Re: XCOFF/PE hack removal

[Tom Rix <trix@redhat.com>]

To foo or to .foo? That is the question

Speaking for XCOFF,  I'm ok with the change.

You should add powerpc64-elf to the list of targets in the comment.

Tom

Alan Modra wrote:

> I'd like to apply the following so that errors reported when linking
> powerpc64-elf objects discern between a function descriptor symbol
> and it's associated function code symbol, which only differs by a
> leading dot.  However, this will affect PE and XCOFF error messages,
> so I'd like an OK from the opposition camp first.  If not OK, then
> I'll do something target dependent.
>
> ld/ChangeLog
>         * ldmisc.c (demangle): Put back dots when string not demangled.
>
> --
> Alan Modra
> IBM OzLabs - Linux Technology Centre
>
> Index: ld/ldmisc.c
> ===================================================================
> RCS file: /cvs/src/src/ld/ldmisc.c,v
> retrieving revision 1.8
> diff -u -p -r1.8 ldmisc.c
> --- ldmisc.c    2001/09/26 01:55:44     1.8
> +++ ldmisc.c    2002/01/22 09:08:55
> @@ -69,6 +69,7 @@ demangle (string)
>       const char *string;
>  {
>    char *res;
> +  int dots = 0;
>
>    if (output_bfd != NULL
>        && bfd_get_symbol_leading_char (output_bfd) == string[0])
> @@ -78,9 +79,13 @@ demangle (string)
>       format.  Xcoff has a single '.', while the NT PE for PPC has
>       '..'.  So we remove all of them.  */
>    while (string[0] == '.')
> -    ++string;
> +    {
> +      ++dots;
> +      ++string;
> +    }
>
>    res = cplus_demangle (string, DMGL_ANSI | DMGL_PARAMS);
> +  string -= dots;
>    return res ? res : xstrdup (string);
>  }

--
Tom Rix
GCC Engineer
trix@redhat.com