[PATCH] D98095: [clang] Fix ICE on invalid type parameters for concepts

[Richard Smith - zygoloid via Phabricator via cfe-commits]

rsmith added inline comments.


================
Comment at: clang/lib/Sema/SemaType.cpp:1644-1661
+    TemplateIdAnnotation *TemplateId = DS.getRepAsTemplateId();
+    ConceptDecl *TypeConstraintConcept = nullptr;
+    llvm::SmallVector<TemplateArgument, 8> TemplateArgs;
     if (DS.isConstrainedAuto()) {
-      Result = ConvertConstrainedAutoDeclSpecToType(S, DS,
-                                                    AutoTypeKeyword::Auto);
-      break;
+      if (TemplateId) {
+        TypeConstraintConcept =
+            cast<ConceptDecl>(TemplateId->Template.get().getAsTemplateDecl());
----------------
Minor style nits: reduce scope of `TemplateId` variable and consistently brace the `else` if and only if the `if` is braced.


================
Comment at: clang/lib/Sema/SemaType.cpp:5947-5948
       TL.setNameLoc(DS.getTypeSpecTypeLoc());
       if (!DS.isConstrainedAuto())
         return;
       if (DS.getTypeSpecScope().isNotEmpty())
----------------
Given that we recover from an invalid `TemplateId` by creating an unconstrained `AutoType`, it would be more consistent to skip filling in the `NestedNameSpecifierLoc` here in that case too. (That way, we would populate the `NestedNameSpecifierLoc` if and only if we created a constrained `AutoType`.) I think it doesn't matter right now because no-one should look at this information for an unconstrained `AutoType`, but I could imagine we would switch to storing smaller `TypeLoc` data in the future if the type is unconstrained, and if we did, the writes to the `NestedNameSpecifierLoc` below would break.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D98095/new/

https://reviews.llvm.org/D98095