[clang] [clang] Catch missing format attributes (PR #70024)

[Aaron Ballman via cfe-commits]

================
@@ -6849,6 +6849,71 @@ static void handleSwiftAsyncAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
     checkSwiftAsyncErrorBlock(S, D, ErrorAttr, AsyncAttr);
 }
 
+// Warn if parent function misses format attribute. Parent function misses
+// format attribute if there is an argument format string forwarded to calling
+// function with format attribute, parent function has a parameter which is
+// either char or string or pointer to char and parent function format attribute
+// type does not match with calling function format attribute type.
+void Sema::DiagnoseMissingFormatAttributes(const FunctionDecl *FDecl,
+                                           ArrayRef<const Expr *> Args,
+                                           SourceLocation Loc) {
+  assert(FDecl);
+
+  // Check if function has format attribute with forwarded format string.
+  IdentifierInfo *AttrType;
+  if (!llvm::any_of(
+          FDecl->specific_attrs<FormatAttr>(), [&](const FormatAttr *Attr) {
+            if (!Args[Attr->getFirstArg()]->getReferencedDeclOfCallee())
+              return false;
+
+            AttrType = Attr->getType();
+            return true;
+          }))
+    return;
+
+  const FunctionDecl *ParentFuncDecl = getCurFunctionDecl();
+  if (!ParentFuncDecl)
+    return;
+
+  // Check if parent function has char, string or pointer to char parameter.
+  unsigned int StringIndex = 0;
+  if (!llvm::any_of(
+          ParentFuncDecl->parameters(), [&](const ParmVarDecl *Param) {
+            StringIndex = Param->getFunctionScopeIndex() + 1;
+            QualType Ty = Param->getType();
+            if (isNSStringType(Ty, Context, true))
+              return true;
+            if (isCFStringType(Ty, Context))
+              return true;
+            if (Ty->isPointerType() &&
+                Ty->castAs<PointerType>()->getPointeeType()->isCharType())
----------------
AaronBallman wrote:

That's not particularly strong logic for why we shouldn't diagnose it. It's somewhat idiomatic in C++ to handle string literals via this deduced NTTP pattern, so I think we should support it if it's not too difficult to do so. That said, if it's a pain to support for some reason, I think it's reasonable to not handle it yet (I'm not certain how often we'll find people with interfaces mimicking format strings and using the string literal pattern).

https://github.com/llvm/llvm-project/pull/70024