]> git.ipfire.org Git - thirdparty/gcc.git/commit
tree: Fix up build_function_type for (...) fntypes [PR123977]
authorJakub Jelinek <jakub@redhat.com>
Fri, 6 Feb 2026 10:27:52 +0000 (11:27 +0100)
committerJakub Jelinek <jakub@gcc.gnu.org>
Fri, 6 Feb 2026 10:27:52 +0000 (11:27 +0100)
commit5f75fc37b131c701a5b5c627239fa4e4dfabe230
tree8fa1d1071ff18c0a40ca797058bbfe67a92dc1d4
parent4046caad61a0406e42cae0c38a6aa26f5975a197
tree: Fix up build_function_type for (...) fntypes [PR123977]

The following testcase ICEs in -std=c++26 mode since my r16-4338 C++26
va_start changes.
The problem is that if we have anything non-canonical in a function
type with (...) in C++26 (or C23/C2Y) mode, in this case the R typedef
on return type rather than void, then we try to build TYPE_CANONICAL
for that, but weren't passing in the no_named_args_stdarg_p, so
a type with TYPE_NO_NAMED_ARGS_STDARG_P flag set got TYPE_CANONICAL
with TYPE_NO_NAMED_ARGS_STDARG_P flag cleared and comptypes then
didn't like that as those aren't really compatible types
(...) vs. the C89-ish () which even C++ uses for some type-generic
etc. builtins.

2026-02-05  Jakub Jelinek  <jakub@redhat.com>

PR c++/123977
* tree.cc (build_function_type): Pass no_named_args_stdarg_p
as last argument to recursive call.

* g++.dg/cpp26/stdarg10.C: New test.
gcc/testsuite/g++.dg/cpp26/stdarg10.C [new file with mode: 0644]
gcc/tree.cc