Re: [cxx-abi-dev] Possibly ambiguous mangling of extern "C" functions

[Mark Mitchell <mark@xxxxxxxxxxxxxxxx>]

David Vandevoorde wrote:

> > extern "C" bool IsEmpty(char *); // (un)mangled as IsEmpty
> > template<bool (&)(char *)> struct CB { static int x; };
> > // CB<IsEmpty> is mangled as "2CBIL_Z7IsEmptyEE"
> > int *p = &CB<IsEmpty>::x;

> Normally, we issue the following error on that example:
>
> "t.c", line 4: error: a reference of type "bool (&)(char *)" (not
>           const-qualified) cannot be initialized with a value of type
>           "bool (char *) C"

Yes, that's the right error.  G++ has never implemented extern "C" 
function *types*, so doesn't issue this kind of error.  There are extern 
"C" functions, but the linkage isn't part of the type per se.  This is a 
known bug.

So, now we're talking about mangling for an invalid C++ program, which 
is outside the scope of the ABI.

> When I enable the emulation of GCC, we accept it and mangle it as 
> _ZN2CBIXL_Z7IsEmptyEEE1xE.
>
> We also have a mode to emulate what we consider "GNU ABI bugs".  When I 
> enable that, we generate _ZN2CBIXL7IsEmptyEEE1xE.

The GCC output is actually:

_ZN2CBILZ7IsEmptyEE1xE

Indeed, GCC does not generate the "X...E" form here.  Instead, it uses 
the expr-primary production directly from the template-arg rule, which 
makes sense since there is no complex expression here.  However, I do 
think that G++ is in error not to omit the underscore before the Z.

--
Mark Mitchell
CodeSourcery
mark@xxxxxxxxxxxxxxxx
(650) 331-3385 x713