[Lldb-commits] [clang] [lldb] [UBSan][BoundsSafety] Implement support for more expressive "trap reasons" (PR #154618)

[via lldb-commits]

================
@@ -793,6 +795,36 @@ DiagnosticBuilder::DiagnosticBuilder(const DiagnosticBuilder &D)
   D.Clear();
 }
 
+RuntimeTrapDiagnosticBuilder::RuntimeTrapDiagnosticBuilder(
+    DiagnosticsEngine *DiagObj, unsigned DiagID, TrapReason &TR)
+    : DiagnosticBuilder(DiagObj, SourceLocation(), DiagID), TR(TR) {
+  assert(DiagObj->getDiagnosticIDs()->isTrapDiag(DiagID));
+}
+
+RuntimeTrapDiagnosticBuilder::~RuntimeTrapDiagnosticBuilder() {
+  // Store the trap message and category into the TrapReason object.
+  getMessage(TR.Message);
+  TR.Category = getCategory();
+
+  // Make sure that when `DiagnosticBuilder::~DiagnosticBuilder()`
+  // calls `Emit()` that it does nothing.
+  Clear();
+}
+
+void RuntimeTrapDiagnosticBuilder::getMessage(SmallVectorImpl<char> &Storage) {
+  // Render the Diagnostic
+  Diagnostic Info(DiagObj, *this);
+  Info.FormatDiagnostic(Storage);
+}
+
+StringRef RuntimeTrapDiagnosticBuilder::getCategory() {
+  auto CategoryID =
+      DiagObj->getDiagnosticIDs()->getCategoryNumberForDiag(DiagID);
+  if (CategoryID == 0)
+    return "";
----------------
Sirraide wrote:

```suggestion
```
I don’t think this check is even necessary since `getCategoryNameFromID()` already returns an empty `StringRef` if the ID is invalid.

Side note: The comment on `getCategoryNameFromID()` needs updating because it mentions returning ‘null’, even though that doesn’t make sense for a `StringRef`; it used to return a `const char*` and it seems that whoever refactored it forgot to update the comment.

https://github.com/llvm/llvm-project/pull/154618