Is this a gcc problem?

[Doug Larrick <doug@ties.org>]

I have tracked down a problem I've been seeing with gtk apps for quite
some time, where certain windows don't display properly... see
http://jekyl.ddts.net/~doug/control-center-1.jpg for an example.  The
problem is in the GtkHPaned widget, specifically in how it calculates
a particular subwidget's size.  However, the problem appears to be one
of code generation rather than a problem with the source.  Here's the
relevant snippet, from gtk/gtkhpaned.c in libgtk1.2, near line 219
(the #if'd out part is the original; the #else part is my replacement):

  /* N.B. the way this computation was being done is causing gcc to
   emit bad code on Alpha??? */
#if 0
  child2_allocation.x = paned->groove_rectangle.x + gutter_size / 2 +
                        1;
  child2_allocation.width = MAX (1, (gint) allocation->width -
                                   child2_allocation.x - border_width);
  
#else
  {
    gint child2_x = paned->groove_rectangle.x + gutter_size / 2 + 1;
    child2_allocation.x = child2_x;
    printf("child2_allocation.x as gint16: %d ; as gint: %d\n", 
           child2_allocation.x, child2_x);
    child2_allocation.width = MAX (1, (gint) allocation->width -
                                   child2_x - border_width);
  }
#endif

The interesting thing to note is that child2_allocation looks like
this:
  struct GtkAllocation
   {
     gint16 x;
     gint16 y;
     guint16 width;
     guint16 height;
   };

This looks like a gcc bug (and a bad one) to me.  The two values
printf'd out are nearly always different.

I am going to try to narrow this down to a good test case and submit a
bug against gcc, but first I'm wondering if it rings a bell with
anyone?

Thanks,
-Doug

-- 
Doug Larrick  doug@ties.org  doug.larrick@compaq.com  AIM: DougLarick