On 4/2/26 11:53 AM, Thomas Berger wrote:
Am Donnerstag, 2. April 2026, 17:28:28 Mitteleuropäische Sommerzeit schrieb
Jason Merrill:
diff --git a/gcc/analyzer/region-model.cc b/gcc/analyzer/region-model.cc
index 4a6f5b4800c..75241f23bb1 100644
--- a/gcc/analyzer/region-model.cc
+++ b/gcc/analyzer/region-model.cc
@@ -4116,6 +4116,10 @@ static bool
struct_or_union_with_inheritance_p (tree struc)
{
tree iter = TYPE_FIELDS (struc);
+
+ while (iter && TREE_CODE (iter) == TYPE_DECL)
I still think this should be != FIELD_DECL, rather than == TYPE_DECL.
As i didn't want to change the behaviour and refactor to fit the new TYPE_DECL
ordering in a single commit, I implemented your recommendation in 3/3 [1] to
keep the history clean.
Ah, right, my C++ patch filter was hiding the third patch from me.
But it seems better to me to change this function once in the first
patch, rather than change it twice.
Unfortunately it's not suitable for GCC 16 at this point since this
isn't a regression and the change affects non-experimental code.
I expected as much. As this was an alternative to "[PATCH v2] c++/reflection:
class_member_of/modules ordering fix" [2] i just want to bring up the other
patch again. While I personally don't think we should change the module
streaming interface just for GCC 16 if this will land in 17, i would love
to close this task in my mind to.
Understandable, but it shouldn't be long before GCC 16 branches.
[1] https://gcc.gnu.org/pipermail/gcc-patches/2026-March/710890.html
[2] https://gcc.gnu.org/pipermail/gcc-patches/2026-March/710607.html
Thomas
