Re: XEmacs freezes under virtual windows on Win2K

>>>>> "Andy" == Andy Piper <andyp(a)bea.com&gt; writes: Andy> At 05:08 PM 1/20/02 +0100, Adrian Aichner wrote: >> >>>>> "Andy" == Andy Piper <andyp(a)bea.com&gt; writes: >> Andy> I think we should do what Ben says and try commenting out Andy> the call to update_window_scrollbars. The alternative would >> >> Something like this? >> >> Index: window.c >> =================================================================== >> RCS file: /pack/xemacscvs/XEmacs/xemacs/src/window.c,v >> retrieving revision 1.52.2.1.2.3 >> diff -u -u -r1.52.2.1.2.3 window.c >> --- window.c 2001/12/01 05:54:21 1.52.2.1.2.3 >> +++ window.c 2002/01/20 15:52:49 >> ＠＠ -386,9 +386,6 ＠＠ mir-> current_display_lines = Dynarr_new (display_line); mir-> desired_display_lines = Dynarr_new (display_line); >> >> -#ifdef HAVE_SCROLLBARS >> - update_window_scrollbars (XWINDOW (win), mir, 0, 0); >> -#endif mir-> buffer = NULL; >> } Andy> No I mean in mark_redisplay(). We should update the window mirrors Andy> before going into GC. Perhaps I misunderstood, but: Once we are in mark_redisplay, GC is already in progress. It's too late to do something pre-gc. Where would I put the register_pre_gc_action (...); call which should do the following: FRAME_LOOP_NO_BREAK (frmcons, devcons, concons) { struct frame *f = XFRAME (XCAR (frmcons)); /* #### urk! this does tons o' crap, such as creating lots of structs, doing window system actions, etc. we DO NOT want to be doing this -- marking should never change any state. i think we can just delete this. --ben */ /* Prompted by Ben's note above and following advice from Andy I am trying to do this via newly implemented pre_gc_action -- Adrian. */ update_frame_window_mirror, f); } post_gc actions are set up like this: if (gc_in_progress) /* #### way bogus! need to remove the offending call. see mark_redisplay(). */ register_post_gc_action (unshow_that_mofo, (void *) SCROLLBAR_MSW_HANDLE (sb)); but we can't do the same for pre: if (gc_in_progress) register_post_gc_action (...); couldn't possibly work, right :-) It would only do it previous to the next GC. Are you suggesting the following should also be reworked to pre_gc? static void mswindows_release_scrollbar_instance (struct scrollbar_instance *sb) { if (gc_in_progress) /* #### way bogus! need to remove the offending call. see mark_redisplay(). */ register_post_gc_action (unshow_that_mofo, (void *) SCROLLBAR_MSW_HANDLE (sb)); else ShowScrollBar (SCROLLBAR_MSW_HANDLE (sb), SB_CTL, 0); SCROLLBAR_MSW_SIZE (sb) = 0; } Ready for a vacation, Adrian Andy> be to do this before we go into GC. We could do this by Andy> having a run_pre_gc_actions which does the same as Andy> run_post_gc_actions and then does the update there. I don't >> >> I wonder why the implementer of run_post_gc_actions has not >> implemented run_pre_gc_actions as well. Could you think of a reason >> why this would not be a good idea? Andy> The reason Ben did not do this is because run_post_gc_actions is Andy> supposed to be used for cleaning up things that cannot be deleted in Andy> GC (e.g. GUI windows). Before GC is run this condition does not Andy> exist. However, there is no reason I can think of why this isn't a Andy> good idea. >> What would we set up for register_pre_gc_action? >> >> Just a function doing >> >> #ifdef HAVE_SCROLLBARS >> update_window_scrollbars (XWINDOW (win), mir, 0, 0); >> #endif >> >> perhaps? Andy> Right, but looping over all frames as per mark_redisplay() Andy> andy -- Adrian Aichner mailto:adrian＠xemacs.org http://www.xemacs.org/