https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90172
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jason at gcc dot gnu.org
--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
So, I guess the rejects-valid part is a P2 8/9 regression (if the testcase is
really valid) and the ICE is error recovery regression for that (9 only).
For the latter, I guess something like:
--- gcc/cp/pt.c.jj 2019-04-22 16:03:23.000000000 +0200
+++ gcc/cp/pt.c 2019-04-23 17:21:01.898950417 +0200
@@ -18869,7 +18869,8 @@ tsubst_copy_and_build (tree t,
/* We aren't going to do normal overload resolution, so force the
template-id to resolve. */
function = resolve_nondeduced_context (function, complain);
- for (unsigned i = 0; i < nargs; ++i)
+ unsigned int n_call_args = call_args->length ();
+ for (unsigned i = 0; i < n_call_args; ++i)
{
/* In a thunk, pass through args directly, without any
conversions. */
@@ -18881,9 +18882,10 @@ tsubst_copy_and_build (tree t,
if (thisarg)
{
/* Shift the other args over to make room. */
- vec_safe_push (call_args, (*call_args)[nargs-1]);
- for (int i = nargs-1; i > 0; --i)
- (*call_args)[i] = (*call_args)[i-1];
+ tree last_arg = (*call_args)[n_call_args - 1];
+ vec_safe_push (call_args, last_arg);
+ for (int i = n_call_args - 1; i > 0; --i)
+ (*call_args)[i] = (*call_args)[i - 1];
(*call_args)[0] = thisarg;
}
ret = build_call_a (function, call_args->length (),
could do the job, nargs doesn't take into account if there are more arguments
in call_args due to some pack expansion and also the vec_safe_push is broken
because (*call_args)[nargs-1] is just a reference and trying to push it if it
needs to reallocate is broken.
I have no idea if n_call_args could be 0 and thisarg non-NULL, if yes, we need
to just vec_safe_push (call_args, thisarg); in that case instead of moving
anything.
