From mboxrd@z Thu Jan 1 00:00:00 1970 Path: news.gmane.org!.POSTED!not-for-mail From: Daniel Colascione Newsgroups: gmane.emacs.devel Subject: Re: RFC: flicker-free double-buffered Emacs under X11 Date: Fri, 21 Oct 2016 11:27:29 -0700 Message-ID: References: <9e8ad090-a6a0-c807-95ae-7ec7c3f391cb@dancol.org> <83k2d2rssf.fsf@gnu.org> <831sz9sime.fsf@gnu.org> NNTP-Posting-Host: blaine.gmane.org Mime-Version: 1.0 Content-Type: text/plain X-Trace: blaine.gmane.org 1477074630 25133 195.159.176.226 (21 Oct 2016 18:30:30 GMT) X-Complaints-To: usenet@blaine.gmane.org NNTP-Posting-Date: Fri, 21 Oct 2016 18:30:30 +0000 (UTC) User-Agent: Gnus/5.13 (Gnus v5.13) Emacs/26.0.50 (gnu/linux) Cc: emacs-devel@gnu.org To: Eli Zaretskii Original-X-From: emacs-devel-bounces+ged-emacs-devel=m.gmane.org@gnu.org Fri Oct 21 20:30:25 2016 Return-path: Envelope-to: ged-emacs-devel@m.gmane.org Original-Received: from lists.gnu.org ([208.118.235.17]) by blaine.gmane.org with esmtp (Exim 4.84_2) (envelope-from ) id 1bxea6-0004c6-5g for ged-emacs-devel@m.gmane.org; Fri, 21 Oct 2016 20:30:14 +0200 Original-Received: from localhost ([::1]:33737 helo=lists.gnu.org) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1bxea8-0004nw-Dt for ged-emacs-devel@m.gmane.org; Fri, 21 Oct 2016 14:30:16 -0400 Original-Received: from eggs.gnu.org ([2001:4830:134:3::10]:37474) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1bxeXc-0003WN-3L for emacs-devel@gnu.org; Fri, 21 Oct 2016 14:27:41 -0400 Original-Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1bxeXa-00032I-FU for emacs-devel@gnu.org; Fri, 21 Oct 2016 14:27:40 -0400 Original-Received: from dancol.org ([2600:3c01::f03c:91ff:fedf:adf3]:45486) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1bxeXa-00031S-4W; Fri, 21 Oct 2016 14:27:38 -0400 DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=dancol.org; s=x; h=Content-Type:MIME-Version:Message-ID:In-Reply-To:Date:References:Subject:Cc:To:From; bh=LGfoEkdf/QQN4oyljK1Z1du0ARekI9qaMfbjtnZPRTo=; b=hdL43GCiz/B2oTyi+2TXIRoB8BHf0vnG9YwelTqZSDmWWh8U1GU9GYCaulAjpRYpnqe7buMrr9lc/nYWSK1RJ3bH+3QcU6OUh92MXIPDFj1r2JIh5t+hZA0RC0fu1oWpWp0AJGHJ7fGJrGK52ZwtjMGKsa5GQh/pyfzjevMR3LQYPZCmQHqT5ZJhiqivy8tGxTchCp8OT9aUS9YncbmDV/cl8VKYHyEKE1uzTZUWrfWi/eORnQUbdCBkTyQIhhS3J/QGXm+VnTzZToi7DpBVUldAAu7mgsbuhgTqeYCKywNsv4ndaZyCQteB9QudRNzOtzmPZLRGq2PGnP058eyyLA==; Original-Received: from [2620:0:1008:100b:303b:1201:6da6:b046] (helo=dancol-glaptop0) by dancol.org with esmtpsa (TLS1.2:DHE_RSA_AES_128_CBC_SHA1:128) (Exim 4.84_2) (envelope-from ) id 1bxeXY-0004JL-Mj; Fri, 21 Oct 2016 11:27:36 -0700 In-Reply-To: <831sz9sime.fsf@gnu.org> (Eli Zaretskii's message of "Fri, 21 Oct 2016 20:43:21 +0300") X-detected-operating-system: by eggs.gnu.org: Genre and OS details not recognized. X-Received-From: 2600:3c01::f03c:91ff:fedf:adf3 X-BeenThere: emacs-devel@gnu.org X-Mailman-Version: 2.1.21 Precedence: list List-Id: "Emacs development discussions." List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: emacs-devel-bounces+ged-emacs-devel=m.gmane.org@gnu.org Original-Sender: "Emacs-devel" Xref: news.gmane.org gmane.emacs.devel:208584 Archived-At: Eli Zaretskii writes: >> Cc: emacs-devel@gnu.org >> From: Daniel Colascione >> Date: Fri, 21 Oct 2016 04:04:21 -0700 >> >> > Also, the call to font_flush_frame_caches is unconditional, although >> > only one font back-end supports it. That seems to incur a gratuitous >> > overhead of a function call for the other font back-ends. >> >> We're testing whether the hook function is non-NULL before calling into >> it, so only that one backend gets the call. In any case, the overhead is >> trivial --- it's one indirect function call compared to all of >> redisplay. (Every call into a shared library is the same indirect jump.) > > The code is more clear with the test before the call, otherwise the > code reader needs to look in the hook to understand when it's a no-op. But the test (of whether the hook exists) _is_ before the call. What are you proposing? some_backend_needs_flush = False for backend in font_back_ends: if has_flush_hook(backend): some_back_end_needs_flush = True if some_back_end_needs_flush: for backend in font_back_ends: if has_flush_hook(backend): backend.flush_hook() That's silly. Can you elaborate on what you consider a cleaner approach? > So I'd prefer to have the test outside of the call, or even outside of > the loop. See below. I don't think that giving font back-ends an opportunity to do _something_ in response to a frame being garbaged in certain ways is a bad thing. >> >> + FOR_EACH_FRAME (tail, frame) >> >> + { >> >> + struct frame *f = XFRAME (frame); >> >> + if (FRAME_TERMINAL (f)->redisplay_end_hook) >> >> + (*FRAME_TERMINAL (f)->redisplay_end_hook) (f); >> >> + } >> > >> > This will call the hook for each frame, every redisplay