在 2020/1/21 上午9:31, JunMa 写道:
在 2020/1/20 下午11:49, Nathan Sidwell 写道:
On 1/20/20 12:18 AM, JunMa wrote:
Hi
This patch adds lookup_awaitable_member, it outputs error messages
when any of
the await_ready/suspend/resume functions are missing in awaitable
class.
This patch also add some error check on return value of
build_co_await since we
may failed to build co_await_expr.
Bootstrap and test on X86_64, is it OK?
Regards
JunMa
gcc/cp
2020-01-20 Jun Ma <ju...@linux.alibaba.com>
* coroutines.cc (lookup_awaitable_member): Lookup an
awaitable member.
(build_co_await): Use lookup_awaitable_member instead of
lookup_member.
(finish_co_yield_expr, finish_co_await_expr): Add error
check on return
value of build_co_await.
gcc/testsuite
2020-01-20 Jun Ma <ju...@linux.alibaba.com>
* g++.dg/coroutines/coro1-missing-await-method.C: New test.
1) you're mixing two distinct changes, the one about the error
message and the one about changing error_mark_node.
tree op = build_co_await (kw, a, CO_AWAIT_SUSPEND_POINT);
- TREE_SIDE_EFFECTS (op) = 1;
- SET_EXPR_LOCATION (op, kw);
-
+ if (op != error_mark_node)
+ {
+ TREE_SIDE_EFFECTS (op) = 1;
+ SET_EXPR_LOCATION (op, kw);
+ }
return op;
}
these two fragments are ok, as a separate patch.
ok, I'll split it.
+/* Lookup an Awaitable member, which should be await_ready,
await_suspend
+ or await_resume */
+
+static tree
+lookup_awaitable_member (tree await_type, tree member_id,
location_t loc)
+{
+ tree aw_memb
+ = lookup_member (await_type, member_id,
+ /*protect*/ 1, /*want_type*/ 0, tf_warning_or_error);
fwiw '/*foo=*/expr' is the canonical form -- i know Iain wasn't
consistent, and it looks like I may have missed a few places in review.
ok.
+ if (aw_memb == NULL_TREE || aw_memb == error_mark_node)
+ {
+ error_at (loc, "no member named %qE in %qT", member_id,
await_type);
+ return error_mark_node;
+ }
+ return aw_memb;
+}
This looks wrong. lookup_member is being passed tf_warning_or_error,
so it should already be emitting a diagnostic, for the cases where
something is found, but is ambiguous or whatever. So, I think you
only want a diagnostic here when you get NULL_TREE back.
you are right, I follow with same code of lookup_promise_member, I'll
update both.
Regards
JunMa
nathan
Hi nathan,
attachment is the updated patch which check error_mark_node of
build_co_coawait.
Regards
JunMa
gcc/cp
2020-01-21 Jun Ma <ju...@linux.alibaba.com>
* coroutines.cc (finish_co_await_expr): Add error check on return
value of build_co_await.
(finish_co_yield_expr,): Ditto.
---
gcc/cp/coroutines.cc | 15 ++++++++++-----
1 file changed, 10 insertions(+), 5 deletions(-)
diff --git a/gcc/cp/coroutines.cc b/gcc/cp/coroutines.cc
index 43f03f7c49a..bd3656562ec 100644
--- a/gcc/cp/coroutines.cc
+++ b/gcc/cp/coroutines.cc
@@ -810,8 +810,11 @@ finish_co_await_expr (location_t kw, tree expr)
/* Now we want to build co_await a. */
tree op = build_co_await (kw, a, CO_AWAIT_SUSPEND_POINT);
- TREE_SIDE_EFFECTS (op) = 1;
- SET_EXPR_LOCATION (op, kw);
+ if (op != error_mark_node)
+ {
+ TREE_SIDE_EFFECTS (op) = 1;
+ SET_EXPR_LOCATION (op, kw);
+ }
return op;
}
@@ -874,9 +877,11 @@ finish_co_yield_expr (location_t kw, tree expr)
promise transform_await(). */
tree op = build_co_await (kw, yield_call, CO_YIELD_SUSPEND_POINT);
-
- op = build2_loc (kw, CO_YIELD_EXPR, TREE_TYPE (op), expr, op);
- TREE_SIDE_EFFECTS (op) = 1;
+ if (op != error_mark_node)
+ {
+ op = build2_loc (kw, CO_YIELD_EXPR, TREE_TYPE (op), expr, op);
+ TREE_SIDE_EFFECTS (op) = 1;
+ }
return op;
}
--
2.19.1.3.ge56e4f7
