On Saturday 08 September 2007 15:46:48 Eli Zaretskii wrote:
> > From: Vladimir Prus <vladimir@codesourcery.com>
> > Date: Sat, 8 Sep 2007 02:48:52 +0400
> > 
> > @@ -3397,30 +3448,79 @@ print_one_breakpoint (struct breakpoint 
> >    struct cleanup *old_chain = make_cleanup_ui_out_stream_delete (stb);
> >    struct cleanup *bkpt_chain;
> >  [...]
> > -  ui_out_field_int (uiout, "number", b->number);
> > +  if (part_of_multiple)
> > +    {
> > +      char buf[30];
> 
> Why 30?  GNU coding standards frown on arbitrary constant sizes.

> > Clearly, '30' is sufficient for 2^32 breakpoints, each having 2^32
> > locations.
>
> Are we talking only about 32-bit machines?

The address size is not important -- we only care about the *number*
of breakpoints and locations. And having 2^32 breakpoints each with
2^32 locations is seriously beyond reasonable number.

Anyway, to avoid further discussion, I've changed code to use
asprintf.

> > +      sprintf (buf, "%d.%d", b->number, loc_number);
> > +      ui_out_field_string (uiout, "number", buf);
> 
> I'm worried about this N.M thing: when the number of locations takes
> 2 or 3 digits, won't it disrupt the display alignment?

I've increased the field width from 3 to 7 -- which will tolerate
3 digits of breakpoint number, dot, and 3 digits of breakpoint location
number. Clearly, this still can overflow, but we can get overflow in
the current gdb, too.

> > +/* set_raw_breakpoint() is a low level routine for allocating and
> 
> Another GNU coding standards thing: foo() should not be used as an
> indication that `foo' is a function.  Just say `foo' instead; foo()
> looks like a call to `foo' with no arguments, which is not what you
> mean.

Changed.

I attach revised patch. Does it look OK?

Thanks,
Volodya