cycle. By >> > contrast, redisplay_internal many times updates only the selected >> > window of the selected frame. Is calling the hook for all the frames >> > really needed, or should we only call it for the selected frame in the >> > latter case, or maybe just for frames that got updated? >> >> The hook only does something in the case where someone called update_end >> and we demurred on actually flipping the buffer because we knew we were >> in redisplay and would be getting redisplay_end_hook shortly. That is, >> if we update only one frame, we're only going to do one buffer flip. > > That might be so, but what you say above is in no way apparent from > looking at the code. IME, it is very important to have the idea when > something is being done and when not just by looking at the code in > redisplay_internal; having to find that out by realizing that some > flag is set in update_end and then tested in the hook makes the code > more subtle and its maintenance harder. It's not like keeping this > detail from redisplay_internal makes this detail local to some > functions, or somesuch, so there's really no reason to conceal it, > IMO. This sentiment strikes me as being analogous to the old "no use of hooks in Emacs internals" line --- yet we have facilities like syntax-ppss that rely on hooks in core, and it's worked out fine. We need to do a buffer flip at the end of redisplay for each frame on which update_end was called during redisplay. If someone calls update_end _outside_ redisplay, we should do a buffer flip immediately. The code I've sent is the cleanest way of implementing this model short of changing how update_begin and update_end work. I think thhat what you're proposing is a layering violation. It will make maintenance harder. The only facility that cares about the has-a-frame-been-updated state is the X11 double-buffered back-end, so making xdisp track this state makes everything more complicated, especially because xdisp already has a "needs redisplay" flag and shouldn't need to keep track of a separate "needs buffer flip" flag. It shouldn't have to care. That's not its job. >> Or are you worried about the function call overhead? That, as I >> mentioned above, is trivial. > > No, I worry about maintainability of the display code and about > lowering the risk of bugs introduced due to such subtleties. > I'm also worried about maintainability: that's why I don't want to make redisplay_internal any more of a big ball of mud than it already is. I think it's cleaner to have xterm keep track of state only xterm needs. >> > Also, should >> > we distinguish between visible and iconified frames? >> >> If we do, we should do it in the code that performs the updates, not the >> code (the snippet immediately above) that publishes the updates we've >> already done. > > See above: I don't like such dependencies and find them in general an > obstacle to understanding the overall logic of the display code. I > don't mind adding a test in update_frame and friends, but that > shouldn't prevent us from having a similar (or even identical) test > here. What dependency? You're proposing adding a lot of complexity to the loop that calls redisplay_end_hook, and I still have no idea what this complexity is supposed to accomplish. > >> >> +#ifdef HAVE_XDBE >> >> + if (FRAME_DISPLAY_INFO (f)->supports_xdbe) >> >> + { >> >> + /* If allocating a back buffer fails, just use single-buffered >> >> + rendering. */ >> >> + x_sync (f); >> >> + x_catch_errors (FRAME_X_DISPLAY (f)); >> >> + FRAME_X_DRAWABLE (f) = XdbeAllocateBackBufferName ( >> >> + FRAME_X_DISPLAY (f), >> >> + FRAME_X_WINDOW (f), >> >> + XdbeCopied); >> >> + if (x_had_errors_p (FRAME_X_DISPLAY (f))) >> >> + FRAME_X_DRAWABLE (f) = FRAME_X_WINDOW (f); >> > >> > Shouldn't we turn off the supports_xdbe flag in the case of failure? >> >> supports_xdbe is whether XDBE is supported on a connection at all. What >> if XdbeAllocateBackBufferName fails transiently? > > How can it fail transiently? And why turning off supports_xdbe in > that case would mean trouble? It's an allocation. Allocations can fail. And XDBE isn't guaranteed to work for all visuals. > >> >> +#ifdef HAVE_XDBE >> >> + dpyinfo->supports_xdbe = false; >> >> + { >> >> + int xdbe_major; >> >> + int xdbe_minor; >> >> + if (XdbeQueryExtension (dpyinfo->display, &xdbe_major, >> &xdbe_minor)) >> >> + dpyinfo->supports_xdbe = true; >> >> + } >> >> +#endif >> > >> > No need for braces here, since we now require a C99 compiler. >> >> If we put xdbe_major and xdbe_minor at function level, the names leak >> into function scope. With braces, they exist only around the call to >> XdbeQueryExtension > > We use that in many other places, so I think these precautions are > misguided and generally make our coding style less apparent. Fine.