On 1/20/26 6:02 AM, Marek Polacek wrote:
On Sat, Jan 17, 2026 at 04:31:09PM +0800, Jason Merrill wrote:
On 1/17/26 8:47 AM, Marek Polacek wrote:
I suppose this could/should wait till GCC 17.
Bootstrapped/regtested on x86_64-pc-linux-gnu, ok for trunk?
-- >8 --
During Reflection review it came up that is_std_substitution
handles NAMESPACE_DECLs accidentally: it wants either a class or a
class template, but is looking at the type of any decl.
Are there testcases affected by this?
I added code to check if we now return false for something that we
used to return true for, and found nothing. I ran dg.exp and old-deja.exp.
Let's include such a checking_assert in the return false path.
gcc/cp/ChangeLog:
* mangle.cc (is_std_substitution): Instead of any _DECL, only
accept TYPE_DECL and DECL_CLASS_TEMPLATE_P. Use NULL_TREE
instead of NULL.
---
gcc/cp/mangle.cc | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/gcc/cp/mangle.cc b/gcc/cp/mangle.cc
index 06bd396c03e..4887b7900de 100644
--- a/gcc/cp/mangle.cc
+++ b/gcc/cp/mangle.cc
@@ -478,7 +478,7 @@ add_substitution (tree node)
}
/* Helper function for find_substitution. Returns nonzero if NODE,
- which may be a decl or a CLASS_TYPE, is a template-id with template
+ which may be a class or a class template, is a template-id with template
name of substitution_index[INDEX] in the ::std namespace, with
global module attachment. */
@@ -486,10 +486,10 @@ static bool
is_std_substitution (const tree node,
const substitution_identifier_index_t index)
{
- tree type = NULL;
- tree decl = NULL;
+ tree type = NULL_TREE;
+ tree decl = NULL_TREE;
- if (DECL_P (node))
+ if (TREE_CODE (node) == TYPE_DECL || DECL_CLASS_TEMPLATE_P (node))
{
type = TREE_TYPE (node);
decl = node;
base-commit: e6532f870a741f15082122eaa1120c72569fe6cf
Marek
