[PATCH] D104902: [OpaquePtr] Allow globals with opaque pointer value type

[Duncan P. N. Exon Smith via Phabricator via llvm-commits]

dexonsmith added a comment.

In D104902#2841332 <https://reviews.llvm.org/D104902#2841332>, @dexonsmith wrote:

> In D104902#2840987 <https://reviews.llvm.org/D104902#2840987>, @aeubanks wrote:
>
>> lgtm, though invalid-opaque-ptr.ll is failing and should probably be deleted
>
> I don't understand why the testcase was deleted. It seems to indicate a hole in the verifier. Isn't this still invalid IR?
>
>   define void @f(ptr %a) {
>       %b = bitcast ptr %a to ptr*
>       ret void
>   }

Ah, I see -- the parsed IR will be correct do to the change to `PointerType::get`. So this isn't a verifier bug, it's a parser bug.

Seems like you just need to update:

  // Type ::= Type '*'
  case lltok::star:
    if (Result->isLabelTy())
      return tokError("basic block pointers are invalid");
    if (Result->isVoidTy())
      return tokError("pointers to void are invalid - use i8* instead");
    if (!PointerType::isValidElementType(Result))
      return tokError("pointer to this type is invalid");

to have:

  if (Result->isOpaquePointerTy())
    return tokError("pointer to opaque pointer is invalid");

before the the call to isValidElementType.

You'd need to update the logic in the similar case below for `lltok::kw_addrspace` as well.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104902/new/

https://reviews.llvm.org/D104902