Hello,

This follows up an earler e-mail where I wrote:

Should this be called symtab_complaint(), or perhaphs, should 
complaint() take an extra parameter (the complaint class) so that other 
code can use this mechanism?

I've implemented this.  In the process I've tried to make the code a 
little bit more i18n friendly (it was assuming that things read left to 
right).

Anyway, this is my original description:

The files complaints.[hc] implement a mechanism for supressing warning 
messages that occure more than a small number of times.  GDB's debug 
readers use this mechanism to supress all but the first few warnings 
generated when reading an object file.

The current implementation stores the warning message string in a 
structure vis:

     struct complaint argument_complaint = { "Argument '%d'", };

     DOUBLEST argument;
     complain (&argument_complaint, "argument");
     complain (&argument_complaint, argument);
     complain (&argument_complaint, &argument);

The problem I see with this is that there is nothing (other than the 
human eye) checking for consistency between the format string and the call.

I'd like to propose a new complaints interface:

     void complaint (const char *fmt, ...);

(with a format printf attribute) so that the compiler (GCC with 
-Wformat) can check, at build time, the consistency of the format string 
and its parameter list.  To issue the same complaint from multiple 
points in the code, a wrapper function can be used:

     argument_complaint (int argument)
     {
       complaint ("Argument '%d'", argument);
     }

The process of updating the above should flush out a few bugs :-)

This change would help eliminate problems such as:
http://sources.redhat.com/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gdb&pr=211

-- 

I don't use the most efficient of algorithms when detecting duplicate 
complaints.  But then, I don't know how often complaints are occuring 
and my objective is to fix the format miss-match.

The complaint() interface should probably be documented along with 
error() and warning().

-- 

thoughts?
Andrew