GCC's development branch accepts designated initializers for base classes

On 2026-10-18, GCC’s development branch compiles a designated initializer that names a member of a base class. The line is B{.a = 1, .b = 2}, and it needs -std=c++29. GCC 16.1 and 16.2 in C++26 mode reject the line, and so does Clang trunk in -std=c++2d. In June the Brno post had to leave this example commented out, because no compiler accepted it.

In short

The feature is P2287R6, “Designated-initializers for Base Classes”. The committee adopted it for C++29 at the Brno meeting.
GCC trunk, which is GCC 17 in development, accepts all three forms of it under -std=c++29. It is marked experimental.
Under -std=c++26 the same trunk still rejects the plain form and accepts the other two with a warning.
GCC’s status page lists 8 of the 18 C++29 items it tracks as available in GCC 17.

The example from June

The aggregate B has one base class. Before P2287R6, a designator could not name the inherited member, because it is not a direct member of B:

struct A { int a; };
struct B : A { int b; };

auto b1 = B{.a = 1, .b = 2};    // name the inherited member directly
auto b2 = B{{.a = 1}, .b = 2};  // or brace the base subobject
auto b3 = B{A{1}, .b = 2};      // or initialize the base by type

The embed below compiles that file with GCC trunk in C++29 mode and prints the three results. The second button shows GCC 16.1 in C++26 mode on the same file.

Tested: GCC trunk, -std=c++29 + GCC 16.1, -std=c++26

Run on Compiler ExplorerDesignated initializers for a base classx86-64 gcc (trunk) · -std=c++29 · runs live; the version line changes with the snapshot Also on GCC 16.1, -std=c++26Designated initializers for a base classx86-64 gcc 16.1 · -std=c++26 · fails to compile, exact errors shown

On the snapshot I ran, the program prints 1 2 for each form and reports 17.0.0 20261018 (experimental). Compiler Explorer’s trunk moves, so the version line in the embed will differ.

What each compiler does

Compiler and mode	Plain form	Braced base and base by type
GCC trunk 17.0.0 20261018, `-std=c++29`	compiles	compile
GCC trunk, `-std=c++26`	error: no member named `a`	compile with a `-Wc++29-extensions` warning
GCC 16.1 and 16.2, `-std=c++26`	error: no member named `a`	error: all or none of the clauses should be designated
Clang trunk, `-std=c++2d`	error: field designator `a` does not refer to any field	compile with a `-Wc99-designator` warning

The mixed forms are the ones C++20 already ruled out. GCC trunk accepts them in C++26 mode and says they belong to C++29. Clang accepts them as a C99 extension.

What else is in GCC 17

GCC’s C++ status page, read on 2026-10-18, calls its C++29 support experimental. Of the 18 items it lists, 8 are marked GCC 17:

Paper	Item
P2287R6	Designated initializers for base classes
P2953R5	Restrictions on defaulted assignment operators
P3658R1	Identifiers following new Unicode recommendations
P3733R1	More named universal character escapes
P3424R2	Deallocation functions with a throwing specification are ill-formed
P3822R2	Conditional noexcept in compound requirements
P3540R3	An offset parameter for `#embed`
P3950R1	`return_value` and `return_void` are not mutually exclusive

The last one has its own post from 14 October. The page lists two more items as available, one with no version and one marked GCC 12, and 8 as not yet implemented. I ran only the first row.

Sources: P2287R6 · GCC C++ status page, read 2026-10-18 · Herb Sutter, Brno trip report. Compiler Explorer runs on 2026-10-18: GCC trunk 17.0.0 20261018, GCC 16.1, GCC 16.2 and Clang trunk.