[clang] [clang] Diagnose problematic diagnostic messages (PR #93229)

[Vlad Serebrennikov via cfe-commits]

================
@@ -1213,6 +1213,197 @@ static bool isRemark(const Record &Diag) {
   return ClsName == "CLASS_REMARK";
 }
 
+// Presumes the text has been split at the first whitespace or hyphen.
+static bool isExemptAtStart(StringRef Text) {
+  // Fast path, the first character is lowercase or not alphanumeric.
+  if (isLower(Text[0]) || !isAlnum(Text[0]))
+    return true;
+
+  // If the text is all uppercase (or numbers, +, or _), then we assume it's an
+  // acronym and that's allowed. This covers cases like ISO, C23, C++14, and
+  // OBJECT_MODE. However, if there's only a single letter other than "C", we
+  // do not exempt it so that we catch a case like "A really bad idea" while
+  // still allowing a case like "C does not allow...".
+  if (llvm::all_of(Text, [](char C) {
+        return isUpper(C) || isDigit(C) || C == '+' || C == '_';
+      }))
+    return Text.size() > 1 || Text[0] == 'C';
+
+  // Otherwise, there are a few other exemptions.
+  return StringSwitch<bool>(Text)
+      .Case("AddressSanitizer", true)
+      .Case("CFString", true)
+      .Case("Clang", true)
+      .Case("Fuchsia", true)
+      .Case("GNUstep", true)
+      .Case("IBOutletCollection", true)
+      .Case("Microsoft", true)
+      .Case("Neon", true)
+      .StartsWith("NSInvocation", true) // NSInvocation, NSInvocation's
+      .Case("Objective", true)          // Objective-C (- is a word boundary)
----------------
Endilll wrote:

```suggestion
      .Case("Objective", true)          // Objective-C (hyphen is a word boundary)
```

https://github.com/llvm/llvm-project/pull/93229