bug#4534: assertion failure at window.c:grow_mini_window

[martin rudalics]

>> Would writing
>>
>>              if (height > WINDOW_TOTAL_LINES (w))
>>                {
>>                  freeze_window_starts (f, 1);
>>                  grow_mini_window (w, height - WINDOW_TOTAL_LINES (w));
>>                }
>>
>> instead do better?
>
> In my case, height < WINDOW_TOTAL_LINES (w) at that point (that's what
> triggers the xassert), so adding the above check works exactly like
> removing the if() altogether: it does not crash, but it does not
> shrink the minibuffer to 5 lines, which would be the correct height.

Seems I'm too silly to understand what's going on in resize_mini_window.
Apparently

      if (EQ (Vresize_mini_windows, Qgrow_only))
        ... fails so ...
      else
        ... gets executed ...
        {
          /* Always resize to exact size needed.  */
          if (height > WINDOW_TOTAL_LINES (w))
            ... fails again, so ...
          else if (height < WINDOW_TOTAL_LINES (w))
            ... we're here ...
            {
              int old_height = WINDOW_TOTAL_LINES (w);
              freeze_window_starts (f, 0);
              shrink_mini_window (w);
              ... should resize the mini_window down to one line ...

              if (height)
                {
                  freeze_window_starts (f, 1);
                  ... and the following causes the assertion failure to
                      trigger because the second arguments gets less
                      than zero ...
                  grow_mini_window (w, height - WINDOW_TOTAL_LINES (w));
                }

              window_height_changed_p = WINDOW_TOTAL_LINES (w) != old_height;
            }
        }

Now if I replace the check

              if (height)

by

              if (height > WINDOW_TOTAL_LINES (w))

I should be able to avoid the delta = -1 assertion failure in
grow_mini_window.  Can you confirm my reasoning so far?

The fact that `height' is something like zero or less seems to stem from
some miscalculation (probably caused by the line-wrapping part) earlier
in the code of resize_mini_window.  If that's so could you try to find
out why height is not at least one line?

martin