https://gcc.gnu.org/g:1288283051435c6b23473448aa21fa9ca53fabac
commit r15-494-g1288283051435c6b23473448aa21fa9ca53fabac
Author: Jason Merrill <ja...@redhat.com>
Date: Tue May 14 17:37:08 2024 -0400
c++ comment adjustments for 114935
gcc/cp/ChangeLog:
* decl.cc (wrap_cleanups_r): Clarify comment.
* init.cc (build_vec_init): Update comment.
Diff:
---
gcc/cp/decl.cc | 10 +++++-----
gcc/cp/init.cc | 10 ++++++++--
2 files changed, 13 insertions(+), 7 deletions(-)
diff --git a/gcc/cp/decl.cc b/gcc/cp/decl.cc
index 3fc8835154d9..a139b293e00c 100644
--- a/gcc/cp/decl.cc
+++ b/gcc/cp/decl.cc
@@ -8028,11 +8028,11 @@ wrap_cleanups_r (tree *stmt_p, int *walk_subtrees, void
*data)
Unfortunately, there's no way to express this properly in terms of
nesting, as the regions for the temporaries overlap the region for the
variable itself; if there are two temporaries, the variable needs to be
- the first thing destroyed if either of them throws. However, we only
- want to run the variable's cleanup if it actually got constructed. So
- we need to guard the temporary cleanups with the variable's cleanup if
- they are run on the normal path, but not if they are run on the
- exceptional path. We implement this by telling
+ the first thing destroyed if either of the temporary destructors throws.
+ However, we only want to run the variable's cleanup if it actually got
+ constructed. So we need to guard the temporary cleanups with the
+ variable's cleanup if they are run on the normal path, but not if they
+ are run on the exceptional path. We implement this by telling
honor_protect_cleanup_actions to strip the variable cleanup from the
exceptional path.
diff --git a/gcc/cp/init.cc b/gcc/cp/init.cc
index 7bb98f445c37..906e401974cb 100644
--- a/gcc/cp/init.cc
+++ b/gcc/cp/init.cc
@@ -4685,8 +4685,14 @@ build_vec_init (tree base, tree maxindex, tree init,
for any temporaries in the initialization are naturally within our
cleanup region, so we don't want wrap_temporary_cleanups to do
anything for arrays. But if the array is a subobject, we need to
- tell split_nonconstant_init how to turn off this cleanup in favor of
- the cleanup for the complete object. */
+ tell split_nonconstant_init or cp_genericize_target_expr how to turn
+ off this cleanup in favor of the cleanup for the complete object.
+
+ ??? For an array temporary such as an initializer_list backing array,
+ it would avoid redundancy to leave this cleanup active, clear
+ CLEANUP_EH_ONLY, and not build another cleanup for the temporary
+ itself. But that breaks when gimplify_target_expr adds a clobber
+ cleanup that runs before the build_vec_init cleanup. */
if (cleanup_flags)
vec_safe_push (*cleanup_flags, build_tree_list (iterator, maxindex));
}
