[WIP] C++: Adjust implicit '__cxa_bad_typeid' prototype to reality (was: [v2] C++: Adjust implicit '__cxa_bad_cast' prototype to reality)

[Thomas Schwinge]

Hi!

On 2025-03-21T15:46:01+0100, I wrote:
> On 2025-03-19T14:25:49+0000, Jonathan Wakely <jwakely.gcc@gmail.com> wrote:
>> On Wed, 19 Mar 2025 at 14:21, Marek Polacek <polacek@redhat.com> wrote:
>>> On Wed, Mar 19, 2025 at 12:38:31PM +0100, Thomas Schwinge wrote:
>>> > In 2001 Subversion r40924 (Git commit 52a11cbfcf0cfb32628b6953588b6af4037ac0b6)
>>> > "IA-64 ABI Exception Handling", '__cxa_bad_cast' changed from 'void *' to
>>> > 'void' return type:
>>> >
>>> >     --- libstdc++-v3/libsupc++/exception_support.cc
>>> >     +++ /dev/null
>>> >     @@ -1,388 +0,0 @@
>>> >     -[...]
>>> >     -// Helpers for rtti. Although these don't return, we give them return types so
>>> >     -// that the type system is not broken.
>>> >     -extern "C" void *
>>> >     -__cxa_bad_cast ()
>>> >     -{
>>> >     -  [...]
>>> >     -}
>>> >     -[...]
>>> >
>>> >     --- /dev/null
>>> >     +++ libstdc++-v3/libsupc++/unwind-cxx.h
>>> >     @@ -0,0 +1,163 @@
>>> >     +[...]
>>> >     +extern "C" void __cxa_bad_cast ();
>>> >     +[...]
>>> >
>>> >     --- /dev/null
>>> >     +++ libstdc++-v3/libsupc++/eh_aux_runtime.cc
>>> >     @@ -0,0 +1,56 @@
>>> >     +[...]
>>> >     +extern "C" void
>>> >     +__cxa_bad_cast ()
>>> >     +{
>>> >     +  [...]
>>> >     +}
>>> >     +[...]
>>> >
>>> > The implicit prototype in the C++ front end however wasn't likewise adjusted,
>>> > and so for nvptx we generate code for 'void *' return type:
>>> >
>>> >     // BEGIN GLOBAL FUNCTION DECL: __cxa_bad_cast
>>> >     .extern .func (.param .u64 %value_out) __cxa_bad_cast;
>>> >
>>> >     {
>>> >     .param .u64 %value_in;
>>> >     call (%value_in),__cxa_bad_cast;
>>> >     trap;
>>> >     // (noreturn)
>>> >     exit;
>>> >     // (noreturn)
>>> >     ld.param.u64 %r30,[%value_in];
>>> >     }
>>> >
>>> > ..., which is in conflict with the library code with 'void' return type:
>>> >
>>> >     // BEGIN GLOBAL FUNCTION DECL: __cxa_bad_cast
>>> >     .visible .func __cxa_bad_cast;
>>> >
>>> >     // BEGIN GLOBAL FUNCTION DEF: __cxa_bad_cast
>>> >     .visible .func __cxa_bad_cast
>>> >     {
>>> >     [...]
>>> >     }
>>> >
>>> > ..., and we thus get execution test FAIL for 'g++.dg/rtti/dyncast2.C':
>>> >
>>> >     error   : Prototype doesn't match for '__cxa_bad_cast' in 'input file 7 at offset 51437', first defined in 'input file 7 at offset 51437'
>>> >     nvptx-run: cuLinkAddData failed: device kernel image is invalid (CUDA_ERROR_INVALID_SOURCE, 300)
>>> >
>>> > With this patched, we get the expected: [...]

>>> > --- a/gcc/cp/rtti.cc
>>> > +++ b/gcc/cp/rtti.cc
>>> > @@ -198,7 +198,7 @@ throw_bad_cast (void)
>>> >        fn = get_global_binding (name);
>>> >        if (!fn)
>>> >       fn = push_throw_library_fn
>>> > -       (name, build_function_type_list (ptr_type_node, NULL_TREE));
>>> > +       (name, build_function_type_list (void_type_node, NULL_TREE));
>>> >      }
>>> >
>>> >    return build_cxx_call (fn, 0, NULL, tf_warning_or_error);
>>>
>>> LGTM, matches what I see in abi-eh.html from the itanium-cxx-abi.

> [...] I've now pushed to trunk branch:
> commit 618c42d23726be6e2086d452d6718abe5e0daca8
> "C++: Adjust implicit '__cxa_bad_cast' prototype to reality", [...]

So, a similar problem exists for '__cxa_bad_typeid'.  See the attached
"[WIP] C++: Adjust implicit '__cxa_bad_typeid' prototype to reality" for
what seemed to be the corresponding patch for that one.  However, that
isn't sufficient; we run into internal errors like:

    [...]/g++.dg/rtti/typeid1.C: In function 'int main()':
    [...]/g++.dg/rtti/typeid1.C:9:12: error: lvalue required as unary '&' operand

I got lost in the C++ front end code trying to understand how to resolve
this mismatch.  Anyone able to advise, please?


Grüße
 Thomas


> By the way, I've also tested this with 'fn = get_global_binding (name);'
> commented out, to force always using the C++ front end-synthesized
> prototype (per my understanding).
>
> I've just realized that we might check if some users of 'throw_bad_cast'
> have special code to handle the case of 'void *' return type in addition
> to 'void', which can now be simplified -- I'll try to look into that next
> week.
>
>
> Grüße
>  Thomas


-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0001-WIP-C-Adjust-implicit-__cxa_bad_typeid-prototype-to-.patch
Type: text/x-diff
Size: 3187 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/libstdc++/attachments/20250324/65645517/attachment-0001.bin>