================
@@ -1351,7 +1352,8 @@ def OSConsumesThis : InheritableAttr {
def Cleanup : InheritableAttr {
let Spellings = [GCC<"cleanup">];
- let Args = [DeclArgument<Function, "FunctionDecl">];
+ let Args = [DeclArgument<Function, "FunctionDecl">,
+ ExprArgument<"Expr", /*opt=*/0, /*fake=*/1>];
----------------
erichkeane wrote:
If we're doing 1, we need to make sure we do this at the DeclArgument level,
which means modifying quite a bit of other attributes potentially. PERHAPS we
could just have it cause us to store a `pair` of `DeclRefExpr` and
`FunctionDecl`, that way it minimizes the amount of work.
As far as the extra source-location (2): that is reasonable as well. I don't
think we have an automagic `fake` argument to do here, so it would have to just
be in `AdditionalMembers`.
I would lean AGAINST 3, unless we have a REALLY good reason to. This would
require that Attr have a trailing-objects, and thus store that much extra
information. We COULD potentially change the entire attributes infrastructure
to just store the `ArgsUnion` type (which is a union of
`Expr*`/`IdentifierInfo*` IIRC) then make every access just do the decoding
from those, but that is a LOT of work.
IN THE END, I think 2 is easiest, so hackiest. 1 is IMO the 'best' situation
here (and is only a medium amount of work, with some talbegen/minor other
changes). We might have other argument types that should do something similar,
but no reason we couldn't do that in the future.
https://github.com/llvm/llvm-project/pull/140233
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
