From mboxrd@z Thu Jan  1 00:00:00 1970
From: Keith Seitz <keiths@cygnus.com>
To: Insight Maling List <insight@sources.redhat.com>
Cc: Tom Tromey <tromey@cygnus.com>
Subject: Error message revelation
Date: Mon, 23 Apr 2001 12:25:00 -0000
Message-id: <Pine.SOL.3.91.1010423121040.1894F-100000@ryobi.cygnus.com>
X-SW-Source: 2001-q2/msg00139.html

Hi,

I don't know about you, but I've always wondered why I seem to get a lot 
of messages like "Error:" in the console window... Here's an example:

(gdb) tk gdb_cmd "list" flubber
Error:
(gdb) tk gdb_get_breakpoint_info 12345678
Error:

The reason? When TCL_ERROR is returned by any wrapped call, IT IS ASSUMED 
THAT THE ERROR MESSAGE IS IN THE INTERPRETER.

Here's the interesting bit of code in gdbtk-cmds.c (call_wrapper):

  /*
   * Now copy the result over to the true Tcl result.  If
   * GDBTK_TO_RESULT flag bit is set, this just copies a null object
   * over to the Tcl result, which is fine because we should reset the
   * result in this case anyway.  If the wrapped command returned an
   * error, then we assume that the result is already set correctly.
   */
  if ((result_ptr->flags & GDBTK_IN_TCL_RESULT) || wrapped_returned_error)
    {
      Tcl_DecrRefCount (result_ptr->obj_ptr);
    }
  else
    {
      Tcl_SetObjResult (interp, result_ptr->obj_ptr);
    }

  result_ptr = old_result_ptr;

Anyone know why we only DecrRefCount when wrapped_returned_error is true? 
(In other words, why are we NOT copying result_ptr->obj_ptr into the 
interpreter?)

I can only guess that this was done because we want to do things like:

   Tcl_WrongNumArgs (...);
   return TCL_ERROR;

instead of

   Tcl_WrongNumArgs (...);
   result_ptr->flags &= GDBTK_IN_TCL_RESULT;
   return TCL_ERROR;

However, I think that the latter, more verbose code is much easier to 
remember... The result is ALWAYS taken from result_ptr->obj_ptr EXCEPT if 
GDBTK_IN_TCL_RESULT is specified. That's much simpler than: result is 
taken from result_ptr->obj_ptr EXCEPT if there was an error OR 
GDBTK_IN_TCL_RESULT is set.

If I don't hear any objections, I am going to clean this up to make it 
work. I don't want to remember a bunch of rules. Gdbtk is complex enough 
as it is.

Tom?? You're name is on the change:

2000-11-29  Tom Tromey  <tromey@cygnus.com>

        * gdbtk-cmds.c (call_wrapper): Don't reset result if wrapped
        command returned error.

Any idea what it was supposed to fix?
Keith