Re: [COMMITTED PATCH] Fix %s that should be %B in several message formats.

[Dave Korn <dave dot korn dot cygwin at googlemail dot com>]

Roland McGrath wrote:
>> Your patch is obviously incorrect, except for the last place you
>> change.  Please fix.
> 
> That last place was the actual one I had noticed and (knew how to) test.
> I'm confused about how the other cases I changed differ from that one.
> They pass a bfd * argument to (*_bfd_error_handler) first after the format,
> corresponding to a %s in the format.

  It doesn't "correspond".  The first argument is non-positional; it's always
the relevant bfd, and it's used anywhere in the format string where a %B is
present.  %A is likewise special; remaining arguments are handled like
sprintf.  See the comment in bfd.c:

> /* This is the default routine to handle BFD error messages.
>    Like fprintf (stderr, ...), but also handles some extra format specifiers.
> 
>    %A section name from section.  For group components, print group name too.
>    %B file name from bfd.  For archive components, prints archive too.
> 
>    Note - because these two extra format specifiers require special handling
>    they are scanned for and processed in this function, before calling
>    vfprintf.  This means that the *arguments* for these format specifiers
>    must be the first ones in the variable argument list, regardless of where
>    the specifiers appear in the format string.  Thus for example calling
>    this function with a format string of:
> 
>       "blah %s blah %A blah %d blah %B"
> 
>    would involve passing the arguments as:
> 
>       "blah %s blah %A blah %d blah %B",
>         asection_for_the_%A,
> 	bfd_for_the_%B,
> 	string_for_the_%s,
> 	integer_for_the_%d);
>  */


    cheers,
      DaveK