[clang] [Clang][Diagnose] New feature: Nested/Indented diagnostic message in sarif mode. (PR #174106)

via cfe-commits cfe-commits at lists.llvm.org
Thu Jan 1 08:03:38 PST 2026


================
@@ -55,28 +59,64 @@ class SARIFDiagnostic : public DiagnosticRenderer {
                                   StringRef ModuleName) override;
 
 private:
-  // Shared between SARIFDiagnosticPrinter and this renderer.
-  SarifDocumentWriter *Writer;
-
-  SarifResult addLocationToResult(SarifResult Result, FullSourceLoc Loc,
-                                  PresumedLoc PLoc,
-                                  ArrayRef<CharSourceRange> Ranges,
-                                  const Diagnostic &Diag);
-
-  SarifResult addRelatedLocationToResult(SarifResult Result, FullSourceLoc Loc,
-                                         PresumedLoc PLoc);
-
-  llvm::SmallVector<CharSourceRange>
-  getSarifLocation(FullSourceLoc Loc, PresumedLoc PLoc,
-                   ArrayRef<CharSourceRange> Ranges);
-
-  SarifRule addDiagnosticLevelToRule(SarifRule Rule,
-                                     DiagnosticsEngine::Level Level);
-
-  llvm::StringRef emitFilename(StringRef Filename, const SourceManager &SM);
-
-  llvm::SmallVector<std::pair<FullSourceLoc, PresumedLoc>>
-      RelatedLocationsCache;
+  class Node {
+  public:
+    // Subclasses
+    struct Result {
+      DiagnosticsEngine::Level Level;
+      std::string Message;
+      DiagOrStoredDiag Diag;
----------------
Sirraide wrote:

Ugh, yeah, this API is a bit of a mess: `StoredDiagnostic` has a message and `Diagnostic` may or may not have one, so we just pass it in here. Honestly, `DiagOrStoredDiag` is a bit of an abomination of a type in the first place, so I’d love it if we could just get rid of it entirely and replace all of its uses with `StoredDiagnostic`, and that would allow us to get rid of the `Message` (and some of the other arguments). 

That said, I haven’t looked at this in a few months, so I’m not sure if there’s anything that would prevent that; also, such a refactor should be a separate NFC patch.

https://github.com/llvm/llvm-project/pull/174106


More information about the cfe-commits mailing list