================
@@ -221,21 +221,27 @@ llvm::MDNode *CodeGenTBAA::getTypeInfoHelper(const Type
*Ty) {
PtrDepth++;
Ty = Ty->getPointeeType().getTypePtr();
} while (Ty->isPointerType());
- // TODO: Implement C++'s type "similarity" and consider dis-"similar"
- // pointers distinct for non-builtin types.
+
+ SmallString<256> TyName;
if (isa<BuiltinType>(Ty)) {
llvm::MDNode *ScalarMD = getTypeInfoHelper(Ty);
StringRef Name =
cast<llvm::MDString>(
ScalarMD->getOperand(CodeGenOpts.NewStructPathTBAA ? 2 : 0))
->getString();
+ TyName = Name;
+ } else if (!isa<VariableArrayType>(Ty)) {
----------------
rjmccall wrote:
I feel like we should just ignore all array structure here, whether it's
variable-length or fixed-length. For one, partially honoring array structures
probably violates the VLA compatibility rules — IIUC, the C standard requires
l-values of type `T (*)[n]` to be able to alias objects of type `T (*)[10]` if
`n == 10` dynamically, but with your patch, we'd now create a node for that
pointer-to-constant-array type with no relation to the node we'd use for the
pointer-to-VLA type. But also, I feel like trying to honor the array structure
is just pointlessly strict; we can treat a `T (*)[10]` object like a `T *`
object, it's not really going to hurt anything.
You should look through the array structure before you do the
builtin/non-builtin breakdown. There's a `getBaseElementType` operation you
can use.
https://github.com/llvm/llvm-project/pull/110569
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
