Mercurial > hg > xemacs-beta
view src/event-xlike-inc.c @ 1268:fffe735e63ee
[xemacs-hg @ 2003-02-07 11:50:50 by ben]
fixes for menu crashes + better preemption behavior
This contains two related changes:
(1) Fix problems with reentrant calling of lwlib and associated
crashes when selecting menu items.
(2) Improve redisplay handling of preemption. Turn on lazy lock
and hold down page-down or page-up and you'll see what I mean.
They are related because they both touch on the code that retrieves
events and handles the internal queues.
console-msw.h, event-msw.c, event-stream.c, events.h, menubar-msw.c, menubar-x.c, menubar.h: mswindows_protect_modal_loop() has been generalized to
event_stream_protect_modal_loop(), and moved to event-stream.c.
mswindows_in_modal_loop ->in_modal_loop likewise. Changes in
event-msw.c and menubar-msw.c for the new names and calling format
(use structures instead of static variables in menubar-msw.c).
Delete former in_menu_callback and use in_modal_loop in its place.
Remove emacs_mswindows_quit_check_disallowed_p(), superseded by
in_modal_loop. Use event_stream_protect_modal_loop() in
pre_activate_callback() so that we get no lwlib reentrancy.
Rearrange some of the code in event-msw.c to be grouped better.
Make mswindows_drain_windows_queue() respect in_modal_loop and
do nothing if so.
cmdloop.c, event-stream.c: Don't conditionalize on LWLIB_MENUBARS_LUCID when giving error when
in_modal_loop, and give better error.
event-Xt.c, event-gtk.c: If in_modal_loop, only retrieve process and timeout events.
Don't retrieve any X events because processing them can lead
to reentrancy in lwlib -> death.
event-stream.c: Remove unused parameter to check_event_stream_ok() and change
all callers.
lisp.h, event-stream.c: Rearrange some functions for increased clarity -- in particular,
group all the input-pending/QUIT-related stuff together, and
put right next to next-event stuff, to which it's related.
Add the concept of "HOW_MANY" -- when asking whether user input
is pending, you can ask if at least HOW_MANY events are pending,
not just if any are. Add parameter to detect_input_pending()
for this. Change recursive_sit_for from a Lisp_Object (which
could only be Qt or Qnil) to an int, like it should be.
event-Xt.c, event-gtk.c, event-xlike-inc.c: New file.
Abstract out similar code in event_{Xt/gtk}_pending_p() and write
only once, using include-file tricks. Rewrite this function to
implement HOW_MANY and only process events when not in_modal_loop.
event-msw.c: Implement HOW_MANY and only process events when not in_modal_loop.
event-tty.c: Implement HOW_MANY.
redisplay.c: Add var `max-preempts' to control maximum number of preempts.
(#### perhaps not useful) Rewrite preemption check so that,
rather than preempting when any user events are available, only
preempt when a certain number (currently 4) of them are backed up.
This effectively allows redisplay to proceed to completion in the
presence of a fast auto-repeat (usually the auto-repeating is
generated dynamically as necessary), and you get much better
display behavior with lazy-lock active.
event-unixoid.c: Comment changes.
event-stream.c: Rewrite discard-input much more simply and safely using the
drain-queue functions. I think the old version might loop
forever if called when in_modal_loop.
SEMI-UNRELATED CHANGES:
-----------------------
event-stream.c: Turn QUIT-checking back on when running the pre-idle hook so it
can be quit out of.
indent.c: Document exact functioning of `vertical-motion' better, and its
differences from GNU Emacs.
