[Bug c++/114795] internal compiler error: in finish_member_declaration after module import in gcc 14.0.1 snapshot

["nshead at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114795

Nathaniel Shead <nshead at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           See Also|                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=114600,
                   |                            |https://gcc.gnu.org/bugzill
                   |                            |a/show_bug.cgi?id=99000
                 CC|                            |nshead at gcc dot gnu.org

--- Comment #3 from Nathaniel Shead <nshead at gcc dot gnu.org> ---
Similarly to PR114600 (and PR99000 and maybe some others?) this is because we
don't support textual redefinitions yet (`#include` after `import` naming the
same things). See https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Modules.html:


G++’s modules support is not complete. Other than bugs, the known missing
pieces are: 

[...]

Textual merging of reachable GM entities

    Entities may be multiply defined across different header-units. These must
be de-duplicated, and this is implemented across imports, or when an import
redefines a textually-defined entity. However the reverse is not
implemented—textually redefining an entity that has been defined in an imported
header-unit. A redefinition error is emitted.


A workaround is to ensure that `#include` always goes before `import`, or to
use a header module here rather than `#include`.

FWIW this particular ICE is because we hit this assertion:

  /* Don't add decls after definition.  */
  gcc_assert (TYPE_BEING_DEFINED (current_class_type)
              /* We can add lambda types when late parsing default
                 arguments.  */
              || LAMBDA_TYPE_P (TREE_TYPE (decl)));

but the class has already been defined earlier, by the import.


The other issue here is that we didn't discard the entire GMF here, I suppose
possibly because of a similar issue as described in
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114630#c2.