RE: [Valgrind-users] "commoned up errors"

[Nicholas N. <nj...@ca...>]

On Tue, 23 Dec 2003, Pablo Alvarez wrote:

> >Yes, there is a way. Use option --num-callers=5.
>
> I have tried using that option, but as far as I can tell that does not
> change the way errors are commoned up. Rather, it gives you a lot more
> information on the call chain for the first error, but I am getting a
> report, for example, about 2000 errors coming from a method that I know is
> only called once...
>
> So what I really would like to know is whether someone has tried to hack
> the code to change how errors are commoned up (or to disable this feature
> completely).

Try the following (untested) patch, combined with a high value for
--num-callers.

N



RCS file: /home/kde/valgrind/coregrind/vg_errcontext.c,v
retrieving revision 1.48
diff -u -3 -p -r1.48 vg_errcontext.c
--- coregrind/vg_errcontext.c   4 Jan 2004 16:43:19 -0000       1.48
+++ coregrind/vg_errcontext.c   4 Jan 2004 17:56:21 -0000
@@ -290,7 +290,7 @@ void VG_(maybe_record_error) ( ThreadId
           Error* p;
           Error* p_prev;
           UInt   extra_size;
-          VgRes  exe_res                = Vg_MedRes;
+          VgRes  exe_res                = Vg_HighRes;
    static Bool   stopping_message       = False;
    static Bool   slowdown_message       = False;
    static Int    vg_n_errs_shown        = 0;