On 7/16/24 5:55 PM, Andi Kleen wrote:
On Tue, Jul 16, 2024 at 12:52:31PM -0700, Andi Kleen wrote:
On Tue, Jul 16, 2024 at 02:51:13PM -0400, Jason Merrill wrote:
On 7/16/24 12:18 PM, Andi Kleen wrote:
On Tue, Jul 16, 2024 at 11:17:14AM -0400, Jason Merrill wrote:
On 7/16/24 11:15 AM, Andi Kleen wrote:
In the adjusted test it looks like the types of f and g match, so I wouldn't
expect an error.
Good point! Missing the forest for the trees.
Anyways are the C++ patches ok with this change?
I'm still looking for a test which does error because the types are
different.
Like this?
Where the called function returns C and the callee function does not.
In this case the attribute seems to get lost and it succeeds.
This somewhat hackish patch fixes it here, to handle the case
of a TARGET_EXPR where the CALL_EXPR is in the cleanup. extract_call
bails on that.
The CALL_EXPR in the cleanup is calling the destructor, that's not what
we're trying to tail-call.
I think the problem here is that the call to f<C> is represented with an
AGGR_INIT_EXPR instead of CALL_EXPR, so you need to handle the flag on
that tree_code as well.
diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
index 3b914089a6e2..8753aa51da52 100644
--- a/gcc/cp/pt.cc
+++ b/gcc/cp/pt.cc
@@ -21124,6 +21124,10 @@ tsubst_expr (tree t, tree args, tsubst_flags_t
complain, tree in_decl)
CALL_EXPR_REVERSE_ARGS (call) = rev;
if (TREE_CODE (call) == CALL_EXPR)
CALL_EXPR_MUST_TAIL_CALL (call) = mtc;
+ else if (mtc
+ && TREE_CODE (ret) == TARGET_EXPR
+ && TREE_CODE (TARGET_EXPR_CLEANUP (ret)) ==
CALL_EXPR)
+ CALL_EXPR_MUST_TAIL_CALL (TARGET_EXPR_CLEANUP (ret)) = mtc;
}
if (warning_suppressed_p (t, OPT_Wpessimizing_move))
/* This also suppresses -Wredundant-move. */
