https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121614
Nathaniel Shead <nshead at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|--- |INVALID
Status|UNCONFIRMED |RESOLVED
CC| |nshead at gcc dot gnu.org
--- Comment #4 from Nathaniel Shead <nshead at gcc dot gnu.org> ---
Hi, thanks for the report!
The issue looks to be with the Makefile, GCC is behaving correctly here. A
module TU also should produce an object file which has code emitted into it; in
this case, the ABI says that we should be emitting vtables and typeinfo into
the object file for 'foo.ccm', see
https://itanium-cxx-abi.github.io/cxx-abi/abi.html#vague-vtable:
> Virtual tables for dynamic classes are emitted as follows:
>
> - If the class is templated, the tables are emitted in every
> object that references any of them.
> - Otherwise, if the class is attached to a module, the tables
> are uniquely emitted in the object for the module unit in
> which it is defined.
> - [...]
By producing an object file for foo.ccm (i.e. not passing -fmodule-only) and
linking that into your executable you should get a working build.
Note that this means that there are *two* outputs from building a module: the
object file, with definitions emitted into it, and the module CMI. You can use
a module mapper (see
https://gcc.gnu.org/onlinedocs/gcc/C_002b_002b-Module-Mapper.html) to customise
where the CMI is built, otherwise gcm.cache/modname.gcm is the default
location.
