PATCH: PR libstdc++/14098

[Benjamin Kosnik]

>It is my assumption that our test suite already directly catches
>most/all mis-marked primary X...

Not really. I did some fixups after the LSB discussion.

>Manually checking...
>
>$ objdump --syms -C libstdc++.so.6 | grep typeinfo
>00064d1a l O .rodata 0000001a typeinfo name for std::__moneypunct_cache<char>
>00064d34 l O .rodata 00000018 typeinfo name for std::__numpunct_cache<char>
>00067470 l O .data   0000000c typeinfo for std::__moneypunct_cache<char>
>0006496a l O .rodata 00000019 typeinfo name for std::__timepunct_cache<char>
>00063b60 l O .rodata 0000002f typeinfo name for __gnu_cxx::demangler::implementation_details
>00067490 l O .data   0000000c typeinfo for std::__numpunct_cache<char>
>0006645c l O .data   00000008 typeinfo for __gnu_cxx::demangler::implementation_details
>00066f48 l O .data   0000000c typeinfo for std::__timepunct_cache<char>
>[... all w entries after here...]
>
>Looking at related X entries (automate as new pass in check abi tool?)
>shows that only __gnu_cxx::demangler::implementation_details matches
>as local.  

>At a first-order, seems we could automate looking for cases where X
>was "visible" in the final image but typeinfo X, etc was "hidden".

Yes, this should be a new pass in check_abi. Oui oui absolument oui! If
you cannot do this, can you put in a mention of this into the TODO list?

Thanks for the help.

-benjamin