[cxx-abi-dev] Mangling of anonymous unions?

[Kerch Holt]

Jason Merrill wrote:
> On Wed, 18 Dec 2002 15:54:30 -0500, Jason Merrill <jason at redhat.com> wrote:
> 
> 
>>I just noticed that the ABI doesn't define a mangling for anonymous
>>unions.  This is only really relevant for static local anonymous unions in
>>inline functions; all others are private to a translation unit.
>>
>>The choices would seem to be:
>>
>>1) Use the name of the first member.
>>2) Use the name of the largest member.
>>3) Treat them like string literals, perhaps using 'u' instead of 's'.
>>
>>g++ currently does #2.  But it also fails to allocate space for such a
>>union (q.v. g++.brendan/union1.C), so there's no binary compatibility issue
>>in choosing a different answer.
>>
>>My order of preference is 3, 1, 2.  And yes, I'm volunteering to fix g++.
> 
> 
> After working on this for a bit, I like #3 much less; it doesn't work very
> well for namespace-scope unions, which we might as well mangle the same
> way.
> 
Wouldn't you need some type of factoring of all the members of the union
(instead of first/last) to avoid collisions (of similar unions in the
same inline)?

BTW HP C++ is slightly broken WRT this feature -- so changing to meet
some other spec wouldn't cause too much pain.

-- 
Kerch Holt
HP Cupertino Language & Tools Lab