Version script vs. mangled names on systems where size_type != unsigned int

[Paolo Carlini]

Hi,
> It turns out to be a reasonably big change. This is what it takes to add
> all the missing exports that have different signatures on an LLP64.  (There's
> a strange other issue with exception_ptr related stuff being exported that
> isn't exported in the 32-bit build at all; haven't figured out if it's a
> problem yet but for some reason eh_ptr.o is empty in the 32-bit build.)
>   
Maybe because in 32-bit mode the atomic builtins are not available? That
happens under linux x86_64-32-bit for sure.
> libstdc++-v3/ChangeLog:
>
> 	PR target/42377
> 	* config/abi/pre/gnu.ver: Adjust mangled function signatures to permit
> 	LLP64 sizetypes throughout.
>
>   I haven't tried full build-and-test cycles yet; I'm doing this work on a
> cross compiler so it won't be doing execution tests even if I do; hopefully
> that shouldn't matter because windows exes have to be fully resolved, if
> there's something missing from the import library the linker will fail rather
> than just allow an undefined symbol in the linked executable.  How much/what
> kind of testing would you like me to do for this to be OK?
>   
Not much, I just checked that on x86_64-linux multilib abi_check doesn't
fail, which is good ;) I'm currently running the whole testsuite, but I
don't expect problems, if you don't hear from me the next half hour or
so, please go ahead.

If you could sneak a small comment about 'x' and 'y' in gnu.ver, I think
it could be useful.

Thanks,
Paolo.