Re: Really obscure C++ mangling error with -fnew-abi

["Martin v. Loewis" <martin at loewis dot home dot cs dot tu-berlin dot de>]

>   I fixed this bug with the following patch.  After the patch, this:
> 
>   00000000 W _ZSt26__uninitialized_fill_n_auxIPSsjSsET_S1_T0_RKT1_5_BoolILb0EE
> 
> is the only output from `nm' containing uninitialized_fill_n_aux,
> which is a good sign.  This demangles to:
> 
>   std::string* 
>   std::__uninitialized_fill_n_aux<std::string*, unsigned, std::string>
>     (std::string*, unsigned, const std::string&, _Bool<false>)
> 
> which looks right to me.

I'm having problems with the numbering of things (the S1_ part in
particular); I'd appreciate if anybody could correct me. Building up
the compression dictionary, I get

std::                           - predefined entry, not added
std::__uninitialized_fill_n_aux - function name, not added
std::string                     - predefined entry
std::string*                    - S_ (also first template parameter, T_)
unsigned                        - basic type, not added
std::string                     - predefined entry
std::__u..<types>               - function name, ignored, or S0_?
return type                     - T_, according to definition, added as S1_?

Now, we are up the part of IPSsjSsET_
S1_ clearly is intended as a back-reference to T_, as the first parameter
is the same as the return type.

However, the rationale for not adding the function name to the
dictionary (*) surely applies not only to the name itself (which would
be __uninitialized_fill_n_aux, but also to all components which it is
part of, ie. std::__un... and std::__un...<types>). So I can't see
what S0_ would be in this scenario.

Any enlightenment appreciated,
Martin

(*) It is not possible that there is any back-reference to this name,
so it does not need to be added.