[clang-tools-extra] [clang-tidy] Catch <cctype> function args in bugprone-signed-char-misuse (PR #211403)

[via cfe-commits]

================
@@ -125,11 +125,38 @@ void SignedCharMisuseCheck::registerMatchers(MatchFinder *Finder) {
           .bind("arraySubscript");
 
   Finder->addMatcher(STDArraySubscript, this);
+
+  // Catch signed char values passed to a <cctype>/<ctype.h> classification
+  // or conversion function; any value other than EOF or one representable
+  // as unsigned char is undefined behavior. Unlike the matchers above, this
+  // matches the uncast argument directly instead of going through
+  // charCastExpression(), because hasArgument() strips implicit casts off
+  // the argument before matching it.
+  const auto IntTypedef = qualType(hasDeclaration(typedefDecl(
+      hasAnyName(utils::options::parseStringList(CharTypedefsToIgnoreList)))));
+  const auto CctypeFunctionArgument =
+      callExpr(
+          callee(functionDecl(
+              hasAnyName("isalnum", "std::isalnum", "isalpha", "std::isalpha",
+                         "isblank", "std::isblank", "iscntrl", "std::iscntrl",
----------------
cavdarahmet wrote:

Done — the names now live in a `static constexpr StringRef[]` and are fully qualified (`::isalpha` / `::std::isalpha`), which restricts matching to global and `std` declarations. I went with qualified names rather than `isInStdNamespace()` since the global half would still need a separate condition, while the qualified list covers both with one mechanism — happy to switch if you prefer. `hasName()` treats inline namespaces transparently, so libc++'s `std::__1` declarations still match (verified). Added regression tests for `custom::isalpha` and `Classifier::toupper`.


https://github.com/llvm/llvm-project/pull/211403