        auto MergedFX = FunctionEffectSet::getUnion(Context, OldFX, NewFX);
+
+        FunctionProtoType::ExtProtoInfo EPI = NewFT->getExtProtoInfo();
+        EPI.FunctionEffects = MergedFX;
+        QualType ModQT = Context.getFunctionType(NewFT->getReturnType(),
+                                                 NewFT->getParamTypes(), EPI);
+
+        // TODO: It's ugly to be mutating the incoming const method. It is
+        // mutable in the calling function, though. There is also the
+        // possibility here that we are discarding some other sort of sugar on
+        // the method's type.
+        const_cast<CXXMethodDecl *>(New)->setType(ModQT);
----------------
Sirraide wrote:

Yeah, this seems very scuffed to me as well; `New` should probably be a non-`const` pointer (which may have to propagate up to the callers of this function if need be; doing so seems fine as this happens as part of function declaration merging, so we’re changing `New` anyway).

https://github.com/llvm/llvm-project/pull/84983