author | ben |
---|---|
date | Fri, 07 Feb 2003 11:50:54 +0000 |
parents | |
children | a25c824ed558 |
line wrap: on
line source
/* Shared event code between X and GTK -- include file. Copyright (C) 1991-5, 1997 Free Software Foundation, Inc. Copyright (C) 1995 Sun Microsystems, Inc. Copyright (C) 1996, 2001, 2002, 2003 Ben Wing. This file is part of XEmacs. XEmacs is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2, or (at your option) any later version. XEmacs is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with XEmacs; see the file COPYING. If not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ /* Synched up with: Not in FSF. */ /* For some code it's reasonable to have only one copy and conditionalize at run-time. For other code it isn't. #### Perhaps all code should be included here, not in event-xlike.c. However, event-xlike.c is always X-specific, whereas the following code isn't, in the GTK case. */ static int #ifdef THIS_IS_GTK emacs_gtk_event_pending_p (int how_many) #else emacs_Xt_event_pending_p (int how_many) #endif { Lisp_Object event; int tick_count_val; /* If `how_many' is 0, then this function returns whether there are any X, timeout, or fd events pending (that is, whether emacs_Xt_next_event() would return immediately without blocking). If `how_many' is > 0, then this function returns whether there are that many *user generated* events available (keyboard, mouse click, etc.). This also implies that emacs_Xt_next_event() would not block. */ /* This function used to simply check whether there were any X events (or if user_p was 1, it iterated over all the pending X events using XCheckIfEvent(), looking for keystrokes and button events). That worked in the old cheesoid event loop, which didn't go through XtAppDispatchEvent(), but it doesn't work any more -- X events may not result in anything. For example, a button press in a blank part of the menubar appears as an X event but will not result in any Emacs events (a button press that activates the menubar results in an Emacs event through the stop_next_event mechanism). The only accurate way of determining whether these X events translate into Emacs events is to go ahead and dispatch them until there's something on the dispatch queue. */ if (!how_many) { /* We're being asked for *ALL* events, not just user events. */ /* (1) Any pending events in the dispatch queue? */ if (!NILP (dispatch_event_queue)) return 1; /* (2) Any TTY or process input available? Note that formerly we just checked the value of XtAppPending() to determine if there was file-desc input. This doesn't work any more with the signal_event_pipe; XtAppPending() will says "yes" in this case but there isn't really any input. So instead we keep track of the file descriptors, and call select() ourselves. Another way of fixing this problem is for the signal_event_pipe to generate actual input in the form of an identity eval event or something. (#### maybe this actually happens?) */ if (poll_fds_for_input (non_fake_input_wait_mask)) return 1; #ifndef THIS_IS_GTK /* (3) Any timeout input available? */ if (XtAppPending (Xt_app_con) & XtIMTimer) return 1; #else /* #### Is there any way to do this in Gtk? I don't think there is a 'peek' for events */ #endif } else { /* HOW_MANY > 0 */ EVENT_CHAIN_LOOP (event, dispatch_event_queue) { if (command_event_p (event)) { how_many--; if (how_many <= 0) return 1; } } } /* XtAppPending() can be super-slow, esp. over a network connection. Quantify results have indicated that in some cases the call to detect_input_pending() completely dominates the running time of redisplay(). Fortunately, in a SIGIO world we can more quickly determine whether there are any X events: if an event has happened since the last time we checked, then a SIGIO will have happened. On a machine with broken SIGIO, we'll still be in an OK state -- quit_check_signal_tick_count will get ticked at least every 1/4 second, so we'll be no more than that much behind reality. (In general it's OK if we erroneously report no input pending when input is actually pending() -- preemption is just a bit less efficient, that's all. It's bad bad bad if you err the other way -- you've promised that `next-event' won't block but it actually will, and some action might get delayed until the next time you hit a key.) */ if (!in_modal_loop) { /* quit_check_signal_tick_count is volatile so try to avoid race conditions by using a temporary variable */ tick_count_val = quit_check_signal_tick_count; if (last_quit_check_signal_tick_count != tick_count_val #if !defined (THIS_IS_GTK) && (!defined (SIGIO) || defined (CYGWIN)) || (XtIMXEvent & XtAppPending (Xt_app_con)) #endif ) { last_quit_check_signal_tick_count = tick_count_val; /* We need to drain the entire queue now -- if we only drain part of it, we may later on end up with events actually pending but detect_input_pending() returning false because there wasn't another SIGIO. */ event_stream_drain_queue (); if (!how_many) return !NILP (dispatch_event_queue); EVENT_CHAIN_LOOP (event, dispatch_event_queue) { if (command_event_p (event)) { how_many--; if (how_many <= 0) return 1; } } return 0; } } return 0